3.1. DMPC Control Problem Design
In order to ensure the safe and reliable operation of the VCTS, it is necessary to introduce state constraints and control constraints to limit the safe operation protection of trains. Due to the influence of train performance and other factors, the train control constraints are as follows:
where
,
are the maximum braking deceleration and maximum traction acceleration, respectively. Considering the track conditions and the influence of bad weather, the train speed is limited as follows:
where
is the speed limit of train
at operating displacement
. And in the process of the VCTS operation, in order to exclude the possibility of train tailing, a certain safety distance must be maintained between trains, according to the relative braking curve of the train, and the safety distance constraints between the neighboring trains can be calculated as:
where
is the actual distance between the two trains;
is the braking safety margin; due to the role of the speed limit, it is known that
, and then, Formula (9) can be linearized as follows:
When the two trains reach the same speed or the speed of the front train is greater than that of the rear train, the operating spacing of the two trains only needs to satisfy the inequality shown in (10), and when the speed of the front train is less than that of the rear train, the operating spacing of the two trains should satisfy the inequality shown in (11).
In order to facilitate the design of the subsequent VCTS optimal control problem, the following assumptions are given.
Assumption 1 [24]. For the system model (6), given a positive definite symmetric matrix
, , , there exists a constant , a positive definite symmetric matrix
and a local state feedback control law , such that for any terminal state quantity, there is , where the terminal state matrix satisfies , .
Remark 1. For the system model (6), when , according to the local state feedback control law
, it can be obtained that, then . And since the terminal matrix satisfies , it follows that . Similarly, can be obtained from . Therefore, the system has bounded convergence, which leads to . Also for system stability, the terminal matrix should satisfy and . The proofs can be found in Appendix A and Appendix B of the manuscript. Describe the optimal control problem for the VCTS as follows:
In the formula,
,
are the operation index function and terminal index function, respectively, which are expressed as follows:
where
is the prediction time domain;
,
, and
are positive definite symmetric matrices;
is the positive definite weight matrix of the end states that satisfies Assumption 1;
and
are the estimated state quantities of train
and train
at time
versus time
, respectively; and
and
are the predicted state quantities and the control inputs of train
at time
versus time
, respectively, which are described as the following constraint sets:
where
Remark 2. Due to the fact that VCTS conveys information synchronously, then in this paper, we construct the system estimation state and as follows: 3.2. ET-DMPC Controller Design
The traditional DMPC is a time-triggered control algorithm, i.e., it needs to solve the control optimization problem at each moment and transfer the information periodically to the unit trains connected to its topology, whereas the DMPC algorithm based on the ET mechanism solves the optimization problem and transfers the information at the triggering moment only, so it saves a large amount of communication resources and computational resources. As shown in
Figure 3, the block diagram of the ET-DMPC algorithm designed in this paper, when the system satisfies the triggering conditions, the train state information will be passed to the controller, otherwise it will not be passed. The solid line in
Figure 3 indicates that the information is transmitted regularly, while the dotted line indicates that the information is transmitted non-regularly.
The specific values of the state and control quantities to be transferred to the rear train are as follows, considering the two cases that the system satisfies the trigger condition and does not satisfy the trigger condition.
When the system satisfies the trigger condition at time
, the optimal control problem (12) is solved to obtain the optimal solution sequences
and
, and then, the feasible control quantities and the corresponding feasible state quantities at the next moment can be expressed as follows:
where
is the state feedback gain matrix for train
.
When the system does not satisfy the trigger condition at moment
, then the optimal control problem is not solved at that moment, and then, the feasible control quantities at the next moment and the corresponding feasible state quantities can be expressed as follows:
Therefore, the feasible control sequence and the feasible state sequence of the system between the trigger moment
and the next trigger moment
, i.e., at time
, can be expressed as follows:
where
.
In summary, the VCTS control optimization problem based on the ET-DMPC algorithm is described as follows:
Problem 1.
where
is the upper bound on the error of the feasible state quantity
and the estimated state quantity
.
3.3. Event Trigger Condition
In order to ensure the stability requirement of the system, we will derive the event triggering condition of the system at time in the following.
Consider a system triggered to solve a control optimization problem at two moments and , then is the cost function under the action of the feasible control sequence and the feasible state sequence at moment . If is between and , then is the difference between and .
Theorem 1. For the VCTS system (6), consider that the system satisfies the requirements of Assumption 1, and if its event triggering condition is designed as (26), it can be obtained that .
In the formula,where, , ,
.
Remark 3. Due to , consider forcing the ET-DMPC optimization problem to be solved if no event triggering condition occurs in the system in the prediction time domain . Therefore, the event trigger condition can be described as follows: Remark 4. For the event triggering condition (30), the smaller is, the easier the event triggering condition can be satisfied, but at the same time, it will increase the communication cost and computation cost. In particular, when , ET-DMPC becomes DMPC, and the event triggering mechanism fails. As increases, the number of system triggers decreases, but at the same time, the algorithm control performance decreases. Therefore, it is necessary to choose an appropriate trigger parameter to achieve a balance between the system control performance and resource utilization.
The ET-DMPC algorithm proposed in this paper is described in Algorithm 1 as follows:
Algorithm 1 The ET-DMPC algorithm |
Initialization: Set the train performance parameters, the reference speed curve , the speed limit curve , the fixed inter-company distance , the braking safety margin and the initial state of the train in each unit. Define the system parameters , , , , , ; let , , . |
Step 1: Solve Problem 1 to obtain the optimal control sequence at the moment , and then, obtain the optimal predicted state at that moment, apply the control quantity to the unit train, and transfer the optimal predicted state sequence to the topologically connected unit train. |
Step 2: At the moment , judge whether the event trigger condition (30) is satisfied:
- (1)
If satisfied, then , the controller solves Problem 1 based on the actual state quantities at that moment and the predicted state sequences of the topologically connected trains, obtains the optimal control sequence at that moment and then obtains the optimal predicted state at the corresponding moment, applies the control quantities to the train, and passes the optimal predicted state sequences to the trains of its topologically connected units. - (2)
If not, problem 1 is not solved, the controller acts on the unit train according to the feasible control sequences and feasible state sequences obtained from (23) and (24) and does not pass the information to the topologically connected trains, and the controllers of the topologically connected unit trains also solve the control quantities of the unit train according to the feasible control sequences and feasible state sequences of the forward train obtained from (23) and (24).
Step 3: Let , return to step 2. |
In order to describe the above algorithm steps more intuitively, the flowchart of the designed algorithm is shown in
Figure 4.
3.4. Recursive Feasibility
In order to prove the algorithmic feasibility of ET-DMPC, a recursive feasibility analysis of the ET-DMPC algorithm is given in this section.
Theorem 2. Assuming that there is a feasible solution to the constrained optimization problem 1 at time , then for any time , there is a feasible solution to the constrained optimization problem.
Proof of Theorem 2. Assuming that the system triggers the solving of problem 1 at the moment and solves the optimal control sequence and the corresponding optimal state sequence , it is necessary to prove that at the moment , without triggering the solving problem 1, the feasible control sequences and the feasible state sequences under the operation of (23) and (24) still satisfy the constraints (25b)–(25e), and then, the system is recursively feasible.
For time , the feasible control sequence and the feasible state sequence are constructed according to the and obtained at time . It can be seen from Assumption 1 that when , , and , ;
For time , according to the and constructed at the previous time, the feasible control sequence and the feasible state sequence of the time are constructed. Similarly, when , , and , .
In summary, at , the control constraints, state constraints, and terminal state constraints (25b)–(25d) all meet the requirements.
For time , , when , , therefore, ;
For time , , when , , therefore, .
In summary, at , the error constraint (25e) between the feasible state quantity and the estimated state quantity meets the requirements. □