Network pre-estimation control method based on network fluctuation
By adding network delay module and slope factor to the Smith estimate controller and dynamic matrix control algorithm, the problem of insufficient robustness of the existing technology under random network delay is solved, and more stable remote operation remote control is achieved.
Patent Information
- Application Number
- CN202410986496.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-26
AI Technical Summary
When facing random network delays, the prior art is not robust and it is difficult to effectively handle the time-delay data caused by network fluctuations, affecting the stability of remote operation and remote control.
Using Smith's prediction controller and improved dynamic matrix control algorithm, by building a time-varying time-delay system model, adding network delay modules, eliminating lag, and adding slope factors to dynamic matrix control to optimize control output, achieving more Fast and robust tracking.
It effectively eliminates the impact of network delay on the control system, improves robustness and speed, and ensures stable data processing and control accuracy in network fluctuations.
Smart Images

Figure CN119024687A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of teleoperation and remote control, and is a method for an unmanned vehicle end to process instructions issued by an operator in the case of network fluctuations. Background Art
[0002] In some scenarios that are inaccessible to humans or harmful to humans, the emergence of teleoperated robots has enabled humans to deal with these situations. Teleoperation is now used in many fields. Most of the current remote communications use the Internet. In some cases where the network signal is weak or there is signal interference, the signal will be discontinuous, causing network fluctuations. Therefore, it is very important to study the network fluctuations that occur during remote control of teleoperation.
[0003] At present, there are many prediction control algorithms and improved algorithm applications for processing network fluctuation transmission data. However, most algorithms are not robust when facing random network delays. Therefore, it is very important to process the time-delay data after network fluctuations. Summary of the invention
[0004] In order to overcome the above defects, the present invention provides a network prediction control method based on network fluctuation, which performs network prediction control on a time-varying and time-delay system.
[0005] The present invention is implemented by the following technical solution, comprising the following steps:
[0006] Step 1: Process the data received by the unmanned vehicle, remove outliers and make the data smooth.
[0007] Step 2: Build a Smith predictive controller for data with time-varying delays and add network delay as a feedback module into the controller.
[0008] Step 3: Add the improved dynamic matrix control algorithm to the Smith predictive control system and predict the control output of the control parameters.
[0009] In the step 1, it includes: establishing a random discrete nonlinear system model. First, various parameters of the vehicle are modeled.
[0010] In the step two, it includes: establishing a traditional Smith predictive controller, then adding a network delay module according to the time-varying delay system to update the Smith predictive controller, and obtaining the relevant closed-loop function and closed-loop characteristic equation according to the established model, and simplifying them under the premise of meeting the prerequisites to remove the hysteresis caused by the network delay.
[0011] In the step three, it includes: dynamic matrix algorithm step, setting reference trajectory, adding softening factor, so that the system can smoothly reach the set value. Then establish a prediction output model, and add a slope factor to the classic DMC dynamic matrix algorithm, so that the system can track more quickly according to the slope trend. Then perform rolling optimization on the parameters inside, and use the MPC minimum cost function to solve the optimal solution. The previous steps obtain an open-loop solution, and a closed-loop solution needs to be obtained based on feedback. The error is obtained by subtracting the predicted value from the actual value, and then the predicted value at the next moment is updated, so that feedback correction is achieved. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0013] Figure 1 It is an overall framework diagram of the solution in the specific example provided in the embodiment of the present invention;
[0014] Figure 2 is a schematic diagram of a Smith network prediction controller provided by an embodiment of the present invention;
[0015] Figure 3 is a schematic diagram of unit step sampling provided by an embodiment of the present invention;
[0016] Figure 4 It is a schematic diagram of a dynamic matrix control algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0017] The present invention will be further described in detail below with reference to the accompanying drawings and through specific embodiments. The following embodiments are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.
[0018] Figure 1 It is the overall framework diagram of the present invention.
[0019] Step 1: Process the data received by the unmanned vehicle, add constraints according to the actual situation, such as speed range, acceleration range, etc., remove outliers to avoid huge control fluctuations.
[0020] The state vector of the system is defined as:
[0021] G=[s0,v x , v y ,θ,d x , d y , x, y] (1)
[0022] Where s0 is the distance traveled, v0, v1 are the speeds in the world coordinate system, θ0 is the angle. Because it is an unmanned vehicle, only one angle is needed. d0, d1 represent the direction in the world coordinate system, and x and y represent the position in the world coordinate system.
[0023] Step 2: Build a Smith predictive controller for the unmanned vehicle system with time-varying delay. The data transmission of the unmanned vehicle uses the network for transparent transmission, so the received data has uncertain delay. Therefore, the network delay is added to the controller as a feedback module, such as Figure 2 shown.
[0024] A Smith network predictive controller is designed and a related model is established to eliminate the lag term of the network delay in the closed-loop characteristic equation so that the entire system is not affected by time-varying delays.
[0025] The closed-loop transfer function is:
[0026]
[0027] Among them, G(s) is the controlled object of this system, G m (s) is the estimated model of the controlled object, τ1 is the feedforward delay, τ 1m is the estimated model of the feedforward delay, τ2 is the feedback delay, τ 2m is the estimated model of feedback delay, C(s) is the controller, R(s) is the system input, D(s) is the disturbance input, and Y(s) is the system output. When the model established by the controlled object is correct, that is, τ m =τ and G m (s) = G(s), its transfer function can be simplified as:
[0028]
[0029] The closed-loop characteristic equation can also be simplified to
[0030] 1+C(s)G m (s)=0
[0031] In this way, the Smith network predictive controller is established, eliminating the control impact caused by network delay.
[0032] Step 3: Add the improved dynamic matrix control algorithm to the Smith predictive control system, and predict the control parameters for the control output. The schematic diagram of the dynamic matrix control algorithm is shown in Figure 4 .
[0033] (1) Set the reference trajectory:
[0034] Set its sequence:
[0035] w(t+j)=α j *g(t)+(1-α j )*g r (4)
[0036] w(t) is the reference output of the flexible trajectory of the system, g(t) is the actual output, and g r is the output after subsequent feedback correction, α is the softening factor, which makes the system transition smoothly, 0≤α<1. According to the formula, the smaller α is, the faster the system can reach the set value. The larger α is, the smoother the reference trajectory is, the smaller the control fluctuation is, but the system's rapidity is lower. The system needs to adjust the value of α according to the actual situation, with emphasis on robustness and rapidity.
[0037] (2) Establish a prediction output model:
[0038] The system is sampled by unit step, as shown in the schematic diagram Figure 3 . A is the unit step response matrix and also the unit input control matrix, namely the DMC dynamic control matrix.
[0039] The predicted output is:
[0040]
[0041] in, is the predicted output considering the control increment, It is the predicted output without considering the control increment, Δu(k) is the control increment. Different from the classic DMC algorithm, an operator related to the slope is added to the predicted output. s is the slope of the input, β is the slope factor, which enables the system to track the slope trend more quickly. sgn(s) is the sign function, which changes the positive and negative of the slope effect according to the positive and negative of the slope, and c is the amplification factor of the slope effect.
[0042] Convert it into matrix form:
[0043]
[0044]
[0045] Among them, A is called the dynamic matrix of DMC, P is the rolling optimization time domain length, M is the control time domain length, and C is the slope influence amplification matrix. Adding control increment and slope operator to the prediction output without control increment makes the prediction output fast and robust.
[0046] (3) Scrolling optimization:
[0047] The performance of the optimization indicator can be expressed by the MPC minimum cost function:
[0048]
[0049] Among them, Q is the square weight coefficient of the system's output error, which acts on the error between the predicted value and the actual value. R is the control weight matrix, which acts on the control increment. Q and R focus on rapidity and robustness respectively. Q reflects the system's ability to track the reference trajectory, and R reflects the system's ability to smoothly change the control increment. The minimum cost function here is obtained by calculating ΔU M (k) Calculate the partial differential and obtain the optimal solution, that is, the performance index is optimal.
[0050] (4) Feedback correction:
[0051] The error can be obtained by subtracting the actual value from the predicted value at time k+1:
[0052]
[0053] Then the future error is predicted, and then the predicted output considering the control increment at the kth moment is added to obtain the predicted output at the k+1th moment after correction:
[0054]
[0055] Then introduce the shift matrix to get the next prediction value:
[0056]
[0057]
[0058] S is the shift matrix, which transforms the open-loop solution into a closed-loop feedback solution, which can be adjusted in real time through the error to make more accurate predictions.
Claims
1. A network prediction control method based on network fluctuations is characterized by: Step 1: Process the data received by the unmanned vehicle, remove outliers and make the data smooth. Step 2: Build a Smith predictive controller for data with time-varying delays and add network delay as a feedback module into the controller. Step 3: Add the improved dynamic matrix control algorithm to the Smith predictive control system and predict the control output of the control parameters.
2. The method according to claim 1, characterized in that In the step 1, a discrete nonlinear system model is established for the system, and various parameters are integrated.
3. The method according to claim 1, characterized in that In the step 2, a traditional Smith predictive controller is established, and then a network delay model is added as feedback to perform predictive control on this basis to form a new Smith network predictive controller.
4. The method according to claim 1, characterized in that: In the step 2, the closed-loop function and the closed-loop characteristic equation are simplified based on the established model to remove the network delay under the premise of satisfying the conditions, so that the system is equivalent to a system without network delay.
5. The method according to claim 1, characterized in that In the step three, the controlled object is predictively controlled using the DMC algorithm, thus forming a predictive controller that is a fusion of Smith and DMC.
6. The method according to claim 5, characterized in that Set the reference trajectory and add a softening factor to it so that the system can smoothly reach the set value. w(t+j)=a j *g(t)+(1-a j )*g r (1) w(t) is the reference output of the flexible trajectory of the system, g(t) is the actual output, and g r is the output after subsequent feedback correction, α is the softening factor, which makes the system transition smoothly, 0≤α<1. According to the formula, the smaller α is, the faster the system can reach the set value. The larger α is, the smoother the reference trajectory is, the smaller the control fluctuation is, but the system's rapidity is lower. The system needs to adjust the value of α according to the actual situation, with emphasis on robustness and rapidity.
7. The method according to claim 5, characterized in that The prediction output model is established. The system is sampled by unit step to obtain the dynamic control matrix A. At the same time, a slope factor is added to the prediction output to make the system reach the set value more quickly. The prediction output is: in, is the predicted output considering the control increment, It is the predicted output without considering the control increment, Δu(k) is the control increment. Different from the classic DMC algorithm, an operator related to the slope is added to the predicted output. s is the slope of the input, β is the slope factor, which enables the system to track the slope trend more quickly. sgn(s) is the sign function, which changes the positive and negative of the slope effect according to the positive and negative of the slope, and c is the amplification factor of the slope effect. Convert it into matrix form: Among them, A is called the dynamic matrix of DMC, P is the rolling optimization time domain length, M is the control time domain length, and C is the slope influence amplification matrix. Adding control increment and slope operator to the prediction output without control increment makes the prediction output fast and robust.
8. The method according to claim 5, characterized in that Rolling optimization. The performance of the optimization index can be expressed by the MPC minimum cost function: Among them, Q is the square weight coefficient of the system's output error, which acts on the error between the predicted value and the actual value. R is the control weight matrix, which acts on the control increment. Q and R focus on rapidity and robustness respectively. Q reflects the system's ability to track the reference trajectory, and R reflects the system's ability to smoothly change the control increment. The minimum cost function here is obtained by calculating ΔU M (k) Calculate the partial differential and obtain the optimal solution, that is, the performance index is optimal.
9. The method according to claim 5, characterized in that Feedback correction. Subtracting the actual value from the predicted value at the k+1th moment can give the error: Then the future error is predicted, and then the predicted output considering the control increment at the kth moment is added to obtain the predicted output at the k+1th moment after correction: Then introduce the shift matrix to get the next prediction value: S is the shift matrix, which transforms the open-loop solution into a closed-loop feedback solution, which can be adjusted in real time through the error to make more accurate predictions.
