Load power prediction method for servo pump control system of loader and servo pump control system

Abstract

The invention relates to the technical field of fluid machinery control, and provides a load power prediction method of a loading machine servo pump control system and the servo pump control system.The method comprises the steps that S1, wavelet decomposition and clustering analysis are conducted on real-time operation data of a loading machine, and a training set and a test set are obtained; s2, setting and training an RBF neural network model according to a clustering analysis result of the loader load signal data; s3, predicting load power by using an RBF neural network model, and performing multi-target optimization matching to obtain control parameters; and S4, executing optimal matching control of the servo pump control system of the loading machine. The loader working system load power prediction model is established through the RBF neural network, periodic load power prediction of the loader working system is achieved, the rotating speed of the servo motor and the displacement of the variable pump are cooperatively controlled through the multi-target optimization algorithm, accurate control over the output power of the power source and flow distribution of the multi-way valve is achieved, and the reliability of the system is improved. And a servo pump control system of the loading machine keeps good operation.

Description

Technical Field

[0001] This invention relates to the field of fluid machinery control technology, specifically to a load power prediction method and a servo pump control system for a loader. Background Technology

[0002] In the field of electromechanical control, the dynamic characteristics of the working system load and the energy distribution requirements pose a severe challenge to the precise control of servo pump control systems. The operation of construction machinery involves periodic actions such as digging, lifting, and rotating. Its load is affected by working conditions, such as soil hardness and material weight, as well as operating behavior, exhibiting strong nonlinearity, time-varying characteristics, and intermittent impact characteristics.

[0003] However, the multi-physics coupling characteristics of hydraulic systems make it difficult to model the load power mechanism. Traditional linear control methods, such as PID control, struggle to dynamically match the pump's output power with load demand, resulting in system energy efficiency losses and dynamic response lag. Furthermore, the stringent energy efficiency requirements of electrified construction machinery necessitate dynamic energy-saving optimization of the pump control system through high-precision load prediction. However, noise interference and sudden changes in operating conditions further increase the robustness design difficulty of the prediction model. Secondly, pure electric machinery control places near-stringent demands on energy distribution efficiency. Limited battery capacity directly restricts the equipment's endurance and workload. Therefore, achieving feedforward and dynamic energy-saving optimization of the pump control system through high-precision load prediction becomes a key approach to improving loader energy efficiency. However, the acquisition of load signals is inevitably subject to various noise interferences, and sudden changes in operating conditions during operation can cause drastic jumps in the load signal. These factors significantly increase the design difficulty of the prediction model, requiring it to possess excellent noise suppression capabilities and generalization ability to unknown disturbances. These complex factors make constructing an integer-cycle load power prediction method that balances real-time performance, generalization, and disturbance rejection a key technical bottleneck for upgrading servo pump control systems. Summary of the Invention

[0004] To address the shortcomings of the existing technology, the present invention aims to provide a load power prediction method and a servo pump control system for a loader. By establishing a load power prediction model for the loader's working system through an RBF neural network, the method achieves effective prediction of the periodic load power of the loader's working system. Using a multi-objective optimization algorithm, the method dynamically and collaboratively controls the servo motor speed and variable pump displacement, guiding the controller to achieve precise control of the power source output power and the flow distribution of the multi-way valve. This enables the dual-variable power source output power to track load demand in real time, reducing high-pressure overflow and throttling losses, and achieving dynamic response operation of the loader's servo pump control system throughout the entire operating cycle.

[0005] Specifically, on the one hand, the present invention provides a load power prediction method for a loader servo pump control system, which includes the following steps: S1: Acquire real-time operating data of the loader, perform wavelet decomposition and cluster analysis, and divide it into a training set of loader load signal data. Loader load signal data test set ; S2: Based on the clustering analysis results of the loader load signal data, set the parameters of the hidden layer neurons in the RBF neural network, and obtain the training set of loader load signal data obtained in step S1. Train the RBF neural network model and adjust the output layer weights; use the load power tracking error objective function. Adjust and correct the input layer connection weights until the required number of training iterations is reached; S3: Use the RBF neural network model obtained in step S2 to predict the load power, and generate control parameters for the loader servo pump control system based on the predicted load power; perform multi-objective optimization matching to obtain the optimal control parameter set for the loader servo pump control system. ; S4: Based on the optimal control parameter set determined in step S3 The speed and displacement of the loader's servo pump control system are controlled to achieve optimal matching control of the loader's servo pump control system.

[0006] Preferably, step S2 specifically includes: S21: Obtain the node width parameter of the hidden layer neurons through the nearest neighbor algorithm to adapt to the distribution density of the loader load in the feature space; S22: Choose the diminishing memory recursive least squares method to solve the weight parameters from the hidden layer to the output layer of the neural network, to adapt to the time-varying working conditions of the loader; S23: Obtain the loader's operating conditions and sampled data features, and initialize the RBF neural network parameters; S24: Train the RBF neural network model using the load power tracking error objective function. The input layer connection weights are adjusted and corrected using gradient descent, and the weights are updated iteratively.

[0007] Preferably, the objective function for load power tracking error in step S24 is as follows: ; in, Let the objective function be the load power tracking error. For the first The actual load power value for each load operation mode; For the first The actual load power prediction value for each load operation mode; Connect the weight vectors to the input layer; This represents the number of training samples; This is the sequence number of the load sample.

[0008] Preferably, step S22 specifically includes: ; ; ; in, For the first Update the step size gain vector for each control parameter; For the first Covariance matrix of each control parameter; This is the output control vector of the current hidden layer; Transpose the output control vector of the current hidden layer; It is a gradually diminishing factor; For the first Estimated values ​​of the connection weight vectors of each input layer; This is an index for the number of layers in a neural network. For the first Covariance matrix of each control parameter; For the first Estimates of the input layer connection weight vectors.

[0009] Preferably, in step S24, the gradient descent method is used to iteratively adjust the weights, update the weights, and connect the new input layer to the weight vector. The value assigned is: ; ; in, For the first Input layer connection weight vector for clustering of load job patterns; For the first Training samples for each load job mode; For iterative update rate; For the first The activation function output for each load job pattern cluster; It is an exponential function with the natural constant e as its base; For belonging to the first The number of load samples in each load job pattern cluster; For the first Hidden layer node width feature parameters for clustering of individual load job patterns; Index for cluster groups based on job load patterns; This is the preset total number of cluster centers for load features.

[0010] Preferably, step S3 specifically includes: S31: Generate control parameters for the loader's servo pump control system based on the predicted load power; obtain the predicted load power value sequence using the RBF neural network model. Used for multi-objective optimization matching; S32: Obtain the predicted load power for multi-objective optimization matching, and adjust the control variables of the first-level control parameters. Perform selection, crossover, and update operations to generate secondary control parameters; merge primary and secondary control parameters, and calculate each control variable. The characteristic distance is used to output the optimal control parameter set of the loader servo pump control system. .

[0011] Preferably, the method for obtaining the secondary control variables in step S32 is as follows: ; in, The first control variable is the primary control parameter. The second control variable is the primary control parameter; This is the first control variable for the secondary control parameters after crossover; The second control variable is the secondary control parameter after crossover; These are random numbers generated according to a probability distribution. For time parameters; Use symbols to mark control variables.

[0012] Preferably, each control variable in step S32 Feature distance for: ; in, To control variables The first adjacent variable in the first position The values ​​of each objective function; To control variables The second adjacent variable in the th The values ​​of each objective function; For the first The maximum value of each objective function; For the first The minimum value of an objective function; Number the objective function; For each control variable The characteristic distance.

[0013] Preferably, step S1 specifically includes: S11: Collect real-time operating data during the loader's operation, perform wavelet decomposition, and analyze the detailed parameters of each load signal. Perform soft thresholding to extract detailed parameters of the load signal. The loader load signal data is reconstructed from the original low-frequency parameters and then denoised. ; S12: Acquire the denoised loader load signal data We analyzed the characteristics of the load signal data of the loader and used clustering methods to extract the cluster centers of typical load operation modes. S13: Divide the processed loader load signal data into a loader load signal data training set according to time sequence. Loader load signal data test set .

[0014] On the other hand, the present invention provides a servo pump control system for a loader servo pump control system, which includes: a power source module, a pump control hydraulic circuit module, an actuator module, and a drive control module; The power source module includes a battery and a rectifier-inverter; the battery is connected to the input of the rectifier-inverter via a DC bus to provide and convert drive power; the output of the rectifier-inverter is connected to the input of the permanent magnet synchronous motor. The pump-controlled hydraulic circuit module includes a permanent magnet synchronous motor, an electro-hydraulic proportional variable pump, a relief valve, a first check valve, a second check valve, a back pressure check valve, and an oil tank. The actuator module includes a multi-way valve group, a first hydraulic cylinder, and a second hydraulic cylinder; the first oil inlet of the multi-way valve group is connected to the main pressure oil circuit, and the second oil inlet is connected to the auxiliary circuit; multiple working oil ports of the multi-way valve group are respectively connected to the rodless chamber and the rod chamber of the first hydraulic cylinder and the second hydraulic cylinder, for controlling the movement of the hydraulic cylinder; The drive control module includes a servo driver, a controller, a pressure sensor, and a flow sensor; the output of the servo driver is connected to the control input of the permanent magnet synchronous motor to adjust the motor speed.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention establishes a load power prediction model for the loader working system based on RBF neural network. Wavelet transform theory is used to preprocess the input sample signal to provide highly reliable samples for the neural network model. The parameters of the hidden layer of the neural network model are determined by K-means algorithm, P-nearest algorithm and fading memory recursive least squares algorithm. Finally, the periodic load power of the loader working system is effectively predicted. The predicted data can guide the controller to achieve precise control of the power source output power and the flow distribution of the multi-way valve.

[0016] (2) The present invention uses the load power predicted by the RBF neural network as a feedforward signal and uses multi-objective optimization algorithms such as NSGA-Ⅱ to dynamically and collaboratively control the speed of the servo motor and the displacement of the variable pump, so that the output power of the dual variable power source tracks the load demand in real time and accurately, thereby reducing high pressure overflow and throttling loss from the source and realizing high energy efficiency operation of the loader working system throughout the entire operation cycle.

[0017] (3) The present invention establishes a servo pump control system for a loader. The battery is connected to the rectifier inverter and the permanent magnet synchronous motor via the DC bus to provide power to the hydraulic system. The permanent magnet synchronous motor is connected to the electro-hydraulic proportional variable pump to control the hydraulic cylinder. The controller is connected to the servo driver, the permanent magnet synchronous motor and the electro-hydraulic proportional variable pump variable mechanism to adjust the power distribution of the hydraulic system. A check valve and a relief valve are added to the main circuit to limit the peak pressure of the hydraulic cylinder and quickly release pressure to avoid system overload. A check valve and a back pressure check valve are set in the return oil circuit. During the swing braking, the motor speed is limited by the back pressure to convert the braking energy into electrical energy and feed it back to the battery. At the moment of swing start, the oil on the suction side of the motor is replenished by the check valve to avoid vibration and noise caused by cavitation. Attached Figure Description

[0018] Figure 1 This is a flowchart of the load power prediction method for the electric loader servo pump control system proposed in this invention; Figure 2 This is a schematic diagram of the servo pump control system of the load power prediction method of the loader servo pump control system of the present invention; Figure 3 This is a flowchart of the RBF neural network load power prediction method for the servo pump control system of the loader servo pump control system of the present invention. Figure 4 This is a flowchart of the RBF neural network training process of the servo pump control system for the load power prediction method of the loader servo pump control system of the present invention. Figure 5 This is a structural diagram of the RBF neural network load power prediction model of the servo pump control system of the loader servo pump control system of the present invention. Figure 6 This is a flowchart of the NSGA-Ⅱ multi-objective optimization algorithm of the servo pump control system for the load power prediction method of the loader servo pump control system of the present invention; Figure 7 This is a comparison diagram of the load power prediction method of the servo pump control system of the loader servo pump control system of the present invention, and the RBF neural network prediction of the boom cylinder load pressure. Figure 8 This is a comparison diagram of the RBF neural network for predicting the load pressure of the bucket cylinder in the servo pump control system of the loader servo pump control system of the present invention. Figure 9 This is a comparison diagram of the load power prediction method of the servo pump control system of the loader servo pump control system of the present invention, and the RBF neural network prediction of the boom cylinder load flow. Figure 10 This is a comparison diagram of the RBF neural network for predicting the load flow of the bucket cylinder in the servo pump control system of the loader servo pump control system of the present invention. Figure 11 This is a comparison diagram of the RBF neural network power source output pressure prediction of the servo pump control system of the loader servo pump control system of the present invention. Figure 12 This is a comparison diagram of the RBF neural network power source output total flow prediction of the servo pump control system of the loader servo pump control system of the present invention.

[0019] Key reference numerals: 1. First hydraulic cylinder; 2. Second hydraulic cylinder; 3. Multi-way valve assembly; 41. First check valve; 42. Second check valve; 5. Back pressure check valve; 6. Relief valve; 7. Electro-hydraulic proportional variable pump; 8. Permanent magnet synchronous motor; 9. Rectifier inverter; 10. Battery; 11. Pressure sensor; 12. Flow sensor; 13. Pump variable mechanism; 14. Servo driver; 15. Controller. Detailed Implementation

[0020] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0021] This invention proposes a load power prediction method for an electric loader servo pump control system, such as... Figure 1 As shown, wavelet decomposition and cluster analysis are performed on the real-time operating data of the loader to obtain training and test sets; an RBF neural network model is set and trained based on the cluster analysis results of the loader load signal data; the RBF neural network model is used to predict load power, and multi-objective optimization matching is performed to obtain control parameters; optimal matching control of the loader servo pump control system is executed; specifically, the following steps are included: Step S1: Obtain the real-time operating data of the loader, perform wavelet decomposition and cluster analysis, and divide it into training set and test set.

[0022] Step S11: Collect real-time operating data during the loader's operation as input sample data, and perform sample data preprocessing; the real-time operating data of the loader includes: the load pressure signal of the boom hydraulic cylinder. Load pressure signal of bucket hydraulic cylinder Main pump outlet flow signal Permanent magnet synchronous motor speed signal Multi-way valve pilot control current signal DC bus voltage signal and DC bus current signal In this embodiment of the invention, Daubechies 5 is selected as the wavelet basis function, and the number of decomposition levels is set to 5. Wavelet decomposition is performed on the collected real-time operation data of the loader, specifically as follows: ; in, This refers to the real-time load signal data of the loader, which is the raw time-domain signal, that is, the collected, unprocessed raw load signal sequence, such as the observation data of boom cylinder pressure changing over time. This is a low-frequency signal representing the real-time load data of the loader. This is a high-frequency signal representing the real-time load signal data of the loader. It is a continuous time variable, representing the time coordinates of the signal; The scaling parameter of wavelet decomposition determines the resolution or frequency range of the analysis. The larger the value, the larger the corresponding scale and the lower the center frequency of the analysis. In this embodiment of the invention, the number of decomposition layers is set to 5, that is... The range of values ​​for is related to the specific decomposition structure; The translation parameters for wavelet decomposition determine the position of the wavelet basis functions on the time axis; For the scaling function at the scaling parameter Translation parameters Basis functions; For scaling function The corresponding approximate parameters of the load signal; For wavelet functions at scale parameters Translation parameters Basis functions; To be related to wavelet function The corresponding load signal details.

[0023] For each load signal detailed parameters After performing soft thresholding, the detailed parameters of the processed load signal are as follows: ; ; in, These are the detailed parameters of the load signal after soft thresholding. This is a general threshold; This is an estimate of the standard deviation of the high-frequency parameters corresponding to the current decomposition layer; The length of the high-frequency parameters of the current decomposition layer; For signal functions; To find the maximum value of the function; It is the natural logarithm function.

[0024] Detailed parameters of the load signal after soft threshold processing The loader load signal data is reconstructed from the original low-frequency parameters and then denoised. for: ; in, The loader load signal data after noise reduction; This is an index for high-frequency coefficients.

[0025] Step S12: Obtain the loader load signal data after noise reduction in step S11. The load signal data characteristics of the loader were analyzed, and the basis function centers of the hidden function neurons were determined using the K-means clustering method. The K-means clustering method automatically extracts cluster centers for multiple typical load operation modes from the preprocessed loader load signal data, making the network structure more closely match the actual working conditions, thereby improving the convergence speed and accuracy of the subsequent prediction model. In this embodiment, the number of cluster centers in the hidden layer was determined to be 10; the calculation expression is as follows: ; ; ; in, The nearest center index corresponding to the current load data point; For the first Training samples for each load job mode; Training samples for load job modes; Index for cluster groups based on job load patterns; The load sample number; For the first Cluster center vectors; For belonging to the first The number of load samples in each load job pattern cluster; For the first A sample set with cluster centers; The total number of training samples for the load; The preset total number of cluster centers for load characteristics; This is a function that takes the minimum value.

[0026] Step S13: The loader load signal data collected from the loader servo pump control system and preprocessed in step S11 is divided into a loader load signal data training set according to time sequence. Loader load signal data test set The training set of loader load signal data The clustering analysis in step S12 and the RBF neural network training in step S2 are used. The overall loader load signal data sample set obtained in this embodiment of the invention contains N=26000+ groups, of which the first 15000 groups are the loader load signal data training set. The data is used for training, and the remaining data is used for testing and verification, forming a test set of loader load signal data. .

[0027] Step S2: Based on the clustering analysis results of the loader load signal data, set the parameters of the hidden layer neurons in the RBF neural network, and obtain the training set of loader load signal data obtained in step S1. Train the RBF neural network model.

[0028] Step S21: Obtain the hidden layer neuron node width parameters using the P-norm-based nearest neighbor algorithm to adapt to the load distribution density of the loader in the feature space; calculate the hidden layer node width feature parameters using the P-norm-based nearest neighbor algorithm. Specifically: ; in, For the first Hidden layer node width feature parameters for clustering of individual load job patterns; For the first The first cluster center of the load operation pattern The nearest neighbor center vector; The Euclidean distance between vectors; The load sample number; Number the nearest neighbor center vector.

[0029] Step S22: Select the diminishing memory recursive least squares method to solve for the weight parameters from the hidden layer to the output layer of the neural network, adapting to the time-varying working conditions of the loader; in this embodiment of the invention, the weight parameters from the hidden layer to the output layer of the neural network are set to 0.98, and the specific implementation process is as follows: ; ; ; in, For the first Each control parameter updates the step size gain vector, controlling the update step size; For the first The covariance matrix of each control parameter represents the uncertainty of the estimation; For the first The output control vector of the current hidden layer; For the first Transpose the output control vector of the current hidden layer; The fading factor, with a core parameter range of (0,1], is set to 0.98 to allow the algorithm to "forget" old data and better track system changes. For the first Estimated values ​​of the connection weight vectors of each input layer; This is an index for the number of layers in a neural network. For the first Covariance matrix of each control parameter; For the first Estimates of the input layer connection weight vectors.

[0030] Through simulation experiments, the weight parameters from the hidden layer to the output layer of the neural network were compared. It was found that when the fading factor was set to 0.98, the algorithm achieved the best balance between tracking performance and stability under time-varying conditions.

[0031] Hidden layer output vector The elements are specifically represented as follows: ; in, For the first Each hidden layer output vector; The parameter is the natural base.

[0032] Step S23: Obtain the loader's operating conditions and sampled data features, initialize the RBF neural network parameters, and set the number of nodes in the input layer, hidden layer, and output layer of the RBF neural network model, the initial width of the basis functions of the hidden layer neurons, the connection weights from the input layer to the output layer, the number of training iterations, and the iteration update rate.

[0033] Based on the loader's operating conditions and the characteristics of the sampled data, two complete work cycles are considered as a set of data. The six parameters of the previous cycle—the rodless chamber pressure of the bucket and boom hydraulic cylinders, the inlet flow rate of the bucket and boom hydraulic cylinders, and the total output flow rate and output pressure of the power source—are used as input layer nodes. The same parameters are used for the next cycle's bucket and boom hydraulic cylinders. Therefore, the number of input layer nodes is 6, and the number of output layer nodes is 6. The number of hidden layer nodes is usually determined by trial and error. Based on empirical formulas, the number of input layer nodes is assumed to be... The number of output layer nodes is The number of hidden layer nodes is ,in Since the initial number of hidden layer nodes is a constant ranging from [1, 10], it is set to [4, 13]. The initial number of hidden layer nodes is 4, which serves as a minimum reserve for subsequent trial-and-error adjustments while ensuring basic fitting capability. Based on the cluster centers obtained from the P-norm nearest neighbor algorithm, the width of the hidden layer Gaussian kernel is initialized to 0.5. The connection weights from the input layer to the output layer are initialized to random values ​​within the range of (-0.1, 0.1) to prevent all neurons from learning the same features. The network training iterations are set to 1000 times, and the iteration update rate is initialized to 0.1.

[0034] Step S24: Use the loader load signal data training set obtained in step S1 Train the RBF neural network model and adjust the output layer weights; use the load power tracking error objective function. The input layer connection weights are adjusted and corrected until the required number of training iterations is reached; the load power tracking error objective function is as follows: ; in, Let the objective function be the load power tracking error. For the first The actual load power value for each load operation mode; For the first The actual load power prediction value for each load operation mode; Connect the weight vector to the input layer. ; This represents the number of hidden layer nodes. This represents the number of training samples.

[0035] The weights are iteratively adjusted using gradient descent to minimize the load power tracking error. The new input layer is then connected to the weight vector. The value assigned is: ; ; in, For the first Input layer connection weight vector for clustering of load job patterns; For the first Training samples for each load job mode; The iteration update rate is initialized and set in step S23; For the first The activation function output for each load job pattern cluster; An exponential function with the natural constant e as its base.

[0036] During training, the average gradient is calculated using all training samples in each iteration, and the process is repeated until the preset number of training iterations is reached.

[0037] Step S3: Use the RBF neural network model obtained in step S2 to predict the load power, perform multi-objective optimization matching, and obtain the optimal control parameters of the loader servo pump control system.

[0038] Step S31: Generate control parameters for the loader servo pump control system based on the predicted load power; test the loader load signal data set from step S1. The input is fed into the RBF neural network model that has been trained in step S2 to obtain the corresponding load power prediction sequence. , , The predicted load power value Actual measured load power corresponding to the test set A comparison was conducted, and the following metrics were used for quantitative evaluation to verify the prediction accuracy and reliability of the RBF neural network model constructed and trained in the above steps: ; ; ; in, To measure the average absolute deviation between predicted and actual values; To be more sensitive to large errors in the forecast and to reflect the stability of the forecast; To characterize the extent to which the predicted value explains the changes in the true value, the closer it is to 1, the better the fit. This is a test sample of loader load signal data; For the first The true value of a sample of load job modes; For the first Predicted values ​​for a sample of load operation patterns; For the first The average of the true values ​​of a sample of load operation modes; This is the overall mean of the true values ​​of all samples in the test set.

[0039] When the root mean square error (RMSE) of the prediction model is less than 5 kJ / m² and the coefficient of determination (R²) is less than 5 kJ / m², the prediction model is more than 5 kJ / m². 2 If the accuracy is higher than 0.9, the RBF neural network model is considered to meet the requirements of subsequent multi-objective optimization and can be used for load power prediction. If the accuracy of the RBF neural network model meets the requirements after verification by the above indicators, the RBF neural network model and its output predicted load power sequence will be used for multi-objective optimization matching in the subsequent step S32.

[0040] Step S32: Obtain the predicted load power and perform multi-objective optimization matching to obtain the optimal control parameters of the loader servo pump control system.

[0041] Step S321: Initialize the loader servo pump control system control parameters. The random generation scale is as follows: Initial loader servo pump control system control parameters Set the generation number Gen=1; for each variable This represents a control scheme based on the servo motor speed. The variable pump displacement V is controlled by parameters; the objective function value of each variable in the loader servo pump control system is obtained to minimize the load power tracking error, specifically: ; in, For load power tracking error; The load power predicted for the next cycle by the RBF neural network model; In the control scheme The estimated output power of the power source; This is a function that takes the minimum value.

[0042] Estimated output power of the power source The pump outlet pressure Motor speed and pump displacement The calculation yields the following expression: ; in, The mechanical efficiency of the motor; This refers to the volumetric efficiency of the variable pump.

[0043] Obtain the objective function value for each variable in the control parameters to maximize the overall efficiency of the loader servo pump control system, specifically: ; in, Let be the overall efficiency function of the loader servo pump control system, representing the maximum efficiency value; The electrical power consumed to drive the servo motor; This is a function to find the maximum value. To improve the efficiency of the loader's servo pump control system.

[0044] Step S322: Perform the selection, cross SBX, and update PM operations to generate secondary control parameters; for the selected primary control parameters... , The second-level control variables are generated according to the following probability distribution: ; in, The first control variable is the primary control parameter. The second control variable is the primary control parameter; This is the first control variable for the secondary control parameters after crossover. =1 indicates rotational speed. =2 indicates displacement; The second control variable is the secondary control parameter after crossover; These are random numbers generated according to a probability distribution. Use symbols to mark control variables.

[0045] For control variables A certain control variable That is, one of the rotational speed or displacement is updated according to the following formula: ; in, The values ​​of the control variables after the mutation; The values ​​of the control variables before the mutation; This is the upper bound of the control variable; To control the lower bound of the variable; These are the probability distribution parameters.

[0046] Step S323: If the iteration count Gen=1, directly form new control parameters; otherwise, merge the first- and second-level control parameters; perform non-dominated sorting on the merged control parameters, stratifying them into frontier levels such as F1, F2, ...; obtain the characteristic distance of variables within each non-dominated level and evaluate the solution set distribution density; to maintain the diversity of the solution set within the same non-dominated level, calculate the distribution density of each control variable. Feature distance for: ; in, To control variables The first adjacent variable in the first position The values ​​of each objective function; To control variables The second adjacent variable in the th The values ​​of each objective function; For the current entire non-dominated layer, the first The maximum value of each objective function; For the current entire non-dominated layer, the first The minimum value of an objective function; Number the objective function; For each control variable The characteristic distance.

[0047] Select the new level of control parameters based on the non-dominance level and congestion level. The evolutionary iteration count is updated by `Gen = Gen + 1`; the crowding level prioritizes variables with large feature distances within the same layer. If the maximum number of generations is not reached, the iteration returns to start over; otherwise, the optimal control parameter set of the loader servo pump control system is output. .

[0048] Step S4: Obtain the optimal control parameter set obtained in step S3. Execute optimal matching control of the loader servo pump control system; from the optimal control parameter set The final control parameters are selected based on the real-time strategy as follows: ; in, For the control variables of the loader servo pump control system; For the speed control variable of the loader's servo pump control system; For the displacement control variable of the loader servo pump control system; This is the optimal set of control parameters for the loader's servo pump control system.

[0049] The selected control variables are sent to the servo drive and pump variable mechanism for execution, completing the dynamic matching of power source and load power. The RBF neural network integer-cycle load power prediction method for the servo pump control system of this invention is used to realize the servo pump control system schematic diagram as shown below. Figure 2 As shown, the core components and energy flow path of the servo pump control system of the present invention are illustrated, revealing the hardware foundation for the present invention to achieve efficient matching and intelligent control.

[0050] The flowchart of RBF neural network load power prediction is as follows: Figure 3 As shown, the calculation process from raw data to final prediction results is illustrated, clarifying the determination method and optimization algorithm of key parameters of RBF neural network, and improving the robustness and accuracy of prediction model. The key parameters of RBF neural network are the center, width, and output weights of the basis function.

[0051] The RBF neural network training flowchart is as follows: Figure 4 As shown in the figure, this diagram illustrates the complete computational process from sample data preprocessing to the final output prediction result. The process begins with preprocessing the original sample data to repair and reduce noise, followed by the initialization stage of the neural network parameters. The core step is to dynamically adjust and obtain the optimal parameters of the hidden and output layers of the neural network, the basis function width, and the connection weights through iterative training, using the training error as an evaluation metric to determine whether the termination condition is met. Once the condition is met, the process enters the final stage, using test samples to predict load power and quantitatively evaluating the prediction performance. This visualization clearly and systematically elucidates the complete logical architecture and execution sequence of the intelligent control method proposed in this invention, clarifying the data flow and collaborative relationships between various algorithm modules, such as data preprocessing, the RBF neural network, and the evaluation module.

[0052] The load power prediction model structure is as follows: Figure 5 As shown in the figure, this diagram clearly illustrates the key physical quantities in the electromechanical control servo pump control system: the rodless chamber pressure of the bucket and boom hydraulic cylinders in the previous cycle. , Oil inlet flow rate of bucket and boom hydraulic cylinders , Total output flow and output pressure of the power source , Next cycle bucket and boom hydraulic cylinder rodless chamber pressure , Oil inlet flow rate of bucket and boom hydraulic cylinders , Total output flow and output pressure of the power source , ; For input layer nodes; Hidden layer nodes; Output layer node. These are mapped to the corresponding relationships of the six input nodes of the RBF neural network.

[0053] The flowchart of the NSGA-II multi-objective optimization algorithm is as follows: Figure 6 As shown in the figure, this diagram illustrates the complete process of the algorithm, starting from the initialization of control parameters and iteratively evolving to find the optimal solution set. It clearly reveals the inherent optimization mechanism of the NSGA-II algorithm in addressing the conflicting objectives of power point tracking and system energy efficiency in this invention.

[0054] The comparison chart of boom cylinder load pressure prediction is as follows: Figure 7As shown in the figure, the load pressure variation of the boom cylinder during a typical work cycle is illustrated. Throughout the work cycle, the boom cylinder load pressure exhibits typical periodic and drastic fluctuations, with multiple steep upward and downward peaks. The data in the figure demonstrates that the predicted curve of the model in this invention closely matches the measured curve.

[0055] The comparison chart of bucket cylinder load pressure prediction is as follows: Figure 8 As shown in the figure, this diagram compares the predicted load pressure of the bucket hydraulic cylinder by the present invention with the measured data. Compared to the boom cylinder, the pressure dynamic characteristics of the bucket cylinder are typically more complex and frequent. During the time interval of 12s to 16s, the load pressure exhibited multiple high-frequency, small-amplitude, violent pulsations. Nevertheless, the predicted output curve of the present invention still shows a high degree of agreement with these pulsations.

[0056] The comparison chart of boom cylinder load flow prediction is as follows: Figure 9 As shown in the figure, the load flow of the boom cylinder experienced multiple large fluctuations during the 0-20 second working cycle, with a wide range of dynamic fluctuations. The experimental data curve and the predicted data curve highly overlapped throughout the entire time period.

[0057] The comparison chart of bucket cylinder load flow prediction is as follows: Figure 10 As shown, compared to the boom cylinder, the bucket cylinder exhibits different dynamic characteristics in terms of load flow variation, with more frequent and less regular fluctuations. The specific data curves in the figure show that the bucket cylinder flow rate switches frequently within 0-20 seconds. Despite the dramatic changes, the predicted data curve (dashed line) and the experimental data curve (solid line) maintain a very high degree of agreement.

[0058] The power source output pressure prediction comparison chart is as follows: Figure 11 As shown in the figure, the specific data curves reveal that the power source's output pressure dynamically adjusts according to load demand throughout the entire operating cycle. The experimental data curves exhibit several distinct pressure jumps and plateau periods. The predicted data curves closely match the experimental curves.

[0059] The power source output flow prediction comparison chart is shown below. Figure 12 As shown in the figure, the specific data curves demonstrate that the total output flow of the power source is the sum of the flow demands of each actuator, and its changes reflect the loader's operational intensity. The experimental data curves appear as a series of complex flow pulses with varying amplitudes. The predicted data curves successfully tracked this complex variation. The accuracy of the predicted data for each part of the working system compared to the experimental data is over 87%.

[0060] The second aspect of this invention proposes a load power prediction method for a servo pump control system of a loader. Figure 2This is a schematic diagram of the servo pump control system for a loader according to the present invention, which includes: a power source module, a pump control hydraulic circuit module, an actuator module and a drive control module.

[0061] The power source module includes a battery 10 and a rectifier inverter 9; the battery 10 is connected to the input terminal of the rectifier inverter 9 via a DC bus to provide and convert drive power; the output terminal of the rectifier inverter 9 is connected to the input terminal of the permanent magnet synchronous motor 8.

[0062] The pump-controlled hydraulic circuit module includes a permanent magnet synchronous motor 8, an electro-hydraulic proportional variable pump 7, a relief valve 6, a first check valve 41, a second check valve 42, a back pressure check valve 5, and an oil tank. The output end of the permanent magnet synchronous motor 8 is connected to the mechanical input end of the electro-hydraulic proportional variable pump 7 to drive the pump. The electro-hydraulic proportional variable pump 7 has a first oil outlet and a second oil outlet. The first oil outlet is connected to the main pressure oil circuit, and the second oil outlet is connected to the auxiliary circuit and the replenishment oil circuit. The oil inlet of the relief valve 6 is connected to the main pressure... The oil circuit includes an overflow valve 6 whose outlet is connected to the oil tank for system protection; the inlet of the first check valve 41 is connected to the second outlet of the electro-hydraulic proportional variable pump 7, and the outlet of the first check valve 41 is connected to the main pressure oil circuit, forming a replenishment oil circuit; the inlet of the second check valve 42 is connected to the oil tank, and the outlet of the second check valve 42 is connected to the auxiliary circuit; the inlet of the back pressure check valve 5 is connected to the auxiliary circuit, and the outlet of the back pressure check valve 5 is connected to the oil tank to maintain the back pressure of the circuit.

[0063] The actuator module includes a multi-way valve group 3, a first hydraulic cylinder 1, and a second hydraulic cylinder 2; the first oil inlet of the multi-way valve group 3 is connected to the main pressure oil circuit, and the second oil inlet is connected to the auxiliary circuit; the multiple working oil ports of the multi-way valve group 3 are respectively connected to the rodless chamber and the rod chamber of the first hydraulic cylinder 1 and the second hydraulic cylinder 2 to control the movement of the hydraulic cylinders.

[0064] The drive control module includes a servo driver 14, a controller 15, a pressure sensor 11, and a flow sensor 12. The output of the servo driver 14 is connected to the control input of the permanent magnet synchronous motor 8 to adjust the motor speed. The first control output of the controller 15 is connected to the input of the servo driver 14 to send speed control commands to the servo driver 14. The second control output of the controller 15 is connected to the pump variable mechanism 13 of the electro-hydraulic proportional variable pump 7 (the pump variable mechanism 13 is an internal component of the electro-hydraulic proportional variable pump 7) to adjust the pump displacement. The other control output of the controller 15 is connected to the pump variable mechanism 13 of the electro-hydraulic proportional variable pump 7 to adjust the pump displacement. The input of the controller 15 is connected to the pressure sensor 11 and the flow sensor 12 to obtain the system status. The output of the controller 15 is connected to the input of the servo driver 14. The pressure sensor 11 is installed in the main pressure oil circuit to detect the system pressure. The flow sensor 12 is installed in the output oil circuit of the electro-hydraulic proportional variable pump 7 to detect the pump output flow.

[0065] The rectifier-inverter 9 of the power source module drives the permanent magnet synchronous motor 8 of the pump-controlled hydraulic circuit module; the motor drives the electro-hydraulic proportional variable pump 7 to generate hydraulic oil flow, which is regulated, replenished and unloaded through the valve group composed of the overflow valve 6, the first check valve 41, the second check valve 42 and the back pressure check valve 5 of the pump-controlled hydraulic circuit module, and then supplies the multi-way valve group 3 of the actuator module to drive the first hydraulic cylinder 1 and the second hydraulic cylinder 2 to work; the pressure sensor 11 and the flow sensor 12 of the drive control module monitor the hydraulic circuit status in real time and feed it back to the controller 15. The controller 15 calculates the load power prediction method of the loader servo pump control system proposed in the first aspect of the present invention, and performs coordinated control of the permanent magnet synchronous motor 8 of the power source module and the electro-hydraulic proportional variable pump 7 of the pump-controlled hydraulic circuit module through the servo driver 14 and the pump variable mechanism 13, thereby realizing the dynamic matching and efficient operation of the system power.

[0066] The beneficial effects of this invention are as follows: This invention proposes a load power prediction method and a servo pump control system for a loader. An RBF neural network is used to establish a load power prediction model for the loader's working system, ultimately achieving effective prediction of the periodic load power of the loader's working system. In this embodiment, multi-objective optimization algorithms such as NSGA-II are used to dynamically and collaboratively control the servo motor speed and variable pump displacement, guiding the controller to achieve precise control of the power source output power and multi-way valve flow distribution. This enables the dual-variable power source output power to track load demand in real time and accurately, fundamentally reducing high-pressure overflow and throttling losses, and achieving high-efficiency operation of the loader's working system throughout the entire operating cycle. This invention also establishes a servo pump control system for the loader, realizing the application of the load power prediction method for the loader's servo pump control system and solving engineering technical problems.

[0067] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting the load power of a loader servo pump control system, characterized in that, It includes: S1: Acquire real-time operating data of the loader, perform wavelet decomposition and cluster analysis, and divide it into a training set of loader load signal data. Loader load signal data test set ; S2: Based on the clustering analysis results of the loader load signal data, set the parameters of the hidden layer neurons in the RBF neural network, and obtain the training set of loader load signal data obtained in step S1. Train the RBF neural network model and adjust the output layer weights; use the load power tracking error objective function. Adjust and correct the input layer connection weights until the required number of training iterations is reached; S3: Use the RBF neural network model obtained in step S2 to predict the load power, and generate control parameters for the loader servo pump control system based on the predicted load power; perform multi-objective optimization matching to obtain the optimal control parameter set for the loader servo pump control system. ; S4: Based on the optimal control parameter set determined in step S3 The speed and displacement of the loader's servo pump control system are controlled to achieve optimal matching control of the loader's servo pump control system.

2. The load power prediction method for the loader servo pump control system according to claim 1, characterized in that: Step S2 is as follows: S21: Obtain the node width parameter of the hidden layer neurons through the nearest neighbor algorithm to adapt to the distribution density of the loader load in the feature space; S22: Choose the diminishing memory recursive least squares method to solve the weight parameters from the hidden layer to the output layer of the neural network, to adapt to the time-varying working conditions of the loader; S23: Obtain the loader's operating conditions and sampled data features, and initialize the RBF neural network parameters; S24: Train the RBF neural network model using the load power tracking error objective function. The input layer connection weights are adjusted and corrected using gradient descent, and the weights are updated iteratively.

3. The load power prediction method for the loader servo pump control system according to claim 2, characterized in that: The objective function for load power tracking error in step S24 is as follows: ； in, Let the objective function be the load power tracking error. For the first The actual load power value for each load operation mode; For the first The actual load power prediction value for each load operation mode; Connect the weight vectors to the input layer; This represents the number of training samples; This is the sequence number of the load sample.

4. The load power prediction method for the loader servo pump control system according to claim 2, characterized in that: Step S22 is as follows: ； ； ； in, For the first Update the step size gain vector for each control parameter; For the first Covariance matrix of each control parameter; This is the output control vector of the current hidden layer; Transpose the output control vector of the current hidden layer; It is a gradually diminishing factor; For the first Estimated values ​​of the connection weight vectors of each input layer; This is an index for the number of layers in a neural network. For the first Covariance matrix of each control parameter; For the first Estimates of the input layer connection weight vectors.

5. The load power prediction method for the loader servo pump control system according to claim 3, characterized in that: Step S24 uses gradient descent to iteratively adjust the weights, updates the weights, and connects the new input layer to the weight vector. The value assigned is: ； ； in, For the first Input layer connection weight vector for clustering of load job patterns; For the first Training samples for each load job mode; For iterative update rate; For the first The activation function output for each load job pattern cluster; It is an exponential function with the natural constant e as its base; For belonging to the first The number of load samples in each load job pattern cluster; For the first Hidden layer node width feature parameters for clustering of individual load job patterns; Index for cluster groups based on job load patterns; This is the preset total number of cluster centers for load features.

6. The load power prediction method for the loader servo pump control system according to claim 1, characterized in that: Step S3 is as follows: S31: Generate control parameters for the loader's servo pump control system based on the predicted load power; obtain the predicted load power value sequence using the RBF neural network model. Used for multi-objective optimization matching; S32: Obtain the predicted load power for multi-objective optimization matching, and adjust the control variables of the first-level control parameters. Perform selection, crossover, and update operations to generate secondary control parameters; merge primary and secondary control parameters, and calculate each control variable. The characteristic distance is used to output the optimal control parameter set of the loader servo pump control system. .

7. The load power prediction method for the loader servo pump control system according to claim 6, characterized in that: The method for obtaining the secondary control variables in step S32 is as follows: ； in, The first control variable is the primary control parameter. The second control variable is the primary control parameter; This is the first control variable for the secondary control parameters after crossover; The second control variable is the secondary control parameter after crossover; These are random numbers generated according to a probability distribution. For time parameters; Use symbols to mark control variables.

8. The load power prediction method for the loader servo pump control system according to claim 7, characterized in that: Each control variable in step S32 Feature distance for: ； in, To control variables The first adjacent variable in the first position The values ​​of each objective function; To control variables The second adjacent variable in the th The values ​​of each objective function; For the first The maximum value of each objective function; For the first The minimum value of an objective function; Number the objective function; For each control variable The characteristic distance.

9. The load power prediction method for the loader servo pump control system according to claim 1, characterized in that: Step S1 is as follows: S11: Collect real-time operating data during the loader's operation, perform wavelet decomposition, and analyze the detailed parameters of each load signal. Perform soft thresholding to extract detailed parameters of the load signal. The loader load signal data is reconstructed from the original low-frequency parameters and then denoised. ; S12: Acquire the denoised loader load signal data We analyzed the characteristics of the load signal data of the loader and used clustering methods to extract the cluster centers of typical load operation modes. S13: Divide the processed loader load signal data into a loader load signal data training set according to time sequence. Loader load signal data test set .

10. A servo pump control system for a load power prediction method in a loader servo pump control system according to any one of claims 1 to 9, characterized in that, It includes: Power source module, pump-controlled hydraulic circuit module, actuator module, and drive control module; The power source module includes a battery and a rectifier-inverter; the battery is connected to the input of the rectifier-inverter via a DC bus to provide and convert drive power; the output of the rectifier-inverter is connected to the input of the permanent magnet synchronous motor. The pump-controlled hydraulic circuit module includes a permanent magnet synchronous motor, an electro-hydraulic proportional variable pump, a relief valve, a first check valve, a second check valve, a back pressure check valve, and an oil tank. The actuator module includes a multi-way valve group, a first hydraulic cylinder, and a second hydraulic cylinder; the first oil inlet of the multi-way valve group is connected to the main pressure oil circuit, and the second oil inlet is connected to the auxiliary circuit; multiple working oil ports of the multi-way valve group are respectively connected to the rodless chamber and the rod chamber of the first hydraulic cylinder and the second hydraulic cylinder, for controlling the movement of the hydraulic cylinder; The drive control module includes a servo driver, a controller, a pressure sensor, and a flow sensor; the output of the servo driver is connected to the control input of the permanent magnet synchronous motor to adjust the motor speed.

Images