Vertical Switching Algorithm for Unmanned Aerial Vehicle in Power Grid Heterogeneous Communication Networks

Abstract

The rapid development of wireless network technology has led to the coexistence of various heterogeneous wireless networks (HWNs). To ensure that users enjoy normal and diversified services, research on vertical switching technology has become an inevitable trend. However, most current vertical switching algorithms only consider static situations or single services, which are not suitable for power grid scenarios. This paper studies the vertical switching problem of wireless heterogeneous networks for unmanned aerial vehicles (UAVs) performing inspection tasks in power grid scenarios. In this model, a UAV for power grid inspection needs to plan its flight trajectory, avoid obstacles, and find the optimal trajectory to reach each inspection point. Throughout the UAV inspection process, we must ensure the quality of communication services for the UAV. The UAV dynamically selects different networks for access at different locations, presenting a dynamic network selection and vertical switching problem. This paper proposes a method that combines trajectory planning and network selection, which first utilizes the A-star algorithm to obtain suitable trajectories, and then evaluates and judges networks based on the Fuzzy Analytic Hierarchy Process (FAHP) to determine the most appropriate network. It is worth noting that this paper considers three service requirements and seven network attributes under three types of heterogeneous wireless networks. Numerical results show that this method can better meet the requirements of UAV inspection tasks and reduce the number of switches, thus addressing the issue of terminal vertical switches in power grid scenarios.

1. Introduction

Today, the power system plays a crucial role as one of the infrastructures. As a trend in the current development of power grid technology, the power grid faces challenges such as wide geographical distribution of electricity service and difficulty in ensuring communication reliability. A single network system can no longer satisfy the service requirements of the power grid [1]. To improve the reliability and operational efficiency of the power system, the power grid is no longer provided by only one operator or a single network standard, but rather in a complex network environment where multiple heterogeneous wireless networks (HWNs) coexist [2,3,4]. In order to integrate the performance advantages of different networks, excellent heterogeneous wireless vertical switching algorithms are essential. However, the current mainstream network vertical switching algorithms remain in the field of wireless communication networks and have not specifically studied the power grid scenarios. Research in this area still needs to be supplemented. At the same time, with the improvement of infrastructure, the scale and coverage of the power grid continue to expand. In areas such as forests, there are uncontrollable factors such as natural disasters like extreme weather and invasion of animals and plants [5]. The existing electric power inspection methods, which mainly rely on manual inspections, have obvious disadvantages such as high cost, low efficiency, long inspection time, and the inability to effectively ensure the safety of inspection personnel [6]. Therefore, safer and more efficient unmanned aerial vehicle (UAV) inspections are necessary [7,8]. The use of multi-mode integrated communication network switching technology and UAV inspection complement each other can achieve reliable terminal access [9,10,11], effectively ensure the normal operation of the smart grid system and improve its reliability and robustness.

Therefore, the combination of vertical switches and UAV inspection is of great significance for achieving the construction and maintenance of power grids, improving the adaptability of the power system, and is an inevitable trend in the development of the power system. Nowadays, many researchers have conducted research on vertical switching algorithm problems from different perspectives and using different methods, and have made a series of progress and achievements.

1.1. Related Works

The authors in [12] determined whether to perform switches based on the received signal strength (RSS) and residence time of adjacent networks, which improved the throughput of some networks and reduced packet delay. The work in [13] introduced the signal-to-noise ratio into the switching parameters, and the proposed scheme enhanced the throughput of the system and achieved a relatively low packet discard probability. A self-adaptive hysteresis algorithm was proposed in [14], which dynamically adjusted the switching threshold based on the movement speed of terminal devices. The results showed that the algorithm is very effective in reducing the impact of speed. The authors in [15] predicted the movement trend of each mobile terminal by predicting the RSS. It is considered that the candidate network will only be connected to the network when the average RSS is greater than the average RSS of the current network over a period of time, but does not take into account service requirements and user preferences. The authors in [16] used historical information to assist in network selection. This method mapped the wireless network environment into a topology graph, where users move and compare with historical trajectory information for network switches. This algorithm effectively improved the success rate of vertical switches, but was only applicable to HWN environments with more historical trajectory information. When there is less historical information, it still relies only on the RSS. It is evident that vertical switching algorithms based on RSS [17] are computationally simple and easy to implement, but they do not consider comprehensive switching decision factors. Therefore, they are unable to support the service requirements of UAV inspection tasks or achieve greater performance improvements.

The work in [18] used evolutionary game and bankruptcy game models to select users and network operators, respectively, in order to obtain a suitable network. The authors proposed a game theory framework in [19] that comprehensively considers user switches needs to maximize the utilization of network resources. The work in [20] used a repeated stochastic game and developed a Lyapunov optimization algorithm to select the target network. However, the output of vertical switching algorithms based on game theory may not necessarily be the solution for maximizing network utility. Additionally, these algorithms require a large amount of computation and high requirements for terminal equipment resources. Therefore, this type of algorithm is not suitable for power grid terminal scenarios.

The authors in [21] obtained the network rating value by multiplying the weights and indicators, which is a simple weighting method that improves switching efficiency but lacks accuracy. The authors in [22] designed a switching algorithm for multiplicative empirical weighting, which is more sensitive than using the simple weighting method. Therefore, the authors in [23] proposed a switching algorithm based on the network fitness function, using bandwidth and other factors as inputs to the function, and the output result is the network score. The work in [24,25] used the analytic hierarchy process (AHP) to obtain the weights of various attribute values of wireless networks, and such methods are relatively common. The authors in [26] used AHP to calculate subjective weights and the entropy method to calculate objective weights. In an environment with uncertain network attribute values, the most suitable network can still be selected. The authors in [27] also calculated subjective and objective weights to obtain network scores. The authors in [28] considered user preferences using entropy and fuzzy analytical hierarchy process (FAHP), calculated the weights and utility values of network attributes and then used a simple additive weighting function to select the appropriate network. The work in [29] proposed a new multi-criteria utility function model and proved its effectiveness through mathematical reasoning. The authors in [30] associated the FAHP, standard deviation method, and grey relational analysis method. Subjective weights were calculated using FAHP, while objective weights were calculated using the standard deviation method. Finally, the priority order of candidate networks was determined, and the algorithm reduced the number of vertical switches. The work in [31] proposed an algorithm based on two-layer fuzzy logic, which effectively reduces unnecessary switching. The authors in [32] applied fuzzy multi-attribute decision-making to network selection problems, which had a certain degree of adaptability. The authors in [33] also considered attributes such as RSS, delay, and terminal energy consumption, and combined the technique for order preference by similarity to an ideal solution with fuzzy logic to select the optimal wireless network, greatly reducing decision time. The work in [34] introduced a ranking method to effectively reduce the number of switches while adopting a multi-attribute decision-making method. The authors in [35] reduced the complexity of the algorithm through combinatorial fusion and outperformed other similar algorithms in terms of timeliness. The work in [36] established a performance evaluation model that includes methods such as FAHP and grey relational analysis, with ten evaluation criteria including bandwidth, delay, etc. The authors in [37,38] considered user preferences while controlling power consumption and cost, reducing unnecessary system overhead and being more friendly to terminal device energy consumption. Overall, research based on fuzzy logic considers the multi-attribute characteristics of networks and can scientifically calculate the optimal selection for the current state. However, the previous works do not consider the impact of terminal mobility on the selection results, and further optimization is needed before applying these to UAV inspections.

1.2. Contributions

This paper mainly studies the heterogeneous network vertical switching problem for UAV inspections in the power grid scenarios, aiming to timely and effectively select a more suitable network for forest inspection services based on the specific scenarios of UAV, satisfy the requirements of service, reduce the occurrence of ping-pong effects, and ensure the normal development of inspection work. It has certain practical significance and research value. The main contributions of this paper are as follows:

• We consider a scenario involving power grid inspection using a UAV, where there is heterogeneous wireless network coverage, including long-term evolution (LTE), wireless local area network (WLAN), and satellite wireless networks. During the inspection, the UAV needs to navigate around obstacles and dynamically select the most suitable network based on their location to ensure communication quality. We develop a joint design for UAV trajectory planning, network selection, and vertical switches, which comprehensively considers seven network attributes such as signal strength, bandwidth, delay, jitter, loss rate, network load, and cost. Taking into account the service requirements and user preferences in the inspection scenario and the requirements for network performance for three different types of services (voice, video, and web browsing), we enable the power grid UAV to dynamically select and switch networks in real-time, ensuring communication service quality throughout the inspection process.
• We use the A-star algorithm to plan the flight trajectory of the UAV so that the UAV can avoid obstacles during the flight and find the shortest trajectory to the inspection points. The algorithm plans the feasible area of the UAV flight through the known obstacles and the terminal position information and finds the relative distance between the landing point and the end point of each flight so that the UAV is closest to the endpoint while avoiding the obstacle in each flight. The algorithm can realize the shortest trajectory of the global flight of the UAV, effectively saving flight power consumption and task execution time, enabling the UAV to avoid obstacles and effectively reach each power grid inspection point.
• We propose a network selection algorithm based on FAHP to calculate the comprehensive utility value for each network. During the inspection process, the UAV selects the current optimal network based on the utility value. This method comprehensively considers the interrelationships between seven network attributes and three service requirements: voice, video, and web. It computes the utility scores for each switching scheme, effectively addressing the uncertainty and ambiguity in the network selection decision process.

The rest of this paper is organized as follows: Section 2 provides the system model and algorithm framework, and Section 3 presents a detailed introduction to the A-star trajectory detection algorithm and the network selection algorithm based on FAHP. The simulation results were discussed in Section 4. Finally, in Section 5, we draw the conclusions.

2. System and Algorithm Model

In this paper, we consider a power grid scenario deployed in a forest, where the inspection UAVs move in a heterogeneous wireless network environment as shown in Figure 1. Based on the inspection requirements of the UAV and the possible obstacles in the forest, there are locations that need to be inspected (inspection points) and positions that need to be avoided (avoidance points) in the scenario.

We have provided specific details of the environment in this scenario: A set of candidate networks is given, which consists of R different types of wireless networks, denoted as $R=\{r_1,r_2,···,r_M\}$, with $M\ge 2$. These include LTE, WLAN, and satellite networks, totaling three types of wireless networks. When users carrying multi-service multimode terminals (MMT) and UAVs are within the coverage of these networks, terminal devices are served by the aforementioned network types, meaning that the terminal devices can move freely within the scenario and connect to any type of network. The set of network attributes considered in network selection decisions is $C=\{c_1,c_2,···,c_N\}$, with $N\ge 2$. This paper considers seven network attributes, i.e., RSS, bandwidth, delay, jitter, loss rate, network load, and cost. These network attributes can affect the communication and service quality of UAVs or terminal devices. For instance, bandwidth represents the capacity of the network to transmit data, while delay refers to the transmission latency from the sender to the receiver. Jitter reflects the instability of delay variations during data transmission, and the loss rate indicates the proportion of packets lost during the transmission process. Changes in these network attributes directly impact communication quality. Terminal devices can receive different network services $S=\{s_1,s_2,···,s_L\}$ from different networks, with $L\ge 2$. Based on the UAV inspection issues in the power grid scenario, network services are broadly categorized into three types: voice, video, and web browsing. Additionally, a $W_C^S=\{w_1^l,w_2^l,···,w_N^l\}$ set is established to represent the weight values of different attributes for different network services, where the superscript S represents the corresponding service type, the subscript C represents the corresponding network attribute and $w_n^l$ represents the $c_n$ attribute of the network service $s_l$. Additionally, we define the real-time network attribute utility value set as $A={\left(a_{ij}\right)}_{M\times N}$, where $a_{ij}$ represents the attribute value $c_j$ of network $r_i$. Based on the trajectory data provided by the method shown in the next section, we can obtain A. By processing it through attribute utility functions, we can obtain the normalized attribute utility value vector $U_l^i=[u_l^{i,1},u_l^{i,2},···,u_l^{i,N}]$ considering that the UAV requires different service priorities for different inspection contents during inspections, we define the service priority set $P=\{p_1,p_2,···,p_k\}$, where $p_k$ represents the priority of network service $s_l$. Therefore, the comprehensive score $Q_i$ for network i can be described as

$$Q_i=\sum _{l=1}^L{p}_{k,l}·W_C^l·{U_l^i}^T.$$

(1)

The problem addressed in this paper is as follows: In the power grid scenario, when a UAV patrols and inspects electrical equipment based on inspection points and avoidance points, it needs to select the most suitable network to support the completion of inspection tasks based on its current location and network status, where $R_t$ can be obtained by

$$R_t=\underset{t\in \{1,2,···,M\}}{argmax}{Q}_i.$$

(2)

The entire algorithm process is shown in Figure 2.

Firstly, the UAV uses the A-star algorithm to find the shortest trajectory to each inspection point while avoiding obstacles in the scene. Secondly, based on fuzzy decision-making, the algorithm calculates the attribute weights for different service types. Then, it normalizes the various network attributes using attribute utility functions. By combining the weight results with the utility values of the network attributes in the candidate network set, the algorithm computes the score for each network and selects the most ideal network accordingly. Finally, the decision is made based on a threshold value to determine whether to maintain the existing network state or switch to a new network.

3. Vertical Switching Strategy for Power Grid

3.1. UAV Trajectory Planning

In this subsection, we will introduce the shortest obstacle avoidance trajectory planning scheme of UAVs based on the A-star algorithm. In mountainous areas, there are some necessary equipment or projects that require regular inspections.

Table 1. Inspection points.

Serial Number	Items	What Needs to Be Checked
1	Power poles	Whether the tower is deformed or tilted
2	Power poles base	Nearby ground conditions
3	Power wires	Whether the installed nuts or bolts pop out
4	Transmission lines	Whether it is damaged, rusted, or entangled in foreign objects
5	Lightning rods, grounding devices	Whether the discharge gap between the two has changed significantly

indicates some specific items to be inspected. Figure 3 shows a schematic diagram of the UAV inspecting power grid equipment in a mountainous environment.

As shown in Figure 3, the UAV needs to avoid obstacles in the mountains and accurately find the shortest trajectory to each inspection point during the inspection process. In order to achieve this goal, the A-star algorithm is used to achieve the shortest obstacle avoidance trajectory planning. Since this paper considers that the altitude of the UAV is unchanged when flying, this paper first divides the UAV flight range into a two-dimensional grid network, there are several nodes in the grid. The UAV needs to check the inspection points in the grid in a predetermined order during the flight, where the access order of the inspection points is subjectively selected from the inspection point set according to the distance from the start and end points of the inspection points and is also selected according to the needs of the power grid scenario. Therefore, the shortest obstacle avoidance trajectory of the UAV from the current inspection point to the next inspection points is planned by the A-star algorithm, and the shortest obstacle avoidance trajectory to the next inspection point needs to be re-planned by the A-star algorithm every time a new inspection point is reached. Through this method, the influence of dynamic obstacles in the scene on the flight trajectory of the UAV can be effectively reduced. In addition, the set of inspection points is a predefined set that does not change in each trajectory plan, and the only dynamic change factor present in the entire scene is the position of the obstacle. It is important to note that the effect of distance/proximity between inspection points on this scenario is not considered in this paper. In our subsequent work, we will consider adding considerations for the relationship between inspection points. Each node represents a position coordinate, and the UAV can reach eight nodes adjacent to the current node for each flight, which is called the adjacent node. Put the coordinates of known obstacles in the environment into the obstacles table.

Each time the UAV decides the flight trajectory, the current node of the UAV is put into the OPEN table and the eight adjacent nodes of the current node are accessed in turn if the adjacent node is in the CLOSE table or in the obstacles table, the node is skipped, and if the adjacent node is not in the CLOSE table, the adjacent node is put into the OPEN table and the current node is set as the parent node of these nodes. After each adjacent node is accessed, the current node is removed from the OPEN table and placed in the CLOSE table. Then, for each node in the OPEN table, its evaluation function $\mathbf{F}$ is calculated. The evaluation function F can be expressed as

$$\mathbf{F}=\mathbf{G}+\mathbf{H},$$

(3)

where $\mathbf{G}$ represents the distance traveled from the parent node to the adjacent node, and $\mathbf{H}$ represents the distance traveled from the adjacent node to the endpoint.

For adjacent nodes, if the lateral or longitudinal movement distance is 1, the diagonal movement distance is 1.414, from which the $\mathbf{G}$ value of the adjacent node can be calculated. For the $\mathbf{H}$ value, the Manhattan method is used to calculate only the number of nodes that pass through the lateral or longitudinal movement from the adjacent node to the endpoint and ignore the diagonal movement and obstacles. After calculating the $\mathbf{F}$ value of a point in the OPEN table, find the node with the smallest $\mathbf{F}$ value in the OPEN table, delete the remaining nodes from the OPEN table, and put them in the CLOSE table. Determines if the point with the smallest $\mathbf{F}$ value is the endpoint. If not, the node is used as the parent node to continue searching for the trajectory. Figure 4 represents the flow chart of the A-star algorithm.

3.2. Network Selection Algorithm Based on FAHP

In this section, we will introduce the process of calculating the weight values of each network attribute through the FAHP method and then calculate the network score. FAHP was proposed by van Laarhoven and Pedrycz [39], who combined AHP and fuzzy logic to enable it to represent the fuzziness of preferences. Therefore, we use FAHP to convert the pairwise relative importance between network attributes into attribute weight values.

Firstly, triangular fuzzy numbers (TFNs) [28] are introduced to represent the pairwise relative importance between network attributes. The TFN can be defined as

$$\alpha =(l,o,u),l\le o\le u,$$

(4)

where l, o, and u represent the lower limit value, the optimal value, and the upper limit value of relative importance.

The calculation rules of TFN are shown in Equation (5).

$$\begin{array}{cc}\hfill {\alpha }_1+{\alpha }_2& =(l_1+l_2,o_1+o_2,u_1+u_2),\hfill \\ \hfill {\alpha }_1\otimes {\alpha }_2& =(l_1\times l_2,o_1\times o_2,u_1\times u_2),\hfill \\ \hfill \frac{1}{{\alpha }_1}& =\left(\frac{1}{l_1},\frac{1}{o_1},\frac{1}{u_1}\right).\hfill \end{array}$$

(5)

The steps to determine the weight value of network attributes using FAHP are as follows:

Step 1: Calculate the comprehensive fuzzy value $F_i$ of attribute $c_i$, which can be represented as

$$F_i=\sum _{j=1}^N{\alpha }_{ij}\otimes {\left[\sum _{i=1}^N\sum _{j=1}^N{\alpha }_{ij}\right]}^{-1},$$

(6)

where ${\sum }_{j=1}^N{\alpha }_{ij}=\left({\sum }_{j=1}^N{l}_{ij},{\sum }_{j=1}^N{o}_{ij},{\sum }_{j=1}^N{u}_{ij}\right)$, and ${\left[{\sum }_{i=1}^N{\sum }_{j=1}^N{\alpha }_{ij}\right]}^{-1}$= $\left(\frac{1}{{\sum }_{i=1}^N{\sum }_{j=1}^N{u}_{ij}},\frac{1}{{\sum }_{i=1}^N{\sum }_{j=1}^N{o}_{ij}},\frac{1}{{\sum }_{i=1}^N{\sum }_{j=1}^N{l}_{ij}}\right)$.

Step 2: Calculate the probability P that $F_j$ is greater than $F_i$, which is given by

$$P(F_j\ge F_i)=\left\{\begin{array}{ccc}1& ,& o_j\ge o_i\\ \frac{(o_j-u_j)-(o_j-l_i)}{l_j-u_i}& ,& o_j\le o_i,l_i\le u_j\\ 0& ,& \mathrm{otherwise}\end{array}\right..$$

(7)

Step 3: Compare the weight values of the attribute $c_j$ of the network service $s_l$, which can be written as

$$w_l^{j^{\prime }}=minP(F_j\ge F_i)=minP(F_j\ge F_1,F_2,···,F_N),j=1,2,···,N.$$

(8)

Step 4: Calculate the normalized weight value $w_l^j$ to satisfy ${\sum }_{j=1}^N{w}_l^j=1$, we have

$$w_l^j=\frac{w_l^{j^{\prime }}}{{\sum }_{j=1}^N{w}_l^{j^{\prime }}},j=1,2,···,N.$$

(9)

Finally, we can obtain the weight matrix $W_l=[w_l^1,w_l^2,···,w_l^N]$ of network service $s_k$.

In the context of UAV inspections for power grids, we consider seven network attributes: RSS, bandwidth, delay, jitter, loss rate, network load, and cost. Divide them into two categories for normalization purposes, namely benefit attributes and cost attributes. We use different utility functions for these two attributes. Generally, the utility value of a benefit attribute is positively correlated with its network attribute value, meaning that higher network attribute values correspond to higher utility values. RSS and bandwidth belong to this category. Conversely, cost attributes are inversely proportional.

When there are bilateral constraints on the values of network attributes, $f_1\left(x\right)$ represents the utility attributes, and $g_1\left(x\right)$ represents the cost attributes, they can be defined as

$$f_1\left(x\right)=\frac{1}{1+e^{-k(x-m)}},$$

(10)

$$g_1\left(x\right)=1-f_1\left(x\right),$$

(11)

where a and b are constant coefficients, which are set according to the requirements of the network properties. When there is only one threshold value for the network attribute, the utility function $f_2\left(x\right)$ for benefit attributes and the cost function $g_2\left(x\right)$ for cost attributes can be defined as

$$f_2\left(x\right)=1-n/\phantom{nx}x,$$

(12)

$$g_2\left(x\right)=1-n·x,$$

(13)

where c is a constant coefficient, acting similarly to a and b. In this paper, the value ranges and utility functions for the seven network attributes of three types of service requirements are shown in

Table 2. QoS requirements, utility functions and parameters of multiple services.

Attributes/Service	Voice	Video	Web
RSS (dBm)	$-85$∼$-30$	$-85$∼$-30$	$-85$∼$-30$
	$f_1\left(x\right)$	$f_1\left(x\right)$	$f_1\left(x\right)$
	$k=0.15,m=-80$	$k=0.15,m=-80$	$k=0.15,m=-80$
Bandwidth (kbs)	32∼64	512∼5000	128∼1000
	$f_1\left(x\right)$	$f_1\left(x\right)$	$f_1\left(x\right)$
	$k=0.25,m=48$	$k=0.003,m=2000$	$k=0.05,m=564$
Delay (ms)	50∼100	75∼150	250∼500
	$g_1\left(x\right)$	$g_1\left(x\right)$	$g_1\left(x\right)$
	$k=0.1,m=75$	$k=0.1,m=112.5$	$k=0.03,m=375$
Jitter (ms)	50∼100	40∼70	10∼150
	$g_1\left(x\right)$	$g_1\left(x\right)$	$g_1\left(x\right)$
	$k=0.185,m=65$	$k=0.175,m=55$	$k=0.05,m=80$
Loss Rate (%)	<30	<30	<30
	$g_2\left(x\right)$	$g_2\left(x\right)$	$g_2\left(x\right)$
	$n=1/30$	$n=1/30$	$n=1/30$
Net Load (%)	<80	<80	<80
	$g_2\left(x\right)$	$g_2\left(x\right)$	$g_2\left(x\right)$
	$n=1/80$	$n=1/80$	$n=1/80$
Cost	<50	<50	<50
	$g_2\left(x\right)$	$g_2\left(x\right)$	$g_2\left(x\right)$
	$n=1/50$	$n=1/50$	$n=1/50$

.

4. Numerical Results and Analysis

In this section, we will verify and analyze the proposed vertical switching algorithm. We set up a scenario of UAV inspections in the power grid in Figure 5.

In our simulation experiments, the satellite network can cover the entire trajectory of the UAV inspection, with the received satellite signal power at the UAV being $-90$∼−130 dbm. The scenario includes five LTE base stations and seventeen WLAN transmitters, with the RSS of both following the cost231-hata model [40]. The transmit power of the LTE base stations and WLAN transmitters is set at 43 W and 23 W, respectively. The fluctuation values corresponding to other network attributes are provided in

Table 3. Network parameter settings.

	Bandwidth (kbs)	Delay (ms)	Jitter (ms)	Loss Rate (%)	Net Load (%)	Cost
Satellite	1200–4000	90–150	50–80	10–25	15–50	30–60
LTE	800–4000	40–80	15–40	6–20	30–50	10–45
WLAN	1000–8000	70–100	30–70	4–15	20–50	0–20

.

The trajectory data are observed using the A-star algorithm, and they can provide the position information of the UAV at a given moment. From this, we can obtain the real-time attribute values of each network. Then, by calculating the network scores based on the respective weights, we can finally select the optimal network. The UAV can receive three types of heterogeneous wireless networks: LTE, WLAN, and satellite networks. The inspection tasks involve three service requirements, including voice, video, and web. We use ‘voice-prefer’, ‘video-prefer’, and ‘web-prefer’ to represent three scenarios. In the ‘voice-prefer’ scenario, the weight ratio for voice services is $0.8$, while the weight ratios for video and web services are $0.1$ and $0.1$, respectively. Similarly, we have set corresponding weight ratios for ‘video-prefer’ and ‘web-prefer’ scenarios, i.e., $[0.1,0.8,0.1]$ for ‘video-prefer’ and $[0.1,0.1,0.8]$ for ‘web-prefer’. Finally, considering seven network attributes, we can derive the fuzzy comparison matrix and weights in

Table 4. Fuzzy comparison matrix and weights for voice service.

Voice	RSS	Bandwidth	Delay	Jitter	Loss Rate	Net Load	Cost	Weight
RSS	(1,1,3)	(3,5,7)	(0.2,0.33,1)	(1,3,5)	(1,3,5)	(5,7,9)	(0.25,0.5,1)	0.1660
Bandwidth	(0.14,0.2,0.33)	(1,1,3)	(0.25,0.5,1)	(0.17,0.25,0.5)	(1,2,4)	(2,4,6)	(0.125,0.17,0.25)	0.1113
Delay	(1,3,5)	(1,2,4)	(1,1,3)	(0.2,0.33,1)	(1,3,5)	(3,5,7)	(0.14,0.2,0.33)	0.1427
Jitter	(0.2,0.33,1)	(2,4,6)	(1,3,5)	(1,1,3)	(3,5,7)	(5,7,9)	(0.2,0.33,1)	0.1702
Loss Rate	(0.2,0.33,1)	(0.25,0.5,1)	(0.2,0.33,1)	(0.14,0.2,0.33)	(1,1,3)	(1,1,3)	(0.11,0.14,0.2)	0.0958
Net Load	(0.11,0.14,0.2)	(0.17,0.25,0.5)	(0.14,0.2,0.33)	(0.11,0.14,0.2)	(1,1,3)	(1,1,3)	(0.11,0.14,0.2)	0.0869
Cost	(1,2,4)	(4,6,8)	(3,5,7)	(1,3,5)	(5,7,9)	(5,7,9)	(1,1,3)	0.2226

,

Table 5. Fuzzy comparison matrix and weights for video service.

Video	RSS	Bandwidth	Delay	Jitter	Loss Rate	Net Load	Cost	Weight
RSS	(1,1,3)	(0.2,0.33,1)	(3,5,7)	(5,7,9)	(1,3,5)	(4,6,8)	(1,3,5)	0.1903
Bandwidth	(1,3,5)	(1,1,3)	(1,2,4)	(0.17,0.25,0.5)	(4,6,8)	(6,8,9)	(4,6,8)	0.1884
Delay	(0.14,0.2,0.33)	(0.25,0.5,1)	(1,1,3)	(0.14,0.2,0.33)	(1,3,5)	(5,7,9)	(1,3,5)	0.1346
Jitter	(0.11,0.14,0.2)	(2,4,6)	(3,5,7)	(1,1,3)	(5,7,9)	(1,3,5)	(5,7,9)	0.1950
Loss Rate	(0.2,0.33,1)	(0.125,0.17,0.25)	(0.2,0.33,1)	(0.11,0.14,0.2)	(1,1,3)	(1,2,4)	(1,1,3)	0.1028
Net Load	(0.125,0.17,0.25)	(0.11,0.125,0.17)	(0.11,0.14,0.2)	(0.2,0.33,1)	(1,1,3)	(1,1,3)	(1,1,3)	0.0879
Cost	(0.2,0.33,1)	(0.125,0.17,0.25)	(0.2,0.33,1)	(0.11,0.14,0.2)	(1,1,3)	(1,1,3)	(1,1,3)	0.1009

and

Table 6. Fuzzy comparison matrix and weights for web service.

Web	RSS	Bandwidth	Delay	Jitter	Loss Rate	Net Load	Cost	Weight
RSS	(1,1,3)	(0.2,0.33,1)	(3,5,7)	(3,5,7)	(0.2,0.33,1)	(4,6,8)	(0.14,0.2,0.33)	0.1499
Bandwidth	(1,3,5)	(1,1,3)	(3,5,7)	(4,6,8)	(0.25,0.5,1)	(5,7,9)	(1,2,4)	0.1869
Delay	(0.14,0.2,0.33)	(0.14,0.2,0.33)	(1,1,3)	(1,2,4)	(0.125,0.17,0.25)	(1,3,5)	(0.17,0.25,0.5)	0.1037
Jitter	(0.14,0.2,0.33)	(0.125,0.17,0.25)	(0.25,0.5,1)	(1,1,3)	(0.11,0.14,0.2)	(1,1,3)	(0.14,0.2,0.33)	0.0879
Loss Rate	(1,3,5)	(1,2,4)	(4,6,8)	(5,7,9)	(1,1,3)	(6,8,9)	(1,3,5)	0.2097
Net Load	(0.125,0.17,0.25)	(0.11,0.14,0.2)	(0.2,0.33,1)	(1,1,3)	(0.11,0.125,0.17)	(1,1,3)	(0.11,0.14,0.2)	0.0874
Cost	(3,5,7)	(0.25,0.5,1)	(2,4,6)	(3,5,7)	(0.2,0.33,1)	(5,7,9)	(1,1,3)	0.1744

based on the algorithm described in Section 3.

We sampled about 25,000 points on the inspection trajectory and selected the network using the RSS-based method, fuzzy multiple criteria group decision making (Fuzzy MCGDM) method [41], fuzzy technique for order preference by similarity to the ideal situation (FTOPSIS) method [42], and the proposed algorithm.

Figure 6 illustrates the network selection probability for voice services. The data volume of voice services is relatively small, so there is no requirement for excessive bandwidth. However, to ensure its continuity, high requirements are placed on delay and jitter. Therefore, the proposed algorithm tends to select WLAN and LTE while avoiding satellite networks with higher delays. Comparing FAHP and MCGDM methods, FAHP selects networks more evenly, which is beneficial for reducing access load. However, FTOPSIS selects satellite networks too frequently, resulting in resource waste. The RSS-based method completely ignores user preferences and application scenarios.

For streaming media services like video, to ensure smooth playback, there are high requirements on network bandwidth, delay, and other attributes. As shown in Figure 7, the FAHP algorithm used in this paper selects WLAN with the highest probability, LTE as the second choice, and uses satellite networks to compensate for the instability of these two networks. MCGDM and FTOPSIS methods almost exclusively choose WLAN networks, ignoring LTE with higher signal strength, which does not satisfy the requirements of video services.

Figure 8 shows the network selection probability of web services. This type of service does not have particularly high requirements for bandwidth and delay, so it is necessary to minimize the selection of costly satellite networks to save resources. The proposed algorithm hardly selects satellite networks and prefers lower-cost WLAN. MCGDM and FAHP methods select satellite networks excessively, far exceeding the network resources required for web services, consuming more energy, and are not suitable for inspection scenarios.

Figure 9 shows the number of algorithm switches. As can be seen from the figure, as the number of network selections increases, the FAHP algorithm has significantly fewer switching times compared to MCGDM and FTOPSIS methods. This indicates that the algorithm can effectively suppress unnecessary switches, reduce the ping-pong effect, and significantly improve user experience.

In summary, the proposed algorithm is based on the shortest trajectory planning of the UAV, taking into full consideration various services and network attributes. It can precisely match service requirements and select the optimal network, not only fulfilling the stringent requirements of power grid inspections but also significantly reducing network switching times, thereby lowering energy consumption. It brings greater efficiency and stability to power grid inspections.

5. Conclusions

In this paper, we use a vertical switching algorithm based on FAHP to solve the trajectory planning and network selection problems of UAVs in power grid scenarios. Based on the location of the UAV, taking into account both service quality requirements and real-time network requirements, select the candidate network with the highest network utility. During the trajectory planning phase, with the known location information of obstacles and inspection points, the A-star algorithm is employed to obtain the UAV’s flight trajectory. In the network selection phase, we use FAHP to calculate the weights of network attributes and determine the real-time utility of each network. By combining user preferences, the optimal network is selected for access. Simulation results show that the proposed algorithm can better adapt to user preferences compared to RSS-based methods, MCGDM, and FTOPSIS. It achieves a more balanced selection probability of each network under the premise of satisfying service requirements, which is conducive to reducing network load. Additionally, the algorithm can reduce the number of network switches. In the future, we will consider combining historical trajectories with machine learning algorithms to optimize the network score of the overall trajectory, and we will conduct more designs for trajectory planning to clarify the rules followed by UAVs.