An adaptive control system for the cutting transmission system of a coal mining machine under complex working conditions

Abstract

This invention discloses an adaptive control system for a coal mining machine cutting transmission system under complex working conditions. The system uses an extended state observer in the adaptive torque compensation controller to track the state variables of the cutting transmission system, and estimates in real time disturbances such as nonlinearity and time-varying parameters. The real-time effect of the total disturbance is compensated into the feedforward compensation disturbance term in the actual control quantity. A BP network adaptively adjusts the parameters of the second-order active disturbance rejection torque controller, and an adaptive convolutional neural network fuzzy module adaptively adjusts the output compensation of the active disturbance rejection torque compensation controller to obtain expert control rules that satisfy complex impact conditions. Nonlinear state error feedback, combined with the adjustment amount of the adaptive convolutional neural network fuzzy module, calculates the initial electromagnetic torque. After feedforward compensation, a compensated electromagnetic torque is generated, which is superimposed on the torque compensation amount generated by the PI controller. This achieves adaptive compensation control for suppressing the dynamic load of the system and effective suppression of the impact energy of the dynamic load on the cutting transmission system.

Description

Technical Field

[0001] This invention belongs to the field of coal mining machine control technology, specifically relating to an adaptive control system for a coal mining machine cutting transmission system under complex working conditions. Background Technology

[0002] Coal is my country's basic energy source. The mechanization rate of coal mining has reached 85%, and the mechanization rate of tunneling has reached 65%. Technological innovation has further increased its contribution to the industry's development, and new progress has been made in the informatization and intelligentization of coal mines, resulting in the construction of a number of advanced and efficient smart coal mines. Therefore, higher requirements have been placed on the mechanization, automation, and adaptability of coal mining equipment.

[0003] Due to the complexity and variability of underground coal seams, which contain high-strength rock interbeds, hard inclusions, and rock faults, the external loads acting on the drum are characterized by randomness and strong impact. This easily leads to an increase in the dynamic load on the cutting transmission system, making the cutting section of the coal mining machine one of the weakest parts of the entire machine. Given the complex cutting environment, reliable operation and high-efficiency production of the coal mining machine are essential requirements for its automation and unmanned operation. When the coal seam properties change, the adjustable motion parameters are drum height and traction speed, but current technology has not achieved speed regulation of the cutting drum. Under different cutting conditions, when the traction speed varies within a large range, constant-speed drum cutting makes it difficult to ensure that the coal mining machine operates at its optimal comprehensive performance, characterized by high cutting capacity, low specific energy consumption, and high productivity. Therefore, under unmanned operation, it is necessary to achieve optimal comprehensive coal mining performance under different cutting conditions through variable-speed cutting of the coal mining machine drum and coordinated control of the cutting-traction motion parameters.

[0004] Currently, some scholars have used active control in fields such as wind power and rolling mills to reduce fatigue loads in electromechanical transmission systems caused by sudden changes in external loads. This involves designing damping controllers to superimpose compensating torque onto the original generator / motor torque, thereby increasing the electrical damping of the transmission chain. Some scholars have established torsional vibration damping controllers based on bandpass filters (BPFs) to suppress impact loads in wind turbine gear transmission systems caused by sudden changes in external loads and to avoid transmission chain resonance caused by torsional vibrations. They have proposed active fatigue load control strategies based on BPFs, using generator speed as feedback, to suppress multiple resonant frequencies of the system. However, when the controlled object has model parameter disturbances and uncertainties, the stability of BPF-based damping controllers and their performance in suppressing fatigue loads in the transmission system will be affected. To address the shortcomings of the above controllers, some scholars have established damping controllers based on the controlled object model, using generator speed as feedback, and employing Kalman filters to estimate the aerodynamic torque load of the system. However, this method relies on relatively accurate transmission system parameters. In reality, electromechanical transmission systems have complex structures and time-varying parameters, making it difficult to obtain accurate system parameters. To address the shortcomings of the above methods, some scholars have applied active disturbance rejection control (ADRC) technology, which does not rely on a precise mathematical model of the controlled object and can automatically compensate for system parameter disturbances, to the field of torsional vibration suppression in rolling mills and wind power transmission systems. However, these studies simplify the complex transmission system to an inertial mass model during modeling, only considering shaft torsional vibration, without establishing a dynamic model that includes gears. They cannot consider the time-varying meshing stiffness of gears and the dynamic load of the gear transmission system, and cannot analyze the impact of time-varying parameters on the performance of the ADRC. Summary of the Invention

[0005] This invention addresses the problems existing in the prior art by providing an adaptive control system and method for a coal mining machine cutting transmission system under complex working conditions. This control method and system can effectively suppress dynamic load impact energy and achieve adaptive adjustment of system parameters under complex cutting conditions, thereby improving the safety and production efficiency of the coal mining machine under complex cutting conditions.

[0006] To solve the above technical problems, the present invention provides the following technical solution: an adaptive control system for a coal mining machine cutting transmission system under complex working conditions, comprising: a sample construction module, an adaptive torque compensation controller, an overlay module, and a cutting motor speed adjustment module;

[0007] The sample construction module is used to calculate the difference between the actual speed of the cutting motor and the speed of the high-speed gear in the cutting transmission system; based on the speed difference and the corresponding controller output compensation torque, a training sample dataset is constructed; the adaptive torque compensation controller is used to output the controller output compensation torque, and the adaptive torque compensation control neural network is trained using the sample dataset. During the training process, a dynamic adjustment factor is introduced to dynamically adjust the controller output; the adaptive torque compensation controller includes a second-order nonlinear active disturbance rejection torque module, an adaptive convolutional neural network fuzzy module, and a BP neural network;

[0008] The dynamic adjustment factor includes the rate of change factor α, the disturbance factor β, and the error factor γ. The rate of change factor α, the disturbance factor β, and the error factor γ are weighted and fused to obtain the comprehensive adjustment factor λ, λ = s1α + s2β + s3γ, where s1, s2, and s3 are weighting coefficients.

[0009] The second-order nonlinear active disturbance rejection torque module includes a tracking differentiator, an extended state observer, and a nonlinear error feedback law module. The second-order nonlinear active disturbance rejection torque module introduces a dynamic adjustment factor to enhance the adaptability and robustness of the adaptive torque compensation controller. The dynamic adjustment factor dynamically adjusts the control output according to the system's operating state and external disturbances.

[0010] The adaptive convolutional neural network fuzzy module adopts an adaptive neural-fuzzy control system (ANFIS) combined with a multi-layer convolutional neural network structure. Specifically, a multi-layer convolutional neural network (CNN) is introduced before the input layer of the ANFIS system. This adaptive convolutional neural network fuzzy module is used to improve the adaptive capability of the second-order nonlinear active disturbance rejection torque module under different working conditions, and to obtain expert control rules for the cutting transmission system under different coal and rock impacts. It generates an adaptive adjustment amount ΔT according to the changes in working conditions. e ;

[0011] The adaptive convolutional neural network fuzzing module includes: a data input module, which converts the output signal of the tracking differentiator into a format suitable for processing by a multi-layer convolutional neural network (CNN) and performs normalization and standardization processing to form a time slice of the multi-layer convolutional neural network (CNN);

[0012] The multi-layer convolutional neural network (CNN), including multiple convolutional layers, activation function layers, and pooling layers, processes the output signal of the tracking differentiator through multiple layers to extract high-level features of the input data. Finally, the feature vector is output through a fully connected layer. The feature fusion module fuses the features extracted by the multi-layer convolutional neural network (CNN) with other input features of the adaptive neural-fuzzy control system (ANFIS) and inputs them into the adaptive neural-fuzzy control system (ANFIS) for adaptive fuzzy reasoning and decision-making.

[0013] A BP neural network is used to optimize the parameters of the nonlinear error feedback law module and the extended state observer in the second-order nonlinear active disturbance rejection torque module.

[0014] The superposition module superimposes the compensation torque output by the controller onto the torque output by the PI controller to obtain the sum of torques.

[0015] The cutting motor speed adjustment module takes the sum of torques as input to the motor controller, obtains the output of the motor controller, and adjusts the cutting motor speed in real time based on the output.

[0016] Furthermore, the aforementioned tracer differentiator is configured to perform the following actions to arrange a transition process for the input signal:

[0017]

[0018] In the formula, T is the sampling step size; r is the velocity factor; d = rh0; h0 is the filtering factor; ω 11 For the transitioned input signal, ω 22 For ω 11 The derivative of fst is the fastest control synthesis function;

[0019] The extended state observer is configured to perform the following actions:

[0020]

[0021] |e|≤δ

[0022] |e|>δ

[0023] In the formula, y(k) is the system output; e(k) is the difference between the extended state observer's observation output and the system output; β1, β2, and β3 are constants; α1, α2, and α3 are nonlinear parameters, all of which are positive.

[0024] The nonlinear error feedback control law module is configured to perform the following actions:

[0025]

[0026] In the formula, β4 and β5 are the proportional factor and differential factor, respectively.

[0027] Furthermore, the aforementioned dynamic adjustment factors include a rate of change factor α, a disturbance factor β, and an error factor γ; wherein... γ = |e|, where e is the control error. The rate of change factor α, the disturbance factor β, and the error factor γ are weighted and fused to obtain the comprehensive adjustment factor λ, λ = s1α + s2β + s3γ, where s1, s2, and s3 are weighting coefficients.

[0028] Furthermore, the compensation output torque of the aforementioned second-order nonlinear active disturbance rejection torque module is: T e2 =(1+λ)[β4fal(e1(t),a4,δ)+β5fal(e2(t),a5,δ)-z3 / b0].

[0029] Furthermore, the aforementioned BP neural network includes an input layer, a hidden layer, and an output layer; the input layer includes 4 neural nodes to track the differentiator output signal e1, its derivative e2, the total output y of the control system, and 1 as neuron inputs; the hidden layer, combined with the truncated transmission system, has 6 nodes; the output layer has 5 nodes, corresponding to the second-order nonlinear active disturbance rejection torque controller parameters β1, β2, β3, β4, and β5, respectively.

[0030] Furthermore, the aforementioned adaptive convolutional neural network fuzzy module adopts an adaptive neural-fuzzy control system (ANFIS) combined with a multi-layer convolutional neural network structure. Specifically, a multi-layer convolutional neural network (CNN) is introduced before the input layer of the ANFIS. This adaptive convolutional neural network fuzzy module is used to improve the adaptive capability of the second-order nonlinear self-disturbance torque module under different working conditions, obtain expert control rules for the cutting transmission system under different coal and rock impacts, and generate an adaptive adjustment amount ΔT according to the changes in working conditions. e ;

[0031] The adaptive convolutional neural network fuzzing module includes: a data input module, which converts the output signal of the tracking differentiator into a format suitable for processing by a multi-layer convolutional neural network (CNN) and performs normalization and standardization processing to form a time slice of the multi-layer convolutional neural network (CNN);

[0032] A multi-layer convolutional neural network (CNN) consists of multiple convolutional layers, activation function layers, and pooling layers. It processes the output signal of the tracking differentiator through multiple layers to extract high-level features from the input data. Finally, it outputs the feature vector through a fully connected layer. The feature fusion module fuses the features extracted by the CNN with other input features of the adaptive neural-fuzzy control system (ANFIS) and inputs them into the ANFIS for adaptive fuzzy inference and decision-making.

[0033] Furthermore, the aforementioned adaptive neural-fuzzy control system (ANFIS) includes an input layer, a fuzzification layer, a fuzzy rule layer, a fuzzy inference layer, and a defuzzification layer;

[0034] The input layer accepts the fused feature vector, and defines the fused feature vector of the ANFIS input as x. i =[e1(k),e2(k)] T ;

[0035] The fuzzification layer assigns a linguistic variable value to each node using fuzzy inference rules and calculates the membership function of each input component belonging to the fuzzy set of each linguistic variable value. The membership function is represented by a Gaussian function, and its formula is: In the formula, i = 1, 2; j = 1, 2, ..., m j m j It is x i The number of fuzzy segments; c ij and δ ij These are the center and width of the membership function, respectively;

[0036] The fuzzy rule layer is used to define fuzzy rules, i.e. In the formula, i1∈{1,2,...,m1}, ..., i1∈{1,2,...,m1}, j=1,2,...,m, Perform fuzzy reasoning;

[0037] The fuzzy inference layer, in conjunction with fuzzy rules, calculates the applicability of each rule and performs normalized calculations, i.e.

[0038] The defuzzification layer converts the fuzzy inference results into explicit control outputs. In the formula, w ij This represents the connection weight between the fourth-layer node and the output layer node.

[0039] Furthermore, when the aforementioned adaptive neural-fuzzy control system (ANFIS) performs adaptive fuzzy reasoning and decision-making, a performance index function is defined. In the formula, t i Output the target value; calculate using gradient descent. Then, the first-order gradient optimization algorithm is used to adjust w. ij c ij δ ij The first-order gradient is:

[0040]

[0041] The adaptive fuzzy inference parameter learning algorithm is as follows:

[0042]

[0043] Furthermore, the aforementioned adaptive convolutional neural network fuzzing module is configured to perform the following actions: T tra =-1×(T) me -T ste ), where T me T is the electromagnetic torque output by the motor itself. steT is the torque that is opposite in direction to the electromagnetic torque output by the motor. tra The steady-state target value of the motor's electromagnetic torque; by continuously training with the controller's input feature vector, an adaptive convolutional neural network can be obtained to compensate for the electromagnetic torque adjustment ΔT of the control system. e .

[0044] The final output torque of the adaptive torque compensation controller is:

[0045] T e2 =(1+λ)[β4fal(e1(t),a4,δ)+β5fal(e2(t),a5,δ)+ΔT e ]-z3 / b0

[0046] The superposition module superimposes the PI control output torque Te1;

[0047] The cutting motor speed adjustment module outputs the total output compensation torque of the adaptive control system for the coal mining machine as follows:

[0048] T eM =Te1+(1+λ)[β4fal(e1(t),a4,δ)+β5fal(e2(t),a5,δ)+ΔT e ]-z3 / b0.

[0049] Compared with the prior art, the beneficial technical effects of the present invention using the above technical solution are as follows:

[0050] 1. This invention realizes the application of active disturbance rejection technology in the cutting transmission system of a coal mining machine, overcoming the problem that the cutting motor using traditional direct torque control is difficult to quickly and effectively suppress the dynamic load caused by impact loads in the cutting transmission system of a coal mining machine under harsh cutting conditions in deep coal seams. By introducing a dynamic adjustment factor into the second-order nonlinear active disturbance rejection torque compensation controller, the compensation torque can more accurately adapt to changes in actual working conditions. The dynamic adjustment factor can automatically adjust the compensation intensity according to real-time working conditions, improving the system's response speed and control accuracy, thereby more effectively suppressing the impact energy of dynamic loads. This control method does not rely on the mathematical model of the controlled object. Compared with the traditional direct torque control method, this control method not only has good tracking and estimation performance, but is also insensitive to time-varying parameters of the system, reducing the influence of time-varying parameters on the second-order nonlinear active disturbance rejection torque controller.

[0051] 2. This invention introduces a multi-layer convolutional neural network (CNN) before the adaptive neural-fuzzy control system (ANFIS) to perform hierarchical feature extraction on the input data. CNN excels at processing high-dimensional data, extracting multi-level features, and more accurately capturing complex nonlinear relationships. By introducing a multi-layer convolutional neural network (CNN) after the input layer of ANFIS, the system's learning and generalization abilities can be improved, its feature extraction capabilities enhanced, and its processing performance on complex input data improved.

[0052] 3. This invention employs a control method combining an adaptive convolutional neural network fuzzy controller and a second-order nonlinear active disturbance rejection torque compensation controller, effectively solving the problem of adaptive adjustment capability of the output torque compensation of the second-order nonlinear active disturbance rejection torque compensation controller under complex operating conditions. By combining the advantages of multi-layer convolutional neural network control and fuzzy control, adaptive adjustment of the output compensation of the active disturbance rejection torque compensation controller is achieved, and expert control rules satisfying complex impact conditions are obtained. Adaptive adjustment amounts can be generated according to changes in operating conditions, thereby realizing adaptive compensation control for suppressing dynamic loads in the system. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the adaptive control system for a coal mining machine under complex working conditions.

[0054] Figure 2 This is a structural diagram of an adaptive torque compensation controller.

[0055] Figure 3 This is a structural diagram of a second-order nonlinear active disturbance rejection torque compensation controller.

[0056] Figure 4 This is a diagram of a BP neural network structure.

[0057] Figure 5 This is a structural diagram of an adaptive convolutional neural network fuzzy controller.

[0058] Figure 6 This is a structural diagram of the adaptive neural network-fuzzy control module.

[0059] Figure 7 This is a flowchart of the adaptive control system for a coal mining machine under complex working conditions. Detailed Implementation

[0060] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0061] In this invention, various aspects of the invention are described with reference to the accompanying drawings, in which numerous illustrative embodiments are shown. Embodiments of the invention are not limited to those depicted in the drawings. It should be understood that the invention is implemented through any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0062] like Figure 1 The diagram shown is a schematic of the adaptive control system for a coal mining machine under complex working conditions. This control method collects the rotational speed ω at one end of the flexible shaft of the cutting motor. m The difference Δω1 between the rotational speed ω1 of the flexible shaft end in the gear transmission system and the Δω1 is used as the input to the adaptive torque compensation controller. The output Te2 of the adaptive torque compensation controller is superimposed on the output Te1 of the PI controller to form the total compensation torque Te for motor control. m The real-time adjustment of the cutting motor speed is a control method that does not rely on the mathematical model of the controlled object. Compared with the traditional direct torque control method, this control method not only has good tracking and estimation performance, but is also insensitive to the time-varying parameters of the system, and can more effectively suppress the impact energy of external dynamic loads in complex cutting environments.

[0063] This invention provides an adaptive control system for a coal mining machine cutting transmission system under complex working conditions, comprising: a sample construction module, an adaptive torque compensation controller, an overlay module, and a cutting motor speed adjustment module;

[0064] The sample construction module is used to calculate the difference between the actual speed of the cutting motor and the speed of the high-speed gear in the cutting transmission system; and to construct a training sample dataset based on the speed difference and the corresponding controller output compensation torque.

[0065] like Figure 2 The diagram shows the structure of the adaptive torque compensation controller. The controller includes a second-order nonlinear active disturbance rejection torque module, an adaptive convolutional neural network fuzzy module, and a backpropagation (BP) neural network. The controller outputs the compensation torque. The adaptive torque compensation control neural network is trained using a sample dataset, and a dynamic adjustment factor is introduced during the training process to dynamically adjust the controller output.

[0066] The second-order nonlinear active disturbance rejection torque module includes a tracking differentiator, an extended state observer, and a nonlinear error feedback law module. It introduces a dynamic adjustment factor to enhance the adaptability and robustness of the active disturbance rejection controller. This dynamic adjustment factor dynamically adjusts the control output based on the system's operating state and external disturbances. An adaptive convolutional neural network fuzzy module is used to improve the module's adaptability to different operating conditions and to obtain expert control rules for the cutting transmission system under varying coal and rock impacts. It generates an adaptive adjustment amount ΔT based on changes in operating conditions. e A backpropagation neural network is used to optimize the parameters of NLSEF and ESO in a second-order nonlinear active disturbance rejection torque controller.

[0067] The superposition module superimposes the compensation torque output by the controller onto the torque output by the PI controller to obtain the sum of torques.

[0068] The cutting motor speed adjustment module takes the sum of torques as input to the motor controller, obtains the output of the motor controller, and adjusts the cutting motor speed in real time based on the output.

[0069] like Figure 3 The diagram shows the structure of a second-order nonlinear active disturbance rejection torque compensation controller. The tracking differential controller (TD) arranges the transient process for the input signal and is configured to perform the following actions:

[0070]

[0071] In the formula, T is the sampling step size; r is the velocity factor; d = rh0; h0 is the filtering factor. ω 11 For the transitioned input signal, ω 22 For ω 11 The derivative of fst. fst is the fastest control synthesis function.

[0072] The Extended State Observer (ESO) is configured to perform the following actions:

[0073]

[0074] In the formula, y(k) is the system output; e(k) is the difference between the observation output of the extended state observer and the system output; β1, β2, and β3 are constants; α1, α2, and α3 are nonlinear parameters, all of which are positive values; the extended state observer is established based on the measurement output and control input of the cutting transmission system, estimates the system and external disturbances, and compensates for torque in real time through the disturbance estimate.

[0075] The nonlinear error feedback control law module is configured to perform the following actions:

[0076]

[0077] In the formula, β4 and β5 are the proportional factor and differential factor, respectively.

[0078] The introduced dynamic adjustment factors include the rate of change factor α, the disturbance factor β, and the error factor γ; where γ = |e|, where e is the control error. The above factors are weighted and fused to form a comprehensive adjustment factor λ, λ = s1α + s2β + s3γ, where s1, s2, and s3 are weighting coefficients.

[0079] The second-order nonlinear active disturbance rejection torque controller uses disturbance compensation to compensate the real-time effect of the total disturbance into the actual control quantity. After disturbance compensation, the compensated output torque of the second-order nonlinear active disturbance rejection torque controller is:

[0080] T e2 =(1+λ)[β4fal(e1(t),a4,δ)+β5fal(e2(t),a5,δ)-z3 / b0]

[0081] like Figure 4 The diagram shows the structure of a BP neural network. Because the second-order nonlinear active disturbance rejection torque controller has too many fixed parameters, limiting the torque compensation output under different cutting conditions, a BP neural network is introduced to optimize the parameters in the NLSEF and ESO of the active disturbance rejection controller. This BP neural network uses a three-layer structure: an input layer, a hidden layer, and an output layer. The input layer uses four nodes to track the differentiator output signal e1, its derivative e2, the total output y of the control system, and 1 as neuron inputs. The hidden layer has six nodes, determined based on the cutting drive system. The output layer has five nodes, corresponding to the second-order nonlinear active disturbance rejection torque controller parameters β1, β2, β3, β4, and β5. The BP neural network controller adjusts the NLSEF and ESO parameters according to the changes in the controlled object and external disturbances, which helps improve the accuracy of state estimation, enhances system robustness, and improves control performance.

[0082] like Figure 5 The diagram shows the structure of an adaptive convolutional neural network fuzzy controller. To further improve the adaptive capability of a second-order nonlinear active disturbance rejection torque controller, this adaptive convolutional neural network fuzzy module was designed. This controller employs an adaptive neural-fuzzy control system (ANFIS) combined with a multi-layer convolutional neural network (CNN). CNNs excel at processing high-dimensional data, extracting multi-level features, and more accurately capturing complex nonlinear relationships. Introducing a multi-layer convolutional neural network (CNN) after the input layer of the ANFIS improves the system's learning and generalization abilities. Figure 5The specific structure of the adaptive convolutional neural network fuzzy controller includes: a data input module, which converts the output signal of the tracking differentiator into a format suitable for CNN processing and performs normalization and standardization to form CNN time slices; a multi-layer convolutional neural network (CNN), including multiple convolutional layers, activation function layers, and pooling layers, which extracts high-level features from the input data after multi-layer processing, and finally outputs the feature vector through a fully connected layer; a feature fusion module, which fuses the features extracted by the multi-layer convolutional neural network with other input features of the ANFIS, and inputs them into the ANFIS for fuzzy inference and decision-making; and an adaptive neural-fuzzy inference module.

[0083] like Figure 6 The diagram shown is a structural diagram of an adaptive neural network-fuzzy control module, including an input layer that receives the fused feature vector and defines the fused feature vector x as the input to the ANFIS. i =[e1(k),e2(k)] T The fuzzification layer assigns a linguistic variable value to each node using fuzzy inference rules and calculates the membership function of each input component to the fuzzy set of each linguistic variable value. The membership function is represented by a Gaussian function, and its formula is: In the formula, i = 1, 2; j = 1, 2, ..., m j m j It is x i The number of fuzzy segments; c ij and δ ij These represent the center and width of the membership function, respectively. The fuzzy rule layer defines the fuzzy rules, i.e. In the formula, i1∈{1,2,...,m1}, ..., i1∈{1,2,...,m1}, j=1,2,...,m, Perform fuzzy reasoning; in the fuzzy reasoning layer, combine fuzzy rules to calculate the applicability of each rule, and perform normalization calculations, i.e. The defuzzification layer converts the fuzzy inference results into explicit control outputs. In the formula, w ij This represents the connection weight between the fourth-layer node and the output layer node.

[0084] In the adaptive neural-fuzzy inference module, a performance index function is defined. In the formula, t i Output the target value. Calculate using gradient descent. Then, the first-order gradient optimization algorithm is used to adjust w. ij c ij δ ij The first-order gradient is:

[0085]

[0086] The adaptive neural-fuzzy inference module parameter learning algorithm can be obtained as follows:

[0087]

[0088] To obtain expert control rules under different operating conditions, the adaptive convolutional neural network fuzzy controller training algorithm is defined as: T tra =-1×(T) me -T ste ), where T me T is the electromagnetic torque output by the motor itself. ste T is the torque that is opposite in direction to the electromagnetic torque output by the motor. tra Let ΔT be the steady-state target value of the motor's electromagnetic torque. By continuously training with the controller's input feature vector, an adaptive convolutional neural network fuzzy controller can be obtained to compensate for the electromagnetic torque adjustment ΔT in the control system. e .

[0089] The final output torque of the adaptive torque compensation controller is:

[0090] T e2 =(1+λ)[β4fal(e1(t),a4,δ)+β5fal(e2(t),a5,δ)+ΔT e ]-z3 / b0

[0091] The total output compensation torque of the coal mining machine adaptive control system is: (The PI control output torque Te1 is superimposed on the total output torque Te1.)

[0092] T eM =Te1+(1+λ)[β4fal(e1(t),a4,δ)+β5fal(e2(t),a5,δ)+ΔT e ]-z3 / b0

[0093] refer to Figure 7 On the other hand, the present invention also provides a control method based on an adaptive control system for a coal mining machine cutting transmission system under complex working conditions, comprising the following steps:

[0094] S1. Determine whether the cutting resistance of the coal mining machine drum changes. If yes, it is determined to be a full coal working condition and proceeds to step S2. Otherwise, continue to determine whether it is a rock interbedded working condition. If yes, proceed to step S2. Otherwise, it is determined to be a fault working condition and proceeds to step S2.

[0095] S2. Determine the motor speed and impedance, and input the signal to the adaptive torque compensation controller. The controller compensates for the electromagnetic torque in real time according to the signal and outputs the electromagnetic torque to adjust the cutting motor speed, traction speed and drum speed. S3. Determine whether the cutting impedance has changed. If yes, execute step S2; otherwise, end.

[0096] While the present invention has been described above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. An adaptive control system for a coal mining machine cutting transmission system under complex working conditions, characterized in that, include: The module includes a sample construction module, an adaptive torque compensation controller, an overlay module, and a cutting motor speed adjustment module. The sample construction module is used to calculate the difference between the actual speed of the cutting motor and the speed of the high-speed gear in the cutting transmission system; a training sample dataset is constructed based on the speed difference and the corresponding controller output compensation torque; the adaptive torque compensation controller is used to output the controller output compensation torque, and the adaptive torque compensation control neural network is trained using the sample dataset. During the training process, a dynamic adjustment factor is introduced to dynamically adjust the controller output. The adaptive torque compensation controller includes a second-order nonlinear active disturbance rejection torque module, an adaptive convolutional neural network fuzzy module, and a BP neural network; The dynamic adjustment factor includes the rate of change factor α, the disturbance factor β, and the error factor γ. The rate of change factor α, the disturbance factor β, and the error factor γ are weighted and fused to obtain the comprehensive adjustment factor λ, λ = s1α + s2β + s3γ, where s1, s2, and s3 are weighting coefficients. The second-order nonlinear active disturbance rejection torque module includes a tracking differentiator, an extended state observer, and a nonlinear error feedback law module. The second-order nonlinear active disturbance rejection torque module introduces a dynamic adjustment factor to enhance the adaptability and robustness of the adaptive torque compensation controller. The dynamic adjustment factor dynamically adjusts the control output according to the system's operating state and external disturbances. The adaptive convolutional neural network fuzzy module adopts an adaptive neural-fuzzy control system (ANFIS) combined with a multi-layer convolutional neural network structure. Specifically, a multi-layer convolutional neural network (CNN) is introduced before the input layer of the ANFIS system. This adaptive convolutional neural network fuzzy module is used to improve the adaptive capability of the second-order nonlinear active disturbance rejection torque module under different working conditions, and to obtain expert control rules for the cutting transmission system under different coal and rock impacts. It generates an adaptive adjustment amount ΔT according to the changes in working conditions. e ; The adaptive convolutional neural network fuzzing module includes: a data input module, which converts the output signal of the tracking differentiator into a format suitable for processing by a multi-layer convolutional neural network (CNN) and performs normalization and standardization processing to form a time slice of the multi-layer convolutional neural network (CNN); The multi-layer convolutional neural network (CNN), including multiple convolutional layers, activation function layers, and pooling layers, processes the output signal of the tracking differentiator through multiple layers to extract high-level features of the input data. Finally, the feature vector is output through a fully connected layer. The feature fusion module fuses the features extracted by the multi-layer convolutional neural network (CNN) with other input features of the adaptive neural-fuzzy control system (ANFIS) and inputs them into the adaptive neural-fuzzy control system (ANFIS) for adaptive fuzzy reasoning and decision-making. A BP neural network is used to optimize the parameters of the nonlinear error feedback law module and the extended state observer in the second-order nonlinear active disturbance rejection torque module. The superposition module superimposes the compensation torque output by the controller onto the torque output by the PI controller to obtain the sum of torques. The cutting motor speed adjustment module takes the sum of torques as input to the motor controller, obtains the output of the motor controller, and adjusts the cutting motor speed in real time based on the output.

Images