Method for optimizing FOPTID control parameters of a hydraulic jacking system

Abstract

The present application relates to a kind of FOPTID control parameter optimization method of hydraulic jacking system, belong to FOPTID control technical field, solve the problem that PID controller in prior art cannot be applied to complex environment and cannot realize control parameter adaptive adjustment.Construct the simulation model of hydraulic jacking system, based on the simulation model of the hydraulic jacking system described sample data set is constructed;Machine learning model is constructed, and the machine learning model is trained based on the sample data set;Based on the machine learning model, each predicted drag jacking angle from the start of lifting operation to target time is obtained, each actual drag jacking angle from the start of lifting operation to target time is obtained based on actual hydraulic jacking system, FOPTID control parameter is optimized based on each predicted drag jacking angle and each actual drag jacking angle, and the control amount output by actual hydraulic jacking system is calculated based on the FOPTID control parameter obtained by optimization to control drag jacking frame to reach the expected angle.A kind of control parameter optimization method suitable for complex environment and capable of adaptive adjustment proportional coefficient is realized.

Description

Technical Field

[0001] This invention relates to the field of FOPTID control technology, and in particular to a method for optimizing FOPTID control parameters of a hydraulic support system. Background Technology

[0002] The hydraulic erection system is an important component of the vehicle, mainly responsible for adjusting the pitch and tilt of the erection frame. It typically includes components such as hydraulic cylinders, hydraulic pumps, directional valves, and throttle valves. Based on the pressure of hydraulic oil, it pushes the piston of the cylinder to achieve the erection and support of the object. The control output of its controller directly restricts the control accuracy and operational performance of the system.

[0003] Integer-order PID controllers are commonly used to regulate the control output of hydraulic lifting systems due to their simple structure and ease of implementation. However, in complex environments, hydraulic lifting systems are nonlinear, and system parameters change over time. Integer-order PID controllers have limited robustness to parameter changes and external disturbances, resulting in limited effectiveness in handling nonlinear time-varying systems and thus degraded control performance. For situations requiring precise description of system dynamics and handling complex behaviors, fractional-order systems, such as the FOPTID controller, may be more suitable. Furthermore, the proportional gain in a controller refers to the proportional relationship between the controller's output and the error. Different proportional gains are often required for different scenarios to regulate the hydraulic lifting system. However, this regulation process is mostly manual, which significantly reduces production efficiency and wastes considerable time and resources, making automation difficult.

[0004] Therefore, there is a need to provide a control parameter optimization method that can adapt to complex environments and adaptively adjust the proportional coefficient. Summary of the Invention

[0005] Based on the above analysis, the present invention aims to provide a method for optimizing the FOPTID control parameters of a hydraulic lifting system, in order to solve the problem that existing PID controllers are not suitable for complex environments and cannot achieve adaptive adjustment of control parameters.

[0006] This invention provides a method for optimizing the FOPTID control parameters of a hydraulic lifting system, including:

[0007] Construct a simulation model of the hydraulic support system, and build a sample dataset based on the simulation model of the hydraulic support system;

[0008] A machine learning model is constructed and trained based on the sample dataset. The initial angle of the towing frame, the execution time of the lifting operation, the current cylinder inlet pressure, cylinder outlet pressure, towing frame angular velocity, and towing frame acceleration in the sample dataset are used as the input data of the machine learning model. The current towing frame angle in the sample dataset is used as the true value of the output data of the machine learning model.

[0009] Based on the machine learning model, the predicted towing frame angles from the start of the lifting operation to the target time are obtained. Based on the actual hydraulic towing frame system, the actual towing frame angles from the start of the lifting operation to the target time are obtained. Based on the predicted towing frame angles and the actual towing frame angles, the FOPTID control parameters are optimized. Based on the optimized FOPTID control parameters, the control quantity output by the actual hydraulic towing frame system is calculated to control the towing frame to reach the expected angle.

[0010] A further improvement to the above method is the construction of a simulation model of the hydraulic support system, which includes: collecting the dynamic model parameters of the actual hydraulic support system, and using computer simulation software to construct a simulation model of the hydraulic support system based on the dynamic model parameters.

[0011] A further improvement to the above method, the construction of the sample dataset based on the simulation model of the hydraulic support system, includes:

[0012] Multiple sets of different simulation parameters are constructed, including the initial angle of the towing frame, the initial inlet pressure of the hydraulic cylinder, the initial outlet pressure of the hydraulic cylinder, the initial angular velocity of the towing frame, the initial acceleration of the towing frame, and the desired angle of the towing frame.

[0013] The simulation model of the hydraulic lifting system performs simulation operations on each set of simulation parameters until the lifting frame reaches the desired angle, and stores the cylinder inlet pressure, cylinder outlet pressure, lifting frame angular velocity, lifting frame acceleration, lifting operation execution time, and lifting frame angle at each moment during each simulation process.

[0014] The initial angle of the towing frame, the inlet pressure of the hydraulic cylinder, the outlet pressure of the hydraulic cylinder, the angular velocity of the towing frame, the acceleration of the towing frame, the execution time of the lifting operation, and the angle of the towing frame at each moment are taken as one sample data in the sample dataset.

[0015] Based on a further improvement of the above method, the construction of the machine learning model includes: obtaining the hyperparameters of the neural network model based on the particle swarm optimization algorithm.

[0016] A further improvement to the above method, the acquisition of hyperparameters of the neural network model based on the particle swarm optimization algorithm, includes:

[0017] A1: Initialize the particle swarm, treating the fixed learning rate parameter, neuron weight parameter, and batch size parameter in the neural network model as a single particle;

[0018] A2: Calculate the fitness function for each particle to update the individual optimal and global optimal values, and determine whether the termination condition is met. If it is met, execute A4; otherwise, execute A3.

[0019] The fitness function is:

[0020]

[0021] The predicted towing angle y of the i-th sample data output by the neural network model built with fixed learning rate parameters, neuron weight parameters, and batch size parameters for the current particle. i Let be the angle of the gantry frame in the i-th sample data;

[0022] A3: Update the velocity and position of each particle, and return to A3;

[0023] A4: Output the optimal position, where the particle corresponding to the optimal position is the optimal fixed learning rate parameter, neuron weight parameter, and batch size parameter.

[0024] Based on a further improvement to the above method, the velocity of each particle is updated in the following manner:

[0025] v(t+1)=ωv(t)+c1r1(px(t))+c2r2(gx(t)),

[0026]

[0027] c2 = c max -c1,

[0028] The position of each particle is updated in the following way:

[0029] x(t+1)=x(t)+v(t+1),

[0030] Where c1 and c2 are learning factors, c min c max Let r1 and r2 be the minimum and maximum values ​​of the learning factor c1, respectively; r1 and r2 be random numbers; v(t+1) be the particle's updated velocity; v(t) be the particle's current velocity; p be the particle's individual optimal position; g be the global optimal position of the entire swarm; x(t) be the particle's current position; x(t+1) be the particle's updated position; and ω be the adaptive inertia weight. min ω is the minimum value of the inertial weight. maxy represents the maximum inertia weight, d and k are adjustment coefficients, t is the current iteration number, and M is the maximum iteration number; best The predicted towing frame angle is generated by a neural network model built based on the best particle in the current swarm; and a corresponding neural network model is built for each particle in the current swarm, with the average of the predicted towing frame angles output by each neural network model serving as the basis for the prediction.

[0031] A further improvement to the above method is that, after obtaining the trained neural network model, the method also includes: performing a pruning operation on it.

[0032] Based on a further improvement of the above method, the pruning operation includes:

[0033] Obtain the weight set of each hidden layer in the neural network model, optimize the weight set of each layer based on the particle swarm optimization algorithm, calculate the weight change ratio of each weight in the weight set, and sort the weight change ratio of each layer in ascending order to obtain the first sorting result.

[0034] The importance of each hidden layer is calculated based on sensitivity analysis, and the hidden layers are sorted in order of increasing importance to obtain the second ranking result;

[0035] The pruning operation is performed based on the first sorting result and the second sorting result.

[0036] A further improvement to the above method, the step of performing the pruning operation based on the first sorting result and the second sorting result, includes:

[0037] B1: Obtain the first hidden layer in the second sorting result and obtain the first sorting result corresponding to the hidden layer. Delete neurons in the first sorting result that are greater than a preset ratio threshold and calculate whether the current pruning rate meets the preset pruning threshold. If it does, stop pruning. If it does not, execute B2.

[0038] B2: Obtain the next hidden layer in the second sorting result and obtain the first sorting result corresponding to the hidden layer. Delete neurons in the first sorting result that are greater than a preset proportion threshold, and calculate whether the current pruning rate meets the preset pruning threshold. If it does, stop pruning. If it does not, repeat B2 until the preset pruning threshold is met.

[0039] A further improvement to the above method, the optimization of the FOPTID control parameters based on the predicted tow frame angles and the actual tow frame angles, includes:

[0040] C1: Generate the initial population, where each individual {K} in the population... p Kt K i K d} each represents a possible solution;

[0041] C2: Calculate the fitness function for each individual, where the fitness function is:

[0042]

[0043] M1(t)=K p *(target-y(t))+K t *(target-y(t))′+K i ∫(target-y(t))+K d *(target-y(t))″,

[0044]

[0045] in, Let y(t) be the predicted angle of the towing frame at time t, y(t) be the actual angle of the towing frame at time t, target be the expected angle, and J be the moment of inertia of the towing frame. The lifting angular velocity of the towing frame, f is the angular acceleration of the lifting frame. v f is the coefficient of viscous friction. c The coefficient of friction is Coulomb. For dragging the frame at angular velocity Let α be the normal force on the friction surface, α be the clamping force between the piston rod and the towing frame, L3 be the distance from the hinge point of the towing frame to the point of action of the lifting cylinder, L4 be the distance from the hinge point of the towing frame to the center of mass of the towing frame, g be the acceleration due to gravity, m be the mass of the towing frame, and β0 be the initial angle between the towing frame and the horizontal plane. For dynamic perturbation, t is the execution time of the lifting operation, and T is the target time.

[0046] C3: Use the tournament selection algorithm to select individuals with large fitness function values ​​from the current population to enter the next generation;

[0047] C4: Perform crossover and mutation operations to generate a new generation of population;

[0048] C5: Determine if the current iteration meets the preset termination condition. If it does, end the iteration; otherwise, return to C3.

[0049] C6: Select the individual with the largest fitness function value as the optimized parameter and set it as the FOPTID control parameter of the hydraulic support system.

[0050] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0051] 1. This invention provides a method for optimizing the FOPTID control parameters of a hydraulic erecting system. A fractional-order PID controller is used to control the hydraulic erecting system, further improving its adaptability to complex environments and its robustness. During control optimization, a constructed machine learning model can obtain the erecting frame angle under different parameter conditions, eliminating the need to obtain relevant parameters from the actual hydraulic erecting system, thus improving the efficiency of control parameter optimization. Furthermore, the optimization of FOPTID control parameters based on a genetic algorithm improves the accuracy of obtaining these parameters. Simultaneously, it enables adaptive adjustment in complex scenario changes, significantly reducing unnecessary energy waste and improving the overall efficiency of the hydraulic erecting system.

[0052] 2. This invention provides a method for optimizing the FOPTID control parameters of a hydraulic lifting system. When constructing the machine learning model, an optimized particle swarm optimization algorithm is used to optimize the hyperparameters of the machine learning model, finding the optimal learning rate, neuron weights, batch size, and other parameters, further improving the construction efficiency and prediction accuracy of the machine learning model. After obtaining the target machine learning model, pruning operations further reduce computational load and memory requirements, improve model efficiency and generalization ability, and reduce energy consumption. Furthermore, for the particle swarm optimization algorithm, by setting a custom adjustment method for weights and learning factors, the particle swarm optimization algorithm achieves better convergence speed and convergence accuracy.

[0053] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0054] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0055] Figure 1 This is an example diagram illustrating a method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to an embodiment of the present invention. Detailed Implementation

[0056] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0057] A specific embodiment of the present invention discloses a method for optimizing the FOPTID control parameters of a hydraulic lifting system, such as... Figure 1 As shown, it includes:

[0058] S1: Construct a simulation model of the hydraulic support system, and build a sample dataset based on the simulation model of the hydraulic support system.

[0059] The construction of the hydraulic support system simulation model includes: collecting the dynamic model parameters of the actual hydraulic support system, and using computer simulation software to construct the hydraulic support system simulation model based on the dynamic model parameters.

[0060] A hydraulic erection system typically consists of components such as an electric motor, hydraulic pump, electro-hydraulic proportional valve, multi-stage erection hydraulic cylinder, erection frame, tilt sensor, vehicle body support, electronic controller, hydraulic lock, safety valve, and suction filter. Its working principle is as follows: The electronic controller receives the erection reference angle command calculated by the host computer, uses the tilt sensor to measure the launcher angle information in real time, and uses two pressure sensors to acquire the pressure information at the inlet and outlet of the erection cylinder in real time. The obtained signals are fed back to the control module, where a control algorithm calculates the control quantity. This control quantity is amplified by a power amplifier circuit and then used to adjust the opening of the electro-hydraulic proportional valve core, thereby controlling the pressure and flow rate in the two chambers of the erection cylinder, and ultimately achieving control over the piston displacement of the erection cylinder and the angle of the launcher.

[0061] The dynamic model parameters of the hydraulic erection system are collected by various sensors installed on the system. A simulation model of the hydraulic erection system is then constructed using computer simulation software based on these parameters. Specifically, the physical framework of the hydraulic erection system is established in the simulation software. Based on the collected parameters, the physical laws governing the system, and the system's operating characteristics, equations describing the system's dynamic behavior are written, thereby constructing the hydraulic erection system model. The simulation software can be MATLAB / Simulink, ADAMS, AMESim, etc.; the sensors can include acceleration sensors, displacement sensors, pressure sensors, flow sensors, tilt sensors, and distance sensors, etc. It is understood that the process of constructing the simulation model of the hydraulic erection system is not the focus of this invention; as long as a hydraulic erection system constructed by someone skilled in the art based on their technical knowledge meets the requirements, it is acceptable.

[0062] The sample dataset constructed based on the simulation model of the hydraulic support system includes:

[0063] N1: Construct multiple sets of different simulation parameters, including the initial angle of the towing frame, the initial inlet pressure of the hydraulic cylinder, the initial outlet pressure of the hydraulic cylinder, the initial angular velocity of the towing frame, the initial acceleration of the towing frame, and the desired angle of the towing frame.

[0064] For machine learning models, the more training sample data there is, the more accurate the prediction results will be. Typically, technicians can obtain the initial angle, initial cylinder inlet pressure, initial cylinder outlet pressure, initial tow frame angular velocity, initial tow frame acceleration, and the desired angle of the tow frame based on experience. They can then generate multiple sets of different simulation parameters based on the value ranges of each parameter and a random generation algorithm, but it is essential to ensure that each simulation parameter is within its valid range.

[0065] N2: The simulation model of the hydraulic lifting system performs simulation operations on each set of simulation parameters until the lifting frame reaches the desired angle, and stores the cylinder inlet pressure, cylinder outlet pressure, lifting frame angular velocity, lifting frame acceleration, lifting operation execution time, and lifting frame angle at each moment during each simulation process.

[0066] Each set of simulation parameters is used as input parameters for the simulation model of the hydraulic lifting system. The parameters at each moment are recorded and saved to the database for easy retrieval later.

[0067] N3: The initial angle of the towing frame, the inlet pressure of the hydraulic cylinder, the outlet pressure of the hydraulic cylinder, the angular velocity of the towing frame, the acceleration of the towing frame, the execution time of the lifting operation, and the angle of the towing frame at each moment are taken as a sample data in the sample dataset.

[0068] For example, the number of sample data can be 5,000 to 10,000.

[0069] S2: Construct a machine learning model, train the machine learning model based on the sample dataset, and use the initial angle of the towing frame, the execution time of the towing operation, the current cylinder inlet pressure, cylinder outlet pressure, towing frame angular velocity, and towing frame acceleration in the sample dataset as input data for the machine learning model, and use the current towing frame angle in the sample dataset as the true value of the output data of the machine learning model.

[0070] This invention does not limit the type of machine learning model; it can be a neural network model, a deep network model, etc.

[0071] After selecting a machine learning model, data preprocessing is required for the sample dataset. Preprocessing operations include: data cleaning (duplicate removal, handling missing values, handling outliers, handling erroneous values, etc.), data transformation (data normalization, data encoding, data integration, data type conversion, etc.), feature selection and feature extraction (feature selection, feature extraction, feature evaluation, feature dimensionality reduction, etc.).

[0072] After preprocessing, the initial angle of the towing frame during the lifting operation, the execution time of the lifting operation, and the current cylinder inlet pressure, cylinder outlet pressure, towing frame angular velocity, and towing frame acceleration from each sample are used as input data for the model. The current towing frame angle from that sample is used as the ground truth value for the model's output data. The machine learning model is then trained to achieve a preset accuracy. Specifically, a loss function is calculated based on the model's predicted towing frame angle and the ground truth angle from the sample. The model parameters are then updated in reverse based on the loss function. The trained model is obtained when the loss function converges.

[0073] For example, the machine learning model is a neural network model. Furthermore, after designers select the type of neural network model based on experience and set the model structure (parameters such as the number of neurons in the input layer, the number of neurons in the output layer, and the number of hidden layers), they can obtain the hyperparameters of the neural network model based on the particle swarm optimization algorithm before training the model.

[0074] The process of obtaining hyperparameters for a neural network model based on the particle swarm optimization algorithm includes:

[0075] A1: Initialize the particle swarm, treating the fixed learning rate parameter, neuron weight parameter, and batch size parameter in the neural network model as a single particle;

[0076] A2: Calculate the fitness function for each particle to update the individual optimal and global optimal values, and determine whether the termination condition is met. If it is met, execute A4; otherwise, execute A3.

[0077] The fitness function is:

[0078]

[0079] The predicted towing angle y of the i-th sample data output by the neural network model built with fixed learning rate parameters, neuron weight parameters, and batch size parameters for the current particle. i Let be the angle of the gantry frame in the i-th sample data;

[0080] A3: Update the velocity and position of each particle, and return A3; further, update the velocity of each particle in the following way:

[0081] v(t+1)=ωv(t)+c1r1(px(t))+c2r2(gx(t)),

[0082]

[0083] c2 = c max -c1,

[0084] The position of each particle is updated in the following way:

[0085] x(t+1)=x(t)+v(t+1),

[0086] Where c1 and c2 are learning factors, c min c max Let r1 and r2 be the minimum and maximum values ​​of the learning factor c1, respectively; r1 and r2 be random numbers; v(t+1) be the particle's updated velocity; v(t) be the particle's current velocity; p be the particle's individual optimal position; g be the global optimal position of the entire swarm; x(t) be the particle's current position; x(t+1) be the particle's updated position; and ω be the adaptive inertia weight. min ω is the minimum value of the inertial weight. max y represents the maximum inertia weight, d and k are adjustment coefficients, t is the current iteration number, and M is the maximum iteration number; best The predicted towing frame angle is generated by a neural network model built based on the best particle in the current swarm; and a corresponding neural network model is built for each particle in the current swarm, with the average of the predicted towing frame angles output by each neural network model serving as the basis for the prediction.

[0087] A4: Output the optimal position, where the particle corresponding to the optimal position is the optimal fixed learning rate parameter, neuron weight parameter, and batch size parameter.

[0088] When building machine learning models, an optimized particle swarm optimization (PSO) algorithm is used to optimize the hyperparameters of the machine learning model, in order to find the optimal learning rate, neuron weights, batch size, and other parameters, thereby further improving the efficiency of machine learning model construction and prediction accuracy. Furthermore, by setting custom adjustment methods for weights and learning factors, the PSO algorithm achieves better convergence speed and accuracy.

[0089] Understandably, the values ​​of parameters such as population parameters, minimum and maximum values ​​of inertia weight, minimum and maximum values ​​of learning factor, maximum number of iterations, and adjustment coefficient in the particle swarm optimization algorithm can be set according to actual needs, and this invention does not impose any limitations on them.

[0090] Furthermore, after obtaining the trained neural network model, pruning operations are performed on it.

[0091] The pruning operation includes:

[0092] M1: Obtain the weight set of each hidden layer in the neural network model, optimize the weight set of each layer based on the particle swarm optimization algorithm, calculate the weight change ratio of each weight in the weight set, and sort the weight change ratio of each layer from smallest to largest to obtain the first sorting result.

[0093] The optimization of the weight set for each layer based on the particle swarm optimization algorithm, and the calculation of the weight change ratio for each weight in the weight set, include:

[0094] P1: Initialize the particle swarm, use the weights of this layer as the original weights, and use the original weights as a particle. Based on the random generation algorithm, randomly select values ​​in the range of 0 to 1 to generate multiple particles so that the total number of particles meets the preset quantity requirement.

[0095] For example, the preset quantity is 15.

[0096] P2: Calculate the fitness function of each particle to update the individual optimal and global optimal values, and determine whether the termination condition is met. If it is met, proceed to P4; otherwise, proceed to P3.

[0097] The fitness function is:

[0098]

[0099] y new_num To use the corresponding value of the particle as the weight value of the neuron in that layer, the result is obtained by using a neural network model to predict the num-th sample data. i_num The result is obtained by using the original weights of this layer as the weight values ​​of the neurons in this layer, and predicting the num-th sample data using a neural network model. Let num be the angle of the trolley corresponding to the num-th sample data, where num = 1, 2, 3, ..., N.

[0100] P3: Update the velocity and position of each particle, then return to P2.

[0101] P4: Output the optimal position; the particle corresponding to the optimal position is the optimal weight for that layer. P5: Calculate the weight change ratio for each neuron in that layer.

[0102]

[0103] w i_new For the optimal weights of this neuron, w i_old is the original weight of the neuron, and i is the number of neurons in the layer, i = 1, 2, 3, ..., N.

[0104] P6: Perform steps P1-P5 for each hidden layer to obtain the weight change ratio of each neuron in each layer.

[0105] This invention, by applying the particle swarm optimization (PSO) algorithm, obtains the degree of influence (i.e., weight change ratio) of each neuron in each layer on the prediction result, indicating the sensitivity of each neuron to accuracy improvement. A large weight change ratio indicates that the neuron's weights need to be significantly altered to achieve the preset accuracy, thus indicating low importance. Conversely, a small weight change ratio indicates that the neuron's weights only need slight adjustments to achieve the preset accuracy, thus indicating high importance. The speed and position update methods in the PSO algorithm can be implemented using the methods mentioned in step A3, or the update methods defined in the unoptimized PSO algorithm.

[0106] M2: Based on sensitivity analysis, the importance of each hidden layer is calculated, and the hidden layers are sorted in ascending order of importance to obtain the second ranking result.

[0107] The importance of sensitivity analysis can be calculated using methods such as differential analysis, model output variation analysis, and variance decomposition. This invention does not limit these methods, as long as the importance of each layer in the shared layer can be correctly obtained.

[0108] M3: Perform pruning operations based on the first sorting result and the second sorting result. The pruning operations based on the first sorting result and the second sorting result include:

[0109] B1: Obtain the first hidden layer in the second sorting result and obtain the first sorting result corresponding to the hidden layer. Delete neurons in the first sorting result that are greater than a preset ratio threshold and calculate whether the current pruning rate meets the preset pruning threshold. If it does, stop pruning. If it does not, execute B2.

[0110] B2: Obtain the next hidden layer in the second sorting result and obtain the first sorting result corresponding to the hidden layer. Delete neurons in the first sorting result that are greater than a preset proportion threshold, and calculate whether the current pruning rate meets the preset pruning threshold. If it does, stop pruning. If it does not, repeat B2 until the preset pruning threshold is met.

[0111] For example, the preset pruning threshold is 15%.

[0112] S3: Based on the machine learning model, obtain each predicted towing frame angle from the start of the lifting operation to the target time; based on the actual hydraulic towing frame system, obtain each actual towing frame angle from the start of the lifting operation to the target time; optimize the FOPTID control parameters based on the predicted towing frame angles and the actual towing frame angles; and calculate the control quantity output by the actual hydraulic towing frame system based on the optimized FOPTID control parameters to control the towing frame to reach the expected angle.

[0113] The optimization of FOPTID control parameters based on the predicted tow frame angles and the actual tow frame angles includes:

[0114] C1: Generate the initial population, where each individual {K} in the population... p K t K i K d} each represents a possible solution;

[0115] C2: Calculate the fitness function for each individual, where the fitness function is:

[0116]

[0117] M1(t)=K p *(target-y(t))+K t *(target-y(t))′+K i ∫(target-y(t))+K d *(target-y(t))″,

[0118]

[0119] in, Let y(t) be the predicted angle of the towing frame at time t, y(t) be the actual angle of the towing frame at time t, target be the expected angle, and J be the moment of inertia of the towing frame. The lifting angular velocity of the towing frame, f is the angular acceleration of the lifting frame. v f is the coefficient of viscous friction. c The coefficient of friction is Coulomb. For dragging the frame at angular velocity Let α be the normal force on the friction surface, α be the clamping force between the piston rod and the towing frame, L3 be the distance from the hinge point of the towing frame to the point of action of the lifting cylinder, L4 be the distance from the hinge point of the towing frame to the center of mass of the towing frame, g be the acceleration due to gravity, m be the mass of the towing frame, and β0 be the initial angle between the towing frame and the horizontal plane. For dynamic perturbation, t is the execution time of the lifting operation, and T is the target time.

[0120] C3: Use the tournament selection algorithm to select individuals with large fitness function values ​​from the current population to enter the next generation;

[0121] C4: Perform crossover and mutation operations to generate a new generation of population;

[0122] C5: Determine if the current iteration meets the preset termination condition. If it does, end the iteration; otherwise, return to C3.

[0123] C6: Select the individual with the largest fitness function value as the optimized parameter and set it as the FOPTID control parameter of the hydraulic support system.

[0124] Understandably, operations such as crossover and mutation in genetic algorithms are common knowledge in the field, and this invention does not limit them.

[0125] After completing steps C1-C6, the FOPTID control parameters obtained through genetic algorithm optimization are used as the control parameters of the actual hydraulic support system, so that the control quantity output by the hydraulic support system can control the support frame to reach the expected angle.

[0126] Compared with existing technologies, this embodiment provides a method for optimizing the FOPTID control parameters of a hydraulic erecting system. It employs a fractional-order PID controller to control the hydraulic erecting system, further enhancing its adaptability to complex environments and its robustness. When optimizing control inputs, the constructed machine learning model can obtain the erecting frame angle under different parameter conditions, eliminating the need to obtain relevant parameters from the actual hydraulic erecting system, thus improving the efficiency of control parameter optimization. Furthermore, the optimization of FOPTID control parameters based on a genetic algorithm improves the accuracy of obtaining these parameters and enables adaptive adjustment in complex scenarios, significantly reducing unnecessary energy waste and improving the efficiency of the hydraulic erecting system. When constructing the machine learning model, an optimized particle swarm optimization algorithm is used to optimize the hyperparameters of the machine learning model, finding the optimal learning rate, neuron weights, batch size, and other parameters, further improving the construction efficiency and prediction accuracy of the machine learning model. After obtaining the target machine learning model, pruning operations further reduce computational load and memory requirements, improving model efficiency and generalization ability while reducing energy consumption. Furthermore, for the particle swarm optimization algorithm, by setting custom adjustment methods for weights and learning factors, the particle swarm optimization algorithm can achieve better convergence speed and convergence accuracy.

[0127] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0128] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing FOPTID control parameters of a hydraulic vertical support system, characterized in that, include: Construct a simulation model of the hydraulic support system, and build a sample dataset based on the simulation model of the hydraulic support system; A machine learning model is constructed and trained based on the sample dataset. The initial angle of the towing frame, the execution time of the lifting operation, the current cylinder inlet pressure, cylinder outlet pressure, towing frame angular velocity, and towing frame acceleration in the sample dataset are used as the input data of the machine learning model. The current towing frame angle in the sample dataset is used as the true value of the output data of the machine learning model. Based on the machine learning model, the predicted towing frame angles from the start of the lifting operation to the target time are obtained. Based on the actual hydraulic towing frame system, the actual towing frame angles from the start of the lifting operation to the target time are obtained. Based on the predicted towing frame angles and the actual towing frame angles, the FOPTID control parameters are optimized. Based on the optimized FOPTID control parameters, the control quantity output by the actual hydraulic towing frame system is calculated to control the towing frame to reach the expected angle.

2. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 1, characterized in that, The construction of the simulation model of the hydraulic support system includes: The dynamic model parameters of the actual hydraulic support system were collected, and a simulation model of the hydraulic support system was constructed based on the dynamic model parameters using computer simulation software.

3. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 2, characterized in that, The sample dataset constructed based on the simulation model of the hydraulic support system includes: Multiple sets of different simulation parameters are constructed, including the initial angle of the towing frame, the initial inlet pressure of the hydraulic cylinder, the initial outlet pressure of the hydraulic cylinder, the initial angular velocity of the towing frame, the initial acceleration of the towing frame, and the desired angle of the towing frame. The simulation model of the hydraulic lifting system performs simulation operations on each set of simulation parameters until the lifting frame reaches the desired angle, and stores the cylinder inlet pressure, cylinder outlet pressure, lifting frame angular velocity, lifting frame acceleration, lifting operation execution time, and lifting frame angle at each moment during each simulation process. The initial angle of the towing frame, the inlet pressure of the hydraulic cylinder, the outlet pressure of the hydraulic cylinder, the angular velocity of the towing frame, the acceleration of the towing frame, the execution time of the lifting operation, and the angle of the towing frame at each moment are taken as one sample data in the sample dataset.

4. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 3, characterized in that, The construction of the machine learning model includes: Hyperparameters of neural network models are obtained using the particle swarm optimization algorithm.

5. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 4, characterized in that, The process of obtaining hyperparameters for a neural network model based on the particle swarm optimization algorithm includes: A1: Initialize the particle swarm, treating the fixed learning rate parameter, neuron weight parameter, and batch size parameter in the neural network model as a single particle; A2: Calculate the fitness function for each particle to update the individual optimal and global optimal values, and determine whether the termination condition is met. If it is met, execute A4; otherwise, execute A3. The fitness function is: The predicted towing angle y of the i-th sample data output by the neural network model built with fixed learning rate parameters, neuron weight parameters, and batch size parameters for the current particle. i Let be the angle of the gantry frame in the i-th sample data; A3: Update the velocity and position of each particle, and return to A3; A4: Output the optimal position, where the particle corresponding to the optimal position is the optimal fixed learning rate parameter, neuron weight parameter, and batch size parameter.

6. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 5, characterized in that, include: The velocity of each particle is updated in the following way: v(t+1)=ωv(t)+c1r1(px(t))+c2r2(gx(t)), c2=c max -c1, The position of each particle is updated in the following way: x(t+1)=x(t)+v(t+1), Where c1 and c2 are learning factors, c min c max Let r1 and r2 be the minimum and maximum values ​​of the learning factor c1, respectively; r1 and r2 be random numbers; v(t+1) be the particle's updated velocity; v(t) be the particle's current velocity; p be the particle's individual optimal position; g be the global optimal position of the entire swarm; x(t) be the particle's current position; x(t+1) be the particle's updated position; and ω be the adaptive inertia weight. min ω is the minimum value of the inertial weight. max y represents the maximum inertia weight, d and k are adjustment coefficients, t is the current iteration number, and M is the maximum iteration number; best The predicted towing frame angle is generated by a neural network model built based on the best particle in the current swarm; and a corresponding neural network model is built for each particle in the current swarm, with the average of the predicted towing frame angles output by each neural network model serving as the basis for the prediction.

7. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 6, characterized in that, The method further includes: After obtaining the trained neural network model, pruning operations are performed on it.

8. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 7, characterized in that, The pruning operation includes: Obtain the weight set of each hidden layer in the neural network model, optimize the weight set of each layer based on the particle swarm optimization algorithm, calculate the weight change ratio of each weight in the weight set, and sort the weight change ratio of each layer in ascending order to obtain the first sorting result. The importance of each hidden layer is calculated based on sensitivity analysis, and the hidden layers are sorted in order of increasing importance to obtain the second ranking result; The pruning operation is performed based on the first sorting result and the second sorting result.

9. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 8, characterized in that, The pruning operation based on the first sorting result and the second sorting result includes: B1: Obtain the first hidden layer in the second sorting result and obtain the first sorting result corresponding to the hidden layer. Delete neurons in the first sorting result that are greater than a preset ratio threshold and calculate whether the current pruning rate meets the preset pruning threshold. If it does, stop pruning. If it does not, execute B2. B2: Obtain the next hidden layer in the second sorting result and obtain the first sorting result corresponding to the hidden layer. Delete neurons in the first sorting result that are greater than a preset proportion threshold, and calculate whether the current pruning rate meets the preset pruning threshold. If it does, stop pruning. If it does not, repeat B2 until the preset pruning threshold is met.

10. The method for optimizing the FOPTID control parameters of a hydraulic vertical support system according to claim 1, characterized in that, The optimization of FOPTID control parameters based on the predicted tow frame angles and the actual tow frame angles includes: C1: Generate the initial population, where each individual {K} in the population... p K t K i K d } each represents a possible solution; C2: Calculate the fitness function for each individual, where the fitness function is: M1(t)=K p *(target-y(t))+K t *(target-y(t))′+K i ∫(target-y(t))+K d *(target-y(t))″, in, Let y(t) be the predicted angle of the towing frame at time t, y(t) be the actual angle of the towing frame at time t, target be the expected angle, and J be the moment of inertia of the towing frame. The lifting angular velocity of the towing frame, f is the angular acceleration of the lifting frame. v f is the coefficient of viscous friction. c The coefficient of friction is Coulomb. For dragging the frame at angular velocity Let α be the normal force on the friction surface, α be the clamping force between the piston rod and the towing frame, L3 be the distance from the hinge point of the towing frame to the point of action of the lifting cylinder, L4 be the distance from the hinge point of the towing frame to the center of mass of the towing frame, g be the acceleration due to gravity, m be the mass of the towing frame, and β0 be the initial angle between the towing frame and the horizontal plane. For dynamic perturbation, t is the execution time of the lifting operation, and T is the target time. C3: Use the tournament selection algorithm to select individuals with large fitness function values ​​from the current population to enter the next generation; C4: Perform crossover and mutation operations to generate a new generation of population; C5: Determine if the current iteration meets the preset termination condition. If it does, end the iteration; otherwise, return to C3. C6: Select the individual with the largest fitness function value as the optimized parameter and set it as the FOPTID control parameter of the hydraulic support system.

Images