Control method and system for steel support axial force self-servo control joint based on hybrid neural network

Abstract

The invention discloses a control method and system for a steel support axial force self-servo control node based on a hybrid neural network, and the method comprises the steps: automatically adjusting an error value obtained through a steel support axial force self-servo control node algorithm based on a CNN-BiLSTM-Attention hybrid neural network, and carrying out the dynamic adjustment of the weights of different time steps, and converging the error between the output value of the control node and the preset value to a project allowable range. According to the scheme, based on CNN-BiLSTM-Attention hybrid neural network autonomous learning, optimization of the steel support axial force self-servo control algorithm is achieved, the long-term dependency relationship of time sequence data can be better captured, the optimization effect of the control parameter combination is improved, and therefore the control precision of the steel support axial force and the system stability are improved.

Description

Technical Field

[0001] The invention relates to the technical field of building engineering, and in particular to a control technology of a steel support axial force self-servo control section. Background Art

[0002] Steel supports have been widely used in current foundation pit projects. However, due to the large amount of control information for steel supports, the system is relatively complex. Therefore, it is easy for them to fall off due to inadequate execution or environmental changes, thus causing major safety accidents.

[0003] Faced with foundation pit projects with high control requirements, the support effect of steel supports is difficult to meet. The emergence of steel support axial force self-servo control nodes provides a more efficient and safe solution for foundation pit deformation control. However, most of the current control of foundation pit deformation is improved by changing the position, number and angle of the steel support axial force self-servo control nodes. When encountering complex foundation pits, the steel support axial force self-servo control nodes have problems such as simple signal processing, overshoot, and oscillation. These measures cannot meet the control requirements of foundation pit deformation. It is necessary to improve the control accuracy of the self-servo control node itself.

[0004] At present, when the steel support axial force is controlled by the servo control section, the main problems are as follows:

[0005] (1) When encountering complex foundation pits, the steel support axial force self-servo control node has problems such as simple signal processing, overshoot, and oscillation, which cannot meet the control requirements of foundation pit deformation.

[0006] (2) The parameters of the steel support axial force self-servo control algorithm are generally set according to the experience of engineers. The control algorithm has limitations, low accuracy, weak robustness, and simple signal processing, which leads to poor control accuracy.

[0007] Based on this, people proposed a solution to improve the control accuracy of the steel support axial force self-servo control section based on BP neural network. However, the control method based on BP neural network has problems such as limited accuracy and difficulty in capturing long-term dependencies when processing multi-dimensional and time-series-correlated steel support axial force data.

[0008] It can be seen from the above that how to effectively improve the control accuracy of the steel support axial force self-servo control section is a problem that needs to be solved urgently in this field. Summary of the invention

[0009] Aiming at the problem of poor control accuracy in the control scheme of the existing steel support axial force self-servo control joint, the purpose of the present invention is to provide a control method for the steel support axial force self-servo control joint based on deep learning. This control method optimizes the control of the steel support axial force self-servo control joint based on the CNN-BiLSTM-Attention deep learning model, so that the relevant control parameters in the control process reach the best combination, thereby improving the control accuracy and meeting the control of foundation pit deformation. On this basis, the present invention further provides a system capable of implementing this control method.

[0010] To achieve the above object, the control method of the steel support axial force self-servo control joint provided by the present invention autonomously adjusts the error value obtained by the steel support axial force self-servo control joint algorithm based on the CNN-BiLSTM-Attention hybrid neural network, and makes the error between the output value of the control joint and the preset value converge to the engineering allowable range by dynamically adjusting the weights of different time steps.

[0011] In some embodiments of the present invention, the control method includes the following steps:

[0012] (1) Data collection and preprocessing: Collect data related to the axial force of the steel support and preprocess it;

[0013] (2) CNN feature extraction: Use CNN to extract features from the preprocessed data to obtain local feature information;

[0014] (3) BiLSTM time series modeling: Based on the features extracted by CNN, use the BiLSTM model to capture the long-term dependence relationship of the time series;

[0015] (4) Application of the Attention mechanism: Introduce the attention mechanism to weight the temporal features output by BiLSTM;

[0016] (5) Generation and execution of the control strategy: Generate a control strategy according to the model output and adjust the axial force of the steel support in real time.

[0017] In some embodiments of the present invention, the CNN in step (2) includes multiple convolutional layers and pooling layers for extracting multi-scale local features.

[0018] In some embodiments of the present invention, the BiLSTM model in step (3) consists of a forward and a backward LSTM network, which can process the forward and backward information of the time series simultaneously.

[0019] In some embodiments of the present invention, in step (4), the key features are automatically focused by calculating the importance weights of different time steps or features.

[0020] To achieve the above object, the control system of the steel support axial force self-servo control section provided by the present invention includes:

[0021] A data acquisition module for acquiring data related to the axial force of the steel support;

[0022] A data preprocessing module for preprocessing the data acquired by the data acquisition module;

[0023] The CNN-BiLSTM-Attention model is constructed based on the CNN-BiLSTM-Attention hybrid neural network architecture, and can interact with the data preprocessing module to obtain the data preprocessed by the data preprocessing module and complete model training; after the CNN-BiLSTM-Attention model is trained, it can extract features, model time series, and weight key features for the input data;

[0024] A control strategy generation module that interacts with the CNN-BiLSTM-Attention model and generates a control strategy for optimizing the control algorithm of the steel support axial force self-servo control section according to the model output.

[0025] In some embodiments of the present invention, the CNN-BiLSTM-Attention model is composed of five layers: an input layer, a convolutional neural network layer, a bidirectional long short-term memory network layer, an attention mechanism layer, and an output layer.

[0026] In some embodiments of the present invention, the convolutional neural network layer includes multiple convolutional layers and pooling layers for extracting multi-scale local features.

[0027] In some embodiments of the present invention, the bidirectional long short-term memory network layer is composed of a forward and a backward LSTM network, which can simultaneously process the forward and backward information of the time series.

[0028] In some embodiments of the present invention, the attention mechanism layer automatically focuses on key features by calculating the importance weights of different time steps or features.

[0029] The control method of the steel support axial force self-servo control section provided by the present invention is based on the autonomous learning of the CNN-BiLSTM-Attention hybrid neural network, thereby optimizing the algorithm of the steel support axial force self-servo control section, being able to better capture the long-term dependence relationship of time series data, improving the optimization effect of the control parameter combination, and thus enhancing the control accuracy of the steel support axial force and the system stability. Description of the Drawings

[0030] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0031] Figure 1 This is the principle flowchart of the control method for the steel support axial force self-servo control section in the present invention;

[0032] Figure 2 This is the control flowchart of the control method for the steel support axial force self-servo control section in the example of the present invention. Specific embodiments

[0033] In order to make the technical means, creative features, achieved purposes and effects realized by the present invention easy to understand, the present invention will be further described below in conjunction with specific drawings.

[0034] Through full research on the steel support axial force self-servo control section used in the current project construction, it is found that the core of the precision control of the steel support axial force self-servo control section is the control section algorithm.

[0035] Among them, the controlled object obtains an error value through the actual value and the preset value collected, and then the control section algorithm operates on the proportional parameter, integral parameter and differential parameter of the error value. When the three parameters reach the best parameter combination, the control effect can reach the best.

[0036] However, at present, the proportional parameter, integral parameter and differential parameter of the error value obtained by the steel support axial force self-servo control section algorithm are mostly determined by the experience of engineering personnel, and the best parameter combination cannot be matched, resulting in great difficulty in selecting the best parameter combination, low control precision, weak robustness and simple signal processing.

[0037] In response to this, the present invention provides a control method based on the CNN-BiLSTM-Attention deep learning model. This control method is based on the CNN-BiLSTM-Attention hybrid neural network to optimize the steel support axial force self-servo control section algorithm, and can make the proportional parameter, integral parameter and differential parameter of the algorithm error value reach the best combination.

[0038] This control method is deeply integrated with the CNN-BiLSTM-Attention hybrid neural network and the PID control algorithm to realize the intelligent optimization of the steel support axial force control. Through multi-dimensional feature extraction and time series dependence modeling, the error between the output value of the control section algorithm and the preset value converges to the engineering allowable range.

[0039] The error here is the gap between the model preset value and the actual output value, and this gap value corresponds to the fitting ability of the model to the training data.

[0040] In the control process of the steel support axial force self - servo control section, due to the complex actual situation on site, it is difficult to find the appropriate proportion relationship among the three control links of proportional, integral, and differential relying on manual experience; this control method introduces a CNN - BiLSTM - Attention hybrid neural network to study and learn the numerical values of each link of the steel support axial force self - servo control section, so as to determine a set of optimal control section algorithm parameters, thereby improving the stability and efficiency of the system.

[0041] The CNN - BiLSTM - Attention hybrid neural network adopted in this control method uses error backpropagation for learning. It consists of five layers: an input layer, a convolutional neural network layer (CNN layer), a bidirectional long short - term memory network layer (BiLSTM layer), an attention mechanism layer (Attention layer), and an output layer. Through multi - dimensional feature extraction and time - series dependence modeling, this CNN - BiLSTM - Attention hybrid neural network can better capture the long - term dependence relationship of time - series data, improve the optimization effect of control parameter combinations, and thus enhance the control accuracy of steel support axial force and system stability.

[0042] In the CNN - BiLSTM - Attention hybrid neural network architecture of the present invention, its input layer is used to define the format of input data, including batch size, number of time steps, and feature dimension. In this solution, according to the characteristics of axial force control data, the number of time steps and feature dimension are further set.

[0043] Furthermore, the format of the input data defined by the input layer is as follows:

[0044] Batch size: Determined according to the size of training data and memory limitations.

[0045] Number of time steps (t): The length of the time series, that is, how many consecutive time - step data are considered to predict the control parameters of the next time step.

[0046] Feature dimension (n): The number of features of the input data, specifically including preset values, actual output values, deviations, etc. in axial force control.

[0047] Since the control process of axial force is a continuous and dynamic process, historical data has an important impact on current control decisions. In the solution of the present invention, data such as preset values, actual output values, and the deviation between the two in the axial force control requirements are used as time - series data for input.

[0048] The convolutional neural network layer (CNN layer) in this CNN - BiLSTM - Attention hybrid neural network architecture is used to extract local spatial features from the input data of the input layer and generate high - order feature maps;

[0049] For data such as the preset value, actual output value, and the deviation between the two in the axial force control requirement as time series data input in the input layer, a one-dimensional convolutional kernel is used to perform a convolutional operation in the time series direction.

[0050] Furthermore, the size of the convolutional kernel in this convolutional neural network layer (CNN layer) is based on the data characteristics of the time series, preferably 3 or 5 here, thereby capturing local time features.

[0051] Furthermore, after the convolutional operation in this convolutional neural network layer (CNN layer), a pooling operation (such as max pooling) is also performed to reduce the dimension of the data and retain important feature information.

[0052] The bidirectional long short-term memory network layer (BiLSTM layer) in this CNN-BiLSTM-Attention hybrid neural network architecture can perform bidirectional temporal modeling on the feature map based on the features extracted by the CNN layer, construct a BiLSTM model, and use the BiLSTM model to capture the forward and reverse dynamic dependencies in the time series data to determine the influence of historical errors on the current control during the control process of the steel support axial force self-servo control section.

[0053] The BiLSTM model here consists of a forward and a backward LSTM network, which can process the forward and backward information of the time series simultaneously. The forward LSTM network starts from the starting point of the time series and processes the features of each time step in sequence; the backward LSTM network starts from the end point of the time series and processes the features of each time step in reverse order; finally, the outputs of the two directions are concatenated to obtain a hidden state that integrates the front and back information of the time series.

[0054] By training the BiLSTM model, the dynamic change law of the steel support axial force data in the time dimension can be learned, such as the long-term trend of axial force change, seasonal fluctuations, etc., effectively capturing the long-term dependencies of the time series, thereby determining the influence of historical errors on the current control during the control process of the steel support axial force self-servo control section.

[0055] The attention mechanism layer (Attention layer) in this CNN-BiLSTM-Attention hybrid neural network architecture weights the temporal features output by the BiLSTM layer by combining the attention mechanism to highlight key information.

[0056] This attention mechanism layer (Attention layer) can dynamically adjust the weights of different time steps by introducing the attention mechanism, enabling the model to focus on the data of the time steps that are most important for the current control decision.

[0057] Furthermore, the calculation process of this attention mechanism layer (Attention layer) is as follows:

[0058] (1) Calculate the attention weights:

[0059] First, take the output hidden state of the BiLSTM layer as the input of the attention mechanism; then, calculate the attention scores for each time step through a fully connected layer.

[0060] Here, it is preferably to use a feedforward neural network to learn the attention scores:

[0061] scoret = Wa·tanh(Wb·ht + ba),

[0062] where ht is the hidden state of the BiLSTM layer at time step t, Wa and Wb are learnable weight matrices, and ba is the bias term.

[0063] (2) Normalize the attention scores:

[0064] Use the softmax function to normalize the attention scores into a weighted probability distribution.

[0065] (3) Weighted summation:

[0066] Perform a weighted summation of the hidden state of the BiLSTM layer and the attention weights to obtain the attention context vector, which contains the most crucial information for control decisions.

[0067] In the output layer of this CNN - BiLSTM - Attention hybrid neural network architecture, according to the requirements of the control section algorithm, set the number of neurons and the activation function of the output layer.

[0068] For the requirements of the steel support axial force self - servo control section algorithm, this output layer needs to output the proportional parameter (K p ), integral parameter (K i ) and derivative parameter (Kd). The number of neurons in this output layer is 3. At the same time, use a linear activation function (such as a linear function) as the activation function of the output layer to ensure that the output parameter values are continuous.

[0069] The solution of the present invention further gives a corresponding training scheme for the CNN - BiLSTM - Attention hybrid neural network, enabling the CNN - BiLSTM - Attention hybrid neural network to complete research and deep learning on the numerical values of each link of the steel support axial force self - servo control section.

[0070] The training scheme given here mainly includes the data input and pre - processing stage and the model training and optimization stage.

[0071] In the data input and preprocessing stage, first collect relevant data of the self-servo control of the steel support axial force (such as preset values, actual output values, and deviations between the two in the axial force control requirements, etc.). Here, the specific data collection scheme is not limited here and can be determined according to actual needs.

[0072] Next, preprocess the collected data. In this solution, the collected data is normalized or standardized, and the data is scaled to a certain range (such as [0,1] or standard normal distribution) to improve the stability and convergence speed of model training.

[0073] Specifically, here the data is scaled to the range of [0,1], and the specific processing formula is as follows:

[0074]

[0075] Among them, x is the input data, and x′ is the data after normalization.

[0076] Furthermore, in the data input and preprocessing stage, this solution also performs data augmentation, that is, enhances the data through translation, scaling, etc., increases the diversity of training data, and improves the generalization ability of the model.

[0077] Furthermore, in the data input and preprocessing stage, this solution also performs feature engineering, that is, extracts features that have an important impact on the control section algorithm, such as the deviation change trend at different time steps, etc.

[0078] In the model training and optimization stage, first divide the data preprocessed in the data input and preprocessing stage into a training set, a validation set, and a test set; then use the training set to train the model and update the model's parameters through the backpropagation algorithm.

[0079] During the training process, regularly use the validation set to evaluate the performance of the model, and adjust the model's parameters and structure according to the evaluation results.

[0080] On this basis, in order to further improve the model, this solution can adopt the method of combining pre-training and fine-tuning to train the CNN-BiLSTM-Attention hybrid neural network in the model training and optimization stage.

[0081] Specifically, first pre-train the CNN-BiLSTM-Attention model using a large-scale public dataset to learn more general feature representations; then, use a small amount of specific engineering data of steel support axial force control to fine-tune the model to adapt to specific steel support axial force control tasks. This method can significantly reduce the amount of data and time required for training, while improving the performance of the model.

[0082] To further improve the performance of the model, in the model training and optimization stage, this solution can also adopt the method of integrating reinforcement learning and supervised learning to train the CNN-BiLSTM-Attention hybrid neural network.

[0083] Specifically, the training of the model is divided into two stages: the supervised learning stage and the reinforcement learning stage;

[0084] In the supervised learning stage, the model can adjust the control parameters according to the deviation between the preset value and the actual output value; in the reinforcement learning stage, the model can autonomously explore different control strategies, evaluate the effectiveness of the strategies through the reward function, and adjust the model according to the reward signal. In this way, the autonomous learning ability and adaptability of the model can be improved, enabling it to better complete control tasks in complex environments.

[0085] See Figure 1 , the control method of the steel support axial force self-servo control section given by the present invention mainly includes the following steps:

[0086] (1) Determine the basic structure of the CNN-BiLSTM-Attention hybrid neural network, complete the training of the CNN-BiLSTM-Attention hybrid neural network, and initialize the parameters of each layer of the network for the trained CNN-BiLSTM-Attention hybrid neural network.

[0087] (2) According to the axial force control requirement, set the axial force preset value, compare it with the actual output value of the steel support axial force, and calculate the error value.

[0088] (3) Optimize the proportional parameter, integral parameter, and differential parameter of the error value calculated in step (2) by the CNN-BiLSTM-Attention hybrid neural network, and perform operations according to the optimized parameters by the steel support axial force self-servo control section algorithm to output the corresponding error value.

[0089] (4) Determine whether the error value of the output result in step (3) is within the preset specified range: if it is satisfied, perform step 6); if it is not satisfied, perform step 5).

[0090] (5) Calculate the performance index with the error value calculated in step (3), perform autonomous learning in the CNN-BiLSTM-Attention hybrid neural network, and return to step 2) for loop calculation.

[0091] (6) The final output result of the control section algorithm meets the axial force preset requirement, and thus obtain the optimal combination of the proportional parameter, integral parameter, and differential parameter of the error value.

[0092] (7) The calculation ends.

[0093] As can be seen from the above, the solution of the present invention is specifically based on the CNN-BiLSTM-Attention hybrid neural network, which can better capture the long-term dependence relationship of time series data through multi-dimensional feature extraction and time series dependence modeling, improve the optimization effect of the control parameter combination, and thus improve the control accuracy of the steel support axial force and the system stability.

[0094] For the control solution of the steel support axial force self-servo control section provided by the present invention, the following will be further described through specific application examples.

[0095] When the control solution of the steel support axial force self-servo control section provided by the present invention is specifically applied, it can form a corresponding software program to form a control system of the steel support axial force self-servo control section. When the software program runs, it will execute the above-mentioned control method of the steel support axial force self-servo control section, and at the same time be stored in a corresponding storage medium for the processor to retrieve and execute.

[0096] The control system of the present steel support axial force self-servo control section is mainly composed of a data acquisition module, a data preprocessing module, a CNN-BiLSTM-Attention model, and a control strategy generation module that cooperate with each other.

[0097] Among them, the data acquisition module is configured to collect data related to the steel support axial force.

[0098] The data preprocessing module interacts with the data acquisition module and is used to preprocess the collected data.

[0099] Further, the data preprocessing module is specifically configured to preprocess the collected data by using the aforementioned data preprocessing scheme.

[0100] The CNN-BiLSTM-Attention model is constructed by using the aforementioned CNN-BiLSTM-Attention hybrid neural network architecture, and can interact with the data preprocessing module to obtain the data preprocessed by the data preprocessing module and complete model training. After the CNN-BiLSTM-Attention model completes training, it can extract features, model time series, and weight key features for the input data.

[0101] The control strategy generation module can interact with the CNN-BiLSTM-Attention model and generate a control strategy for optimizing the control algorithm of the steel support axial force self-servo control section according to the model output.

[0102] The control strategy generation module generates a corresponding control strategy according to the output of the CNN-BiLSTM-Attention model in combination with the preset control target and safety threshold.

[0103] When the error value generated between the actual output value and the preset value of the steel support axial force predicted by the CNN-BiLSTM-Attention model exceeds the safety threshold, the control strategy generation module will issue an instruction to adjust the axial force. The control instruction is sent to the steel support axial force self-servo control section, and the axial force is adjusted in real time by adjusting the actuators such as jacks to ensure that the error between the output value of the control section algorithm and the preset value converges to the engineering allowable range. At the same time, the feedback module can also be used to monitor and feedback the control effect in real time, and further optimize the control parameters according to the actual change situation to improve the stability and reliability of the control system.

[0104] On this basis, the following uses corresponding application examples to further illustrate the application process of this solution.

[0105] In this example, first, a control software system of the steel support axial force self-servo control section based on the CNN-BiLSTM-Attention hybrid neural network is constructed according to the solution of the present invention. On this basis, the control software system is fused with the PID control algorithm to form an optimized algorithm for the control section of the steel support axial force self-servo control section, realizing intelligent optimization and precise control of the steel support axial force control.

[0106] The process of constructing the control software system of the steel support axial force self-servo control section with the CNN-BiLSTM-Attention hybrid neural network and forming the optimized algorithm for the control section based on this is as follows:

[0107] First, data preprocessing.

[0108] In this example, it is set that the collected steel support axial force measurement data is X = {x1, x2,..., x N}, where x i represents the axial force measurement value at the i-th time step. In order to make the data suitable for model input, the data is normalized here, and the data is scaled to the range of [0, 1]. The normalization formula is as follows:

[0109]

[0110] where, x i is the original data, x i ′ is the normalized data, and min(X) and max(X) respectively represent the minimum value and the maximum value in the data set X.

[0111] Next, CNN feature extraction.

[0112] In this example, a CNN model with two convolutional layers and two pooling layers is constructed. The first convolutional layer uses a convolutional kernel of size 3, and the second convolutional layer uses a convolutional kernel of size 5. After each convolutional layer, a ReLU activation function and a max pooling layer are connected. The specific formula is as follows:

[0113] f1(x) = ReLU(W1 * x + b1),

[0114] f2(x) = ReLU(W2 * f1(x) + b2),

[0115] where W1 and W2 are the convolutional kernel weights, b1 and b2 are the bias terms, and * represents the convolution operation.

[0116] On this basis, the normalized data X′ = {x1′, x2′, …, xN′} is used as the input of the CNN.

[0117] Next, BiLSTM time series modeling is performed.

[0118] Based on the features extracted by the CNN, a two-layer BiLSTM model is constructed. The input features here are X′ = [x1′, x2′, …, x N ′], where N is the number of time steps.

[0119] The forward LSTM and backward LSTM of the BiLSTM are calculated as follows:

[0120] Forward LSTM:

[0121]

[0122] Backward LSTM:

[0123]

[0124] where f t , i t , o t are the activation values of the forget gate, input gate, and output gate respectively, is the candidate cell state, C t is the cell state, and h t is the hidden state.

[0125] The hidden states of the forward and backward LSTMs are concatenated together to obtain the hidden state H = [h1, h2, …, h N that fuses the information before and after the time series.

[0126] Next, the Attention mechanism focuses on the key features.

[0127] The attention mechanism weights the hidden states output by the BiLSTM, that is, calculates the importance weight α for each time step t :

[0128]

[0129] where H t is the hidden state of the BiLSTM, W a and b a are the weights and biases of the attention mechanism, and V a is the output weight of the attention mechanism.

[0130] Multiply the attention weights by the hidden states of the BiLSTM to obtain the weighted features:

[0131]

[0132] Finally, the control strategy is generated and executed.

[0133] In this stage, according to the output of the CNN-BiLSTM-Attention model, combined with the preset control objectives and safety thresholds, corresponding control strategies are generated. When the error value generated between the actual output value and the preset value of the steel support axial force predicted by the CNN-BiLSTM-Attention model exceeds the safety threshold, the control strategy generation module will issue an instruction to adjust the axial force; send the control instruction to the steel support axial force self-servo control section, and adjust the axial force in real time by adjusting the actuators such as jacks to ensure that the error between the output value of the control section algorithm and the preset value converges to the engineering allowable range. At the same time, the feedback module can also be used to monitor and feedback the control effect in real time, and further optimize the control parameters according to the actual changes to improve the stability and reliability of the control system.

[0134] Specifically, according to the output H′ of the CNN-BiLSTM-Attention model, combined with the preset control objectives and safety thresholds, corresponding control strategies are generated. Assuming that the output of the model is y and the control objective is y target , then the control strategy u can be calculated by the following formula:

[0135]

[0136] where K p , K i and K d are the proportional, integral and differential coefficients.

[0137] It should be noted here that in this example, when training the model, the mean square error (MSE) is used as the loss function to optimize the parameters of the CNN-BiLSTM-Attention model

[0138] The loss function here is defined as:

[0139]

[0140] where N is the number of samples, y i is the true value, is the predicted value of the model.

[0141] On this basis, the Adam optimization algorithm is used to update the model parameters, with the learning rate set to 0.001 and the batch size set to 32.

[0142] See Figure 2 For this example, based on the above, the process of specifically controlling the steel support axial force self - servo control section based on the formed control section optimization algorithm is as follows:

[0143] 1) The control section optimization algorithm receives the control signal through the communication protocol, and the control signal triggers the internal program of the control section algorithm.

[0144] 2) After receiving the control instruction, the stepper driver controls the stepper motor and other controlled quantities through the main circuit and the control loop.

[0145] 3) The stepper motor controls the elongation of the jack by controlling the opening and closing of the digital valve.

[0146] 4) The pressure sensor is used as the measurement feedback device, and the error value generated between the actual output value and the preset value is returned to the control section algorithm again to form a closed - loop feedback control system.

[0147] Based on the above example, it can be seen that the solution of the present invention has the following technical effects compared with the prior art:

[0148] High - precision prediction: Through the CNN - BiLSTM - Attention model, the change of the steel support axial force can be predicted more accurately, improving the control accuracy.

[0149] Strong robustness: The model can adapt to the axial force changes under complex working conditions and has strong robustness.

[0150] Real - time control: Through real - time monitoring and feedback, the control strategy can be adjusted in a timely manner to ensure that the foundation pit deformation is always within the controllable range.

[0151] Multi - objective optimization: Through the multi - objective optimization control strategy, the control effect and resource consumption can be balanced, improving the overall performance of the control system.

[0152] Based on the control scheme of the above steel support axial force self-servo control section, an embodiment of the present invention further provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the steps of the control method of the above steel support axial force self-servo control section are implemented.

[0153] An embodiment of the present invention further provides a processor, and the processor is used to run a program, wherein when the program runs, the steps of the control method of the above steel support axial force self-servo control section are executed.

[0154] An embodiment of the present invention further provides a terminal device, the device includes a processor, a memory, and a program stored on the memory and executable on the processor, and the program code is loaded and executed by the processor to implement the steps of the control method of the above steel support axial force self-servo control section.

[0155] The present invention also provides a computer program product, which is suitable for executing the steps of the control method of the above steel support axial force self-servo control section when executed on a data processing device.

[0156] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0157] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and modules can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0158] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0159] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 each process or multiple processes and / or blocks Figure 1means for the functions specified in one or more boxes.

[0160] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction means, and the instruction means implements the process Figure 1 one process or more processes and / or boxes Figure 1 the functions specified in one box or more boxes.

[0161] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable device provide for implementing the process Figure 1 one process or more processes and / or boxes Figure 1 the steps of the functions specified in one box or more boxes.

[0162] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0163] The memory may include non-permanent memory in computer-readable media, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

[0164] Computer-readable media includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0165] It should also be noted that the term "comprise", "include" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements not only includes those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0166] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0167] The above method of the present invention, or a specific system unit, or a part of it, is a pure software architecture and can be distributed through program code on a physical medium such as a hard disk, a CD-ROM, or any electronic device (such as a smart phone, a computer-readable storage medium). When the machine loads the program code and executes it (such as a smart phone loads and executes it), the machine becomes a device for implementing the present invention. The above method and device of the present invention can also be in the form of program code and be transmitted through some transmission media such as cables, optical fibers, or any transmission type. When the program code is received, loaded, and executed by a machine (such as a smart phone), the machine becomes a device for implementing the present invention.

[0168] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A control method for a steel support axial force self-servo control node based on a hybrid neural network, characterized in that: The control method is based on a CNN-BiLSTM-Attention hybrid neural network to autonomously adjust the error value obtained by the steel support axial force self-servo control node algorithm, and by dynamically adjusting the weights of different time steps, the error between the output value of the control node and the preset value converges to the engineering allowable range.

2. The control method of the steel support axial force self-servo control node according to claim 1 is characterized in that: The control method comprises the following steps: (1) Data collection and preprocessing: Collect data related to the axial force of steel supports and preprocess them; (2) CNN feature extraction: Use CNN to extract features from preprocessed data to obtain local feature information; (3) BiLSTM time series modeling: Based on the features extracted by CNN, the BiLSTM model is used to capture the long-term dependencies of time series; (4) Application of Attention Mechanism: Introduce the attention mechanism to weight the temporal features of BiLSTM output; (5) Control strategy generation and execution: Generate control strategies based on model output and adjust the steel support axial force in real time.

3. The control method of the steel support axial force self-servo control node according to claim 2 is characterized in that: The CNN in step (2) includes multiple convolutional layers and pooling layers, which are used to extract multi-scale local features.

4. The control method of the steel support axial force self-servo control node according to claim 2 is characterized in that: The BiLSTM model in step (3) is composed of two LSTM networks, forward and reverse, and can process the forward and backward information of the time series simultaneously.

5. The control method of the steel support axial force self-servo control node according to claim 2 is characterized in that: In the step (4), the key features are automatically focused by calculating the importance weights of different time steps or features.

6. The control system of the steel support axial force self-servo control node based on hybrid neural network is characterized in that: include: Data acquisition module, used to collect data related to the axial force of steel supports; A data preprocessing module preprocesses the data collected by the data acquisition module; The CNN-BiLSTM-Attention model is constructed based on the CNN-BiLSTM-Attention hybrid neural network architecture, and can interact with the data preprocessing module to obtain data preprocessed by the data preprocessing module and complete model training; after completing the training, the CNN-BiLSTM-Attention model can extract features for the input data, model time series, and weight key features; The control strategy generation module interacts with the CNN-BiLSTM-Attention model for data and generates a control strategy for optimizing the steel support axial force self-servo control node control algorithm based on the model output.

7. The control system of the steel support axial force self-servo control node according to claim 6 is characterized in that: The CNN-BiLSTM-Attention model consists of a five-layer structure: an input layer, a convolutional neural network layer, a bidirectional long short-term memory network layer, an attention mechanism layer, and an output layer.

8. The control system of the steel support axial force self-servo control node according to claim 7 is characterized in that: The convolutional neural network layer includes multiple convolutional layers and pooling layers, which are used to extract multi-scale local features.

9. The control system of the steel support axial force self-servo control node according to claim 7 is characterized in that: The bidirectional long short-term memory network layer is composed of two LSTM networks, a forward LSTM network and a reverse LSTM network, and can process the forward and backward information of the time series at the same time.

10. The control system of the steel support axial force self-servo control node according to claim 7, characterized in that: The attention mechanism layer automatically focuses on key features by calculating the importance weights of different time steps or features.

Images