A neural network sliding mode control method for water level in cable tunnels

Abstract

The present invention discloses a neural network sliding mode control method suitable for the water level in a cable tunnel, and belongs to the technical field of water level control. The present invention uses the water level signal as the controlled variable and the drainage pump current signal as the controlled variable, adopts a sliding mode control strategy, and optimizes the variables to achieve rapid stabilization of the tunnel water accumulation system. Since sliding mode control will bring about the problem of chattering, the radial basis function neural network is used to eliminate the interference effects inside and outside the system, effectively alleviating the timeliness problems and parameter accuracy problems brought about by traditional control methods. While monitoring the water level and the drainage pump, the tunnel water accumulation system controls the output of the drainage pump to achieve stable control of the water level in the water accumulation system. The present invention overcomes the current problem of water level fluctuations, adopts advanced sliding mode control strategies, and improves the accuracy of the water accumulation system.

Description

Technical Field

[0001] The present invention belongs to the technical field of water level control, and in particular relates to a neural network sliding mode control method suitable for cable tunnel water level, which is used when the water level of a tunnel water accumulation system is in dynamic fluctuation. Background Art

[0002] Cable tunnels are tunnel-like structures used to store large numbers of cables laid on cable supports. The cables crisscrossing the tunnels are the central nervous system of steel production. Due to their low terrain and long mileage, cable tunnels are prone to water accumulation, making inspections difficult for on-duty personnel. Failure to promptly drain water from these tunnels can lead to prolonged water accumulation and damage to equipment. Prolonged immersion in water can shorten the lifespan of cables and cause failures. Furthermore, moisture from accumulated water can rise to substation equipment levels, causing condensation and compromising safe operation. Therefore, it is imperative to change traditional manual maintenance methods for cable tunnels and implement intelligent tunnel water level control and automated drainage.

[0003] Sliding mode control is a superior control method characterized by fast response and a small number of adjustable parameters. Ideally, with high-precision observables and no interference within the control system, sliding mode control will not cause chattering. However, tunnels are complex structures with numerous interfering factors, including noise and fly ash. Therefore, most sliding mode controllers are susceptible to chattering, resulting in significant fluctuations in system parameters.

[0004] On the other hand, existing solutions focus on dynamic water level control. When implementing water level control, none of them utilize online parameter adjustment, which results in lags in implementation. Furthermore, they fail to consider the time required for the control process. While this reduces production time, it also increases control time.

[0005] After searching, the patent application number is 201910175674.8, and the application date is March 8, 2019. The name of the invention is: Neural Network Integral Sliding Mode Control Method for Electro-Hydraulic Power Steering System. This application adopts nonlinear integral sliding mode technology as the basic control method. Its switchability enables the control system to have strong robustness to parameter uncertainty and external interference. By combining the adaptive RBF neural network method to approximate the dynamic behavior of the electro-hydraulic power steering system in real time, the designed control method no longer requires the measurement of pump source pressure, working pressure, and left and right tire resistance torque. The designed neural network integral sliding mode control method has strong robustness to model uncertainty and external time-varying interference. However, this application mainly solves the problem that the control performance of the electro-hydraulic power steering system needs to be improved, and is not suitable for promotion and application in the field of cable tunnel water level control technology. Summary of the Invention

[0006] 1. Technical problem to be solved by the invention

[0007] To address the long control times and large fluctuations in system parameters associated with existing cable tunnel water level control technologies, this paper proposes a neural network sliding mode control method for cable tunnel water level control. This method employs a sliding mode control strategy to achieve rapid stabilization of the tunnel water accumulation system and utilizes a radial basis function neural network to eliminate interference effects within and outside the system, effectively alleviating the timeliness and parameter accuracy issues associated with traditional control methods.

[0008] 2. Technical solution

[0009] In order to achieve the above object, the technical solution provided by the present invention is:

[0010] A neural network sliding mode control method for a water level in a cable tunnel according to the present invention comprises the following steps:

[0011] Step 1: Initialize the process controller parameters of the tunnel water accumulation system;

[0012] Step 2: Collect tunnel water accumulation system data, including water level signals and drainage pump current signals, and establish a mathematical model based on the collected data, with the water level signal as the control object and the drainage pump current signal as the controlled object;

[0013] Step 3: Introduce the sliding mode controller into the mathematical model and use the radial basis function neural network to optimize the chattering problem caused by the sliding mode control;

[0014] Step 4: Enter the next cycle and repeat steps 2-3 to loop the system operation process.

[0015] 3. Beneficial effects

[0016] Compared with the existing known technologies, the technical solution provided by the present invention has the following significant effects:

[0017] (1) The present invention is a neural network sliding mode control method for cable tunnel water level, which adopts a sliding mode control strategy to achieve rapid stabilization of the tunnel water accumulation system. It uses a radial basis function neural network to eliminate the interference effects inside and outside the system, effectively alleviating the timeliness and parameter accuracy problems caused by traditional control methods. While monitoring the water level and drainage pump, the tunnel water accumulation system controls the output of the drainage pump to achieve stable control of the water level in the water accumulation system. This overcomes the current problem of water level fluctuations and adopts an advanced sliding mode control strategy to improve the accuracy of the water accumulation system.

[0018] (2) The present invention is a neural network sliding mode control method for the water level in a cable tunnel. The method uses the water level signal as the controlled variable and the drainage pump current signal as the controlled variable, and optimizes the variables through sliding mode control. Since sliding mode control may cause chattering problems, a radial basis function neural network is used to optimize the chattering problem, reduce the system error, and ultimately make the controlled variable reach the set value. Compared with the single use of neural network control, the system of the present invention has a faster convergence speed and is more accurate in finding the optimal solution. Compared with the sliding mode controller, the sliding mode control optimized by the neural network has more optimal choices and avoids the possibility of falling into the local optimum during the iteration process. At the same time, compared with the traditional control strategy, the sliding mode control has fewer control parameters, which reduces the calculation work and saves working time. When the system is disturbed, the present invention can re-adjust the chattering parameters through the improved neural network and optimize the sliding mode control to improve the working condition of the water accumulation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a structural diagram of the water accumulation control system of the present invention;

[0020] Figure 2 This is the sliding mode control logic diagram of the present invention. DETAILED DESCRIPTION

[0021] In order to further understand the content of the present invention, the present invention is described in detail with reference to the accompanying drawings and embodiments.

[0022] Combine Figure 1 and Figure 2 , a neural network sliding mode control method for cable tunnel water level in this embodiment includes the following steps:

[0023] Step 1. Initialize the parameters of the tunnel waterlogging system process controller. The main design parameters of the controller include: drainage pump power gain coefficient K1, drainage pump current gain coefficient K2, power change time T1, water level acquisition time T2 of the waterlogging system, system control input, that is, water level parameter u(k), system control output, that is, drainage pump power parameter y(k), and system state variable x(k).

[0024] Step 2: Collect data from the tunnel waterlogging system, including water level signals and drainage pump current signals. A mathematical model is then built based on the collected data, with the water level signal as the control target and the drainage pump current signal as the controlled object. Since the water level signal is an analog quantity, a liquid level sensor is used for measurement. The drainage pump is an electrical device and requires digital signal control, so a solenoid valve is used.

[0025] The specific steps to establish the water system model are as follows:

[0026] Step 2.1. Establish the transfer function G(s) based on the dynamic characteristics and variables of the water system:

[0027]

[0028] By using modern control theory and model transformation, the transfer function is derived into a state space expression.

[0029]

[0030] Step 2.2: It is known that the water accumulation system will encounter unknown interference during operation, so the state space expression is:

[0031]

[0032] Among them, K3 is the appropriate dimension matrix of the unknown interference, and d(k) is a nonlinear function representing some error.

[0033] Step 3: Introduce the sliding mode controller into the mathematical model and use the radial basis function neural network to optimize the chattering problem caused by the sliding mode control. The specific implementation steps are as follows:

[0034] Step 3.1: Sliding mode control requires a sliding surface. Refer to the state space expression obtained in step 2.2 and introduce the sliding surface switching function definition formula:

[0035]

[0036] Among them, σ i is the sliding mode variable, i=1,2,...,m, λ i is a positive constant.

[0037] The definition of the signum function is as follows:

[0038]

[0039] The characteristic of the sliding mode controller is that the sliding mode variable is forced to 0 within a finite transient time, and a discrete feedback system is used for this purpose.

[0040] The Lyapunov function is used to determine the stability of the controller at this time. For the system state equation:

[0041]

[0042] If the input is x=(x1,x2,...,x n )∈R, then

[0043] sgn(x)=(sgn(x1),...,sgn(x n )) T

[0044] The coefficients of the system state equation are 2×2 matrices, select

[0045] x=(x1,x2)

[0046] The system state space equation can be transformed into

[0047]

[0048] Substituting into the sliding surface switching function, we can get

[0049]

[0050] At this time, the Lyapunov function is introduced

[0051]

[0052] When σ(x) = 0, the Lyapunov function satisfies the following two conditions:

[0053]

[0054]

[0055] If the Lyapunov stability condition is satisfied, σ will eventually stabilize on the sliding surface.

[0056] The characteristics of all states x(k) reaching the sliding surface σ(x) = 0 are determined by the switching function σ(x). The switching function is selected to achieve the process of the entire state x(k) converging to 0.

[0057] Step 3.2: Design the input u(k) so that the system satisfies the initial conditions of sliding mode control:

[0058]

[0059] This initial condition enables the sliding mode controller to achieve the best control effect within a limited time and avoid prolonging the control time.

[0060] Take the input:

[0061]

[0062] The input (α+K3d(k)) is the cause of the chattering of the sliding mode control.

[0063] M=(α+K3d(k))

[0064] The input gain M must not only meet the conditions for controlling chattering, but also weaken the impact of unknown errors.

[0065] Step 3.3: Introduce the correction parameter δ and bring it into the switching function in step 3.1 to obtain:

[0066]

[0067] Substituting the sliding mode conditions, we can obtain:

[0068]

[0069] Substituting the input gain, we can get:

[0070]

[0071] When σ i It is positively correlated with M. When it is on the sliding surface, M is in a stable state.

[0072] The radial basis neural network is optimized for the model output M.

[0073] Automatically differentiate the sample parameters of M to obtain the sample gradient

[0074] The formula for automatic differentiation is:

[0075]

[0076] For two samples M and The radial basis function and can be expressed as the eigenvectors of a certain input space:

[0077]

[0078] The composite function is evaluated in forward mode.

[0079] Represents the squared Euclidean distance between two eigenvectors. γ is a free parameter that can be expressed as the width parameter of the function to control the range of the radial basis function.

[0080] Using the inverse quadratic function optimization method, the free parameters are set to optimizable fixed values:

[0081] γ=log(||Mc||)

[0082] c is the optimization center coefficient of the inverse quadratic function.

[0083] The key issue in the learning method of radial basis function neural network is the reasonable selection of the central parameters of hidden neurons and the choice of unsupervised and supervised hybrid learning strategies.

[0084] Determine the input vector:

[0085] M=[m1,m2,…,m n ] T and

[0086] Where p represents the number of neural network layers.

[0087] Step 3.4: Randomly select h samples from the input vector using the clustering algorithm, and obtain h cluster centers through clustering calculation.

[0088] Using the K-means clustering algorithm, h initial samples are created and set as cluster centers.

[0089] For samples M and Calculate the distance from each parameter to the center of h samples and distribute them into categories β i middle.

[0090] Recalculate the cluster center formula:

[0091]

[0092] The cluster center is the centroid of the sample.

[0093] When the error reaches 0.001, the clustering algorithm stops running.

[0094] The minimum mean square error calculation is used to obtain the weights between the hidden layer and the neurons:

[0095]

[0096] When M becomes 0 after multiple trainings or 100 iterations, the neural network stops running.

[0097] Introduce M into the sliding mode control law and obtain the control variable u(k).

[0098] Output u(k) to the drainage pump control end through digital signal transmission.

[0099] Step 4: Enter the next cycle and repeat steps 2-3 to loop the system operation process.

[0100] like Figure 1 As shown, the control algorithm is stored in the host computer, and the control signal is transmitted to the PLC station through a remote wired method. The signal is transmitted to the water accumulation system through the 485 bus and distributed to the water level signal and the drainage pump current signal.

[0101] In response to the problems of low precision and long control time in tunnel waterlogging systems, this embodiment uses the water level signal as the controlled variable and the drainage pump current signal as the controlled variable. Variables are optimized through sliding mode control. Since sliding mode control can cause chattering problems, a radial basis function neural network is used to optimize the chattering problem and reduce system errors. The ultimate goal is to make the controlled variable reach the set value. Compared with the use of a single neural network control, the system of the present invention converges faster and finds the optimal solution more accurately. Compared with the sliding mode controller, the sliding mode control optimized by the neural network has more optimal options and avoids the possibility of falling into the local optimum during the iteration process. At the same time, compared with traditional control strategies, sliding mode control has fewer control parameters, which reduces calculation work and saves working time. When the system is disturbed, this method can re-adjust the chattering parameters through the improved neural network and optimize the sliding mode control to improve the working condition of the waterlogging system.

[0102] The above is a schematic description of the present invention and its embodiments, which is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs a structure and embodiment similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A neural network sliding mode control method for cable tunnel water level, characterized in that: The following steps are involved: Step 1: Initialize the parameters of the tunnel waterlogging system process controller; the controller parameters include: drainage pump power gain coefficient K1, drainage pump current gain coefficient K2, power change time T1, waterlogging system water level acquisition time T2, system control input, i.e., water level parameter u(k), system control output, i.e., drainage pump power parameter y(k), and system state variable x(k); Step 2: Collect tunnel waterlogging system data, including water level signals and drainage pump current signals. Build a mathematical model based on the collected data, with the water level signal as the control object and the drainage pump current signal as the controlled object. The specific steps for building the mathematical model of the waterlogging system are as follows: Step 2.1: Establish the transfer function G(s) based on the dynamic characteristics and variables of the water accumulation system, and use the model transformation method to derive the transfer function into a state space expression; Step 2.2: Introduce the unknown disturbance of the water accumulation system during operation and obtain the state space expression: Where K3 is the appropriate dimension matrix of the unknown interference, d(k) is a nonlinear function representing some error; Step 3: Introduce the sliding mode controller into the mathematical model and use the radial basis function neural network to optimize the chattering problem caused by the sliding mode control; Step 4: Enter the next cycle and repeat steps 2-3 to loop the system operation process.

2. The neural network sliding mode control method for cable tunnel water level according to claim 1, characterized in that: Step 3 specifically includes the following steps: Step 3.1: Based on the state space expression obtained in step 2.2, introduce the sliding surface switching function to realize the process of the entire state variable x(k) converging to 0; Step 3.2: Design the input u(k) so that the system satisfies the initial conditions of sliding mode control. Step 3.3: Introduce the correction parameter δ and substitute it into the switching function, sliding mode condition, and input gain described in step 3.

1. Perform radial basis function neural network optimization on the input gain M output by the model and determine the input vector. Step 3.4: Randomly select h samples from the input vector using the clustering algorithm, and obtain h cluster centers through clustering calculation; Step 3.5: Introduce M into the sliding mode control law to obtain the control variable u(k), and output u(k) to the drainage pump control end through digital signal transmission.

3. The neural network sliding mode control method for cable tunnel water level according to claim 2, characterized in that: The sliding surface switching function in step 3.1 is: Among them, σ i is the sliding mode variable, i=1,2,...,m, λ i is a positive constant; The definition of the signum function is as follows:

4. The neural network sliding mode control method for cable tunnel water level according to claim 3, characterized in that: In step 3.1, the Lyapunov function is used to determine the stability of the sliding mode controller. If the Lyapunov stability condition is met, σ will eventually stabilize on the sliding surface. The characteristic of all state variables x(k) reaching the sliding surface σ(x) = 0 is determined by the switching function σ(x). The switching function is selected to realize the process of the entire state x(k) converging to 0.

5. The neural network sliding mode control method for cable tunnel water level according to claim 4, characterized in that: Initial conditions described in step 3.2: Take the input: Let the input gain be: M=(α+K3d(k)) The input gain M not only satisfies the conditions for controlling chattering, but also needs to weaken the impact of unknown errors.

6. The neural network sliding mode control method for cable tunnel water level according to claim 5, characterized in that: In step 3.3, the correction parameter δ is introduced and substituted into the switching function in step 3.1 to obtain: Substituting the sliding mode conditions into the equation: Substituting the input gain, we get: When σ i It is positively correlated with M. When it is on the sliding surface, M is in a stable state.

7. The neural network sliding mode control method for cable tunnel water level according to claim 6, characterized in that: Step 3.3 performs radial basis neural network optimization on the model output M; automatically differentiate the sample parameters of M to obtain the sample gradient For two samples M and The radial basis function and can be expressed as the eigenvectors of a certain input space: Use forward mode to calculate composite functions; Represents the squared Euclidean distance between two eigenvectors. γ is a free parameter that can be expressed as the width parameter of the function to control the range of the radial basis function. Using the inverse quadratic function optimization method, the free parameters are set to optimizable fixed values: γ=log(||Mc||) c is the optimization center coefficient of the inverse quadratic function; Select unsupervised and supervised hybrid learning strategies and determine the input vector: M=[m1,m2,…,m n ] T and Where p represents the number of neural network layers.

8. The neural network sliding mode control method for cable tunnel water level according to claim 7, characterized in that: Step 3.4 uses the K-means clustering algorithm to create h initial samples and set them as cluster centers; For samples M and Calculate the distance from each parameter to the center of h samples and distribute them into categories β i middle; Recalculate the cluster center formula: The cluster center is the centroid of the sample; When the error reaches 0.001, the clustering algorithm stops running; The minimum mean square error calculation is used to obtain the weights between the hidden layer and the neurons: When M becomes 0 after multiple trainings or 100 iterations, the neural network stops running.

Images