Pressure-adjustable pressure control system

Abstract

The invention discloses a pressure-adjustable pressure control system. The system comprises a data acquisition module, a data preprocessing module, a pressure state detection model building module, a pressure state detection model parameter optimization module and a pressure control strategy generation module. The invention belongs to the technical field of pressure control, and particularly relates to a pressure-adjustable pressure control system, which uses a correlation coefficient and a multi-head attention mechanism, adds spatial context information by using residual connection to obtain a global spatial pressure feature matrix, calculates local spatial pressure features and time pressure features respectively, and splices the local spatial pressure features and the time pressure features to obtain a global spatial pressure feature matrix; obtaining local pressure features, and introducing position coding and a multi-head self-attention mechanism to calculate global space-time pressure features; individuals are divided into excellent individuals, general individuals and poorer individuals based on the division probability, three different updating methods are designed for individual position updating, optimal model parameters are determined, the accuracy of pressure state detection is improved, and the overall effect and the intelligent level of a pressure control system are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of pressure control, and in particular refers to an adjustable pressure control system. Background Art

[0002] The pressure control system uses artificial intelligence algorithms to monitor the pressure state in real time, accurately adjust and control the pressure, and enhance the stability and safety of the system. However, the existing pressure control system has problems such as complex pressure data, incomplete feature extraction, and inability to accurately capture the complex associations and dynamic changes in the pressure data, resulting in inaccurate pressure state detection; the existing pressure control system has problems such as difficulty in accurately optimizing model parameters and low search efficiency, resulting in inaccurate pressure control strategies. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an adjustable pressure control system. In view of the problems in the existing pressure control system, such as complex pressure data, incomplete feature extraction, and inability to accurately capture the complex correlations and dynamic changes in the pressure data, resulting in inaccurate pressure state detection, this solution constructs a pressure map, uses correlation coefficients and a multi-head attention mechanism, and uses residual connections to add spatial context information to obtain a global spatial pressure feature matrix, calculates local spatial and temporal pressure features respectively and splices them to obtain local pressure features, introduces position encoding and a multi-head self-attention mechanism to calculate global spatiotemporal pressure features, completes pressure state detection, effectively integrates information in spatial and temporal dimensions, and improves the accuracy of pressure state detection. Accuracy and reliability, can better adapt to complex and changeable pressure environment and needs; in view of the problem that model parameters in existing pressure control systems are difficult to accurately optimize and the search efficiency is low, resulting in inaccurate pressure control strategies, this scheme divides individuals into excellent, general and poor individuals based on the division probability, introduces random angles and dynamic sensitivity to design the position update method of excellent individuals, introduces control factors and dynamic convergence factors to design the position update method of general individuals, introduces weighting coefficients and Gaussian functions to design the position update method of poor individuals, determines the optimal model parameters, improves the efficiency and accuracy of model parameter optimization, and helps to achieve more accurate and reliable pressure state detection, thereby improving the overall effect and intelligence level of the pressure control system.

[0004] The present invention provides an adjustable pressure control system, comprising a data acquisition module, a data preprocessing module, a pressure state detection model building module, a pressure state detection model parameter optimization module and a pressure control strategy generation module;

[0005] The data acquisition module acquires historical pressure data;

[0006] The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data;

[0007] The pressure state detection model building module builds a pressure map, uses the correlation coefficient and multi-head attention mechanism, and uses residual connection to add spatial context information to obtain a global spatial pressure feature matrix, calculates and concatenates local spatial and temporal pressure features to obtain local pressure features, introduces position encoding and multi-head self-attention mechanism to calculate global spatiotemporal pressure features, and completes pressure state detection;

[0008] The stress state detection model parameter optimization module divides individuals into excellent, general and poor individuals based on the division probability, introduces random angles and dynamic sensitivity to design a position update method for excellent individuals, introduces control factors and dynamic convergence factors to design a position update method for general individuals, introduces weighted coefficients and Gaussian functions to design a position update method for poor individuals, and determines the optimal model parameters;

[0009] The pressure control strategy generation module understands the current pressure state based on the output label of the pressure state detection model and formulates a pressure control strategy.

[0010] Furthermore, the data acquisition module collects historical pressure data; the historical pressure data includes timestamp, pressure value, environmental data, equipment operation status data and pressure status; and the pressure status is used as a data label.

[0011] Furthermore, the data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data; data cleaning includes processing missing values, outliers and duplicate values; data conversion is to convert data into vector form; data normalization is based on the maximum and minimum normalization method to unify the data range; and data set construction is to construct a pressure detection data set based on the processed data.

[0012] Furthermore, the stress state detection model building module specifically includes the following contents:

[0013] Generate a pressure map. Treat each data in the pressure detection data set as a node. Take the data information of each node as the feature vector of the node. Preset a similarity threshold. If the cosine similarity between two data is greater than the similarity threshold, add an edge between the nodes corresponding to the two data. Otherwise, no edge is added between the nodes corresponding to the two data. Finally, the node feature matrix G and the pressure map O=(F, δ, D) are obtained, where F is the node set, δ is the edge set, and D is the adjacency matrix. The elements in D represent the connectivity between nodes. If there is an edge between two nodes, the element value is 1, otherwise the element value is 0.

[0014] Extract global space pressure features, including the following:

[0015] The correlation coefficient between nodes is calculated using the following formula:

[0016] ;

[0017] ;

[0018] In the formula, is the correlation coefficient between the i-th and j-th nodes, i and j are node indices, LeakyReLU (·) is the leaky rectified linear unit activation function, || is the concatenation operation, and W xb is the linear transformation weight matrix, g i and g j are the feature vectors of the i-th and j-th nodes respectively, and They are g i and g j The feature vector after linear transformation mapping, β is the weight vector, and T is the vector transposition operation;

[0019] To calculate the attention weight between nodes, the formula used is as follows:

[0020] ;

[0021] In the formula, is the attention weight between the i-th and j-th nodes;

[0022] Calculate the global spatial pressure feature matrix, embed attention in a multi-head manner, and use residual connections to add spatial context information; through multi-head and residual connections, the spatial features are captured more comprehensively; the formula used is as follows:

[0023] ;

[0024] In the formula, is the global spatial pressure feature matrix, G is the node feature matrix, Head is the number of attention heads, head is the index of the attention head, and ReLU (·) is the ReLU activation function;

[0025] Extract local pressure features, calculate the local spatial pressure feature matrix and the local temporal pressure feature matrix respectively, and then concatenate the two to obtain the local pressure feature matrix , including the following:

[0026] Calculate the local space pressure feature matrix to better extract the local space pressure features. The formula used is as follows:

[0027] ;

[0028] ;

[0029] In the formula, is the local spatial pressure feature matrix, Sigmoid (•) is the Sigmoid activation function, E lap is the Laplacian matrix, W gc is a learnable matrix, I is the identity matrix, and A is the degree matrix of D+I;

[0030] Calculate the local time pressure feature matrix to accurately capture the pressure change characteristics in the time dimension. The formula used is as follows:

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] In the formula, u t 、m t and n t are the reset gate, update gate, and candidate hidden state of the tth time step, t is the time step index, and b u 、b m and b n u t 、m t and n t The bias, B t-1 is the hidden state at the t-1th time step, GC(·) is the graph convolution operation, tanh(·) is the hyperbolic tangent activation function, and are the local time pressure feature matrices extracted at the t-th and t-1-th time steps, respectively; is the global spatial pressure feature matrix extracted at the t-th time step;

[0036] Extract global spatiotemporal pressure features based on the local pressure feature matrix Calculate the global spatiotemporal pressure characteristic matrix, including the following:

[0037] Position encoding, in the local pressure feature matrix Embed position encoding information in the model to enhance the model's perception of time and space. The formula used is as follows:

[0038] ;

[0039] ;

[0040] In the formula, is the local pressure feature matrix after embedding the position encoding information, x is the feature dimension index, and w is the local pressure feature matrix The number of feature dimensions, is the position encoding information;

[0041] Calculate the global spatiotemporal pressure feature matrix and use the multi-head self-attention mechanism from The global spatiotemporal pressure characteristics are calculated in ; the characteristics of different spatiotemporal positions are comprehensively considered; the formula used is as follows:

[0042] ;

[0043] ;

[0044] ;

[0045] In the formula, Q head , K head and V head are the query vector, key vector, and value vector corresponding to the head-th attention head, respectively. , and They are the query weight matrix, key weight matrix, and value weight matrix corresponding to the head-th attention head, is the self-attention score of the head, U 1 is the global spatiotemporal pressure feature matrix, Softmax(·) is the normalized exponential function;

[0046] Pressure state detection, the global spatiotemporal pressure feature matrix is ​​input into the normalization layer, the feedforward neural layer and the output layer for processing to obtain the pressure state, and the loss function of the model is obtained based on the mean square error. The formula used is as follows:

[0047] ;

[0048] ;

[0049] ;

[0050] In the formula, C and are the outputs of the normalization layer and the feedforward neural layer, is the detection result output by the pressure state detection model, LayerNorm is the normalization layer, W1 and W2 are the weight matrices of the first and second linear transformations of the feedforward neural layer, b1 and b2 are the biases of the first and second linear transformations of the feedforward neural layer, W y and b y are the weight matrix and bias of the output layer respectively.

[0051] Furthermore, the pressure state detection model parameter optimization module specifically includes the following contents:

[0052] Initialize the individual position based on the linear transformation weight matrix W in the pressure state detection model xb , weight vector β, learnable matrix W gc , query weight matrix , key weight matrix , value weight matrix , the weight matrix W of the output layer y , the weight matrix W1 of the first linear transformation of the feedforward neural layer and the weight matrix W2 of the second linear transformation are used to establish a search space, the individual position is used as a representative of the pressure state detection model parameters, the loss function of the pressure state detection model established based on the model parameters is used as the individual fitness value, and N individual positions are randomly initialized in the search space;

[0053] Divide individuals. Before each position update, calculate the division probability P of each individual. If the division probability P is greater than 0.7, the individual is divided into excellent individuals and the dynamic sensitivity position update method is used to update the individual position. If the division probability P is greater than and is less than or equal to 0.7, the individual is divided into general individuals and the dynamic convergence position update method is used to update the individual position; if the division probability P is less than or equal to , the individuals are divided into poor individuals, and the distribution estimation update method is used to update the individual positions; through this detailed division, different update strategies can be adopted for different types of individuals to improve the optimization efficiency and effect; the formula used to calculate the individual division probability is as follows:

[0054] ;

[0055] In the formula, and are the partition probability and position of the vth individual at the hth iteration, h is the current iteration number, H is the maximum iteration number, v is the individual index, is the fitness value of the vth individual at the hth iteration, is the relative ranking of the fitness value of the vth individual among N individuals after adding 1 to it in the hth iteration, and rank(·) is the ascending ranking function;

[0056] Individual location updates, including the following:

[0057] Dynamic sensitivity position update method, introduces random angle θ and dynamic sensitivity to design position update method, updates the position of excellent individuals, and enables excellent individuals to move more flexibly in the search space; the formula used is as follows:

[0058] ;

[0059] ;

[0060] In the formula, yes The updated position using the dynamic sensitivity position update method, is the position of the a1th excellent individual at the hth iteration, s(h) is the sensitivity at the hth iteration, s max is the maximum sensitivity, is the position of the global optimal individual at the hth iteration. The global optimal individual is the individual with the minimum fitness value. rand(·) is a random number generator function. The value range of the random angle θ is 0 to 2π.

[0061] The dynamic convergence position update method introduces the control factor and the dynamic convergence factor to design the position update method, and updates the position of general individuals; making the update of general individuals more targeted; the formula used is as follows:

[0062] ;

[0063] ;

[0064] In the formula, yes The updated position using the dynamic convergence position update method, is the position of the a2th general individual at the hth iteration, d min and d max are the minimum and maximum values ​​of the convergence factor, d(h) is the convergence factor at the hth iteration, r1 and r2 are the first and second random numbers in the range of (0, 1), and k is the control factor;

[0065] The distribution estimation position update method introduces weighted coefficients and Gaussian functions to design a position update method to update the position of the poor individuals; guide the poor individuals to move in the direction where the best individuals will appear; the formula used is as follows:

[0066] ;

[0067] ;

[0068] ;

[0069] ;

[0070] In the formula, yes The updated position using the distribution estimation position update method, is the position of the a3th poor individual at the hth iteration, xmean(h) and cov(h) are the weighted mean and weighted covariance at the hth iteration, respectively, ω v is the weight coefficient of the vth individual, Gaussion (·) is the Gaussian function;

[0071] The optimal model parameters are determined, the fitness threshold is set in advance, the individual fitness value and the global optimal individual are updated, when the fitness value corresponding to the global optimal individual is lower than the fitness threshold, the model parameters corresponding to the global optimal individual position are the optimal model parameters, and the pressure state detection model is constructed based on the optimal model parameters; otherwise, if the maximum number of iterations is reached, the individual position is reinitialized; otherwise, the number of iterations is increased by 1 and the individuals are re-divided to update the individual positions.

[0072] Furthermore, the pressure control strategy generation module collects real-time pressure data; the real-time pressure data includes timestamp, pressure value, environmental data and equipment operation status data. After preprocessing the real-time pressure data, it is input into the pressure status detection model constructed based on the optimal model parameters, and the current pressure status is understood based on the output label of the model to formulate a pressure control strategy.

[0073] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0074] (1) In view of the problem that the existing pressure control system has complex pressure data, incomplete feature extraction, and cannot accurately capture the complex correlations and dynamic changes in the pressure data, resulting in inaccurate pressure state detection, this scheme constructs a pressure map, uses correlation coefficients and multi-head attention mechanisms, and uses residual connections to add spatial context information to obtain a global spatial pressure feature matrix. The local spatial and temporal pressure features are calculated and concatenated to obtain local pressure features. Position encoding and multi-head self-attention mechanisms are introduced to calculate global spatiotemporal pressure features to complete pressure state detection, effectively integrating information in spatial and temporal dimensions, and more comprehensively and accurately capturing the characteristics and relationships of pressure data. This improves the accuracy and reliability of pressure state detection, provides strong support for the precise regulation of the pressure control system, and can better adapt to complex and changing pressure environments and demands.

[0075] (2) In view of the problem that model parameters in existing pressure control systems are difficult to accurately optimize and the search efficiency is low, resulting in inaccurate pressure control strategies, this scheme divides individuals into excellent, general and poor individuals based on the division probability, introduces random angles and dynamic sensitivity to design the position update method of excellent individuals, introduces control factors and dynamic convergence factors to design the position update method of general individuals, introduces weighting coefficients and Gaussian functions to design the position update method of poor individuals, determines the optimal model parameters, improves the efficiency and accuracy of model parameter optimization, enhances the targetedness and adaptability of optimization, can better adapt to different situations, and helps to achieve more accurate and reliable pressure state detection, thereby improving the overall effect and intelligence level of the pressure control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A schematic diagram of an adjustable pressure control system provided by the present invention;

[0077] Figure 2 A schematic diagram of a pressure state detection model module;

[0078] Figure 3 Schematic diagram of the parameter optimization module for the stress state detection model.

[0079] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0080] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0081] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0082] Example 1, see Figure 1 , the present invention provides an adjustable pressure control system, including a data acquisition module, a data preprocessing module, a pressure state detection model building module, a pressure state detection model parameter optimization module and a pressure control strategy generation module;

[0083] The data acquisition module collects historical pressure data and sends the data to the data preprocessing module;

[0084] The data preprocessing module receives the data sent by the data acquisition module, performs data cleaning, data conversion, data normalization and data set construction on the collected data, and sends the data to the pressure state detection model construction module;

[0085] The pressure state detection model building module receives the data sent by the data preprocessing module, builds a pressure map, uses the correlation coefficient and the multi-head attention mechanism, and uses the residual connection to add spatial context information to obtain a global spatial pressure feature matrix, calculates and splices the local spatial and temporal pressure features respectively, obtains the local pressure features, introduces position encoding and the multi-head self-attention mechanism to calculate the global spatiotemporal pressure features, completes the pressure state detection, and sends the data to the pressure state detection model parameter optimization module;

[0086] The pressure state detection model parameter optimization module receives the data sent by the pressure state detection model construction module, divides the individuals into excellent, general and poor individuals based on the division probability, introduces random angles and dynamic sensitivity to design a position update method for excellent individuals, introduces control factors and dynamic convergence factors to design a position update method for general individuals, introduces weighted coefficients and Gaussian functions to design a position update method for poor individuals, determines the optimal model parameters, and sends the data to the pressure control strategy generation module;

[0087] The pressure control strategy generation module receives data sent by the pressure state detection model parameter optimization module, understands the current pressure state based on the output label of the pressure state detection model, and formulates a pressure control strategy.

[0088] Example 2, see Figure 1 This embodiment is based on the above embodiment, and the data acquisition module collects historical pressure data; the historical pressure data includes timestamp, pressure value, environmental data, equipment operation status data and pressure status; and the pressure status is used as a data label.

[0089] Example 3, see Figure 1 This embodiment is based on the above embodiment, and the data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data; data cleaning includes processing missing values, abnormal values ​​and duplicate values; data conversion is to convert the data into vector form; data normalization is based on the maximum and minimum normalization method to unify the data range; and data set construction is to construct a pressure detection data set based on the processed data.

[0090] Example 4, see Figure 1 and Figure 2Based on the above embodiment, this embodiment constructs a pressure state detection model module, which specifically includes the following contents:

[0091] Generate a pressure map. Treat each data in the pressure detection data set as a node. Take the data information of each node as the feature vector of the node. Preset a similarity threshold. If the cosine similarity between two data is greater than the similarity threshold, add an edge between the nodes corresponding to the two data. Otherwise, no edge is added between the nodes corresponding to the two data. Finally, the node feature matrix G and the pressure map O=(F, δ, D) are obtained, where F is the node set, δ is the edge set, and D is the adjacency matrix. The elements in D represent the connectivity between nodes. If there is an edge between two nodes, the element value is 1, otherwise the element value is 0.

[0092] Extract global space pressure features, including the following:

[0093] Calculate the correlation coefficient between nodes to effectively capture the degree of association between nodes. The formula used is as follows:

[0094] ;

[0095] ;

[0096] In the formula, is the correlation coefficient between the i-th and j-th nodes, i and j are node indices, LeakyReLU (·) is the leaky rectified linear unit activation function, || is the concatenation operation, and W xb is the linear transformation weight matrix, g i and g j are the feature vectors of the i-th and j-th nodes respectively, and They are g i and g j The feature vector after linear transformation mapping, β is the weight vector, and T is the vector transposition operation;

[0097] Calculate the attention weights between nodes and highlight important node relationships. The formula used is as follows:

[0098] ;

[0099] In the formula, is the attention weight between the i-th and j-th nodes;

[0100] Calculate the global spatial pressure feature matrix, embed attention in a multi-head manner, and use residual connections to add spatial context information; through multi-head and residual connections, more comprehensive spatial features are captured and the expression ability of the model is improved; the formula used is as follows:

[0101] ;

[0102] In the formula, is the global spatial pressure feature matrix, G is the node feature matrix, Head is the number of attention heads, head is the index of the attention head, and ReLU (·) is the ReLU activation function;

[0103] Extract local pressure features, calculate the local spatial pressure feature matrix and the local temporal pressure feature matrix respectively, and then concatenate the two to obtain the local pressure feature matrix , including the following:

[0104] Calculate the local space pressure feature matrix to better extract the local space pressure features. The formula used is as follows:

[0105] ;

[0106] ;

[0107] In the formula, is the local spatial pressure feature matrix, Sigmoid (•) is the Sigmoid activation function, E lap is the Laplacian matrix, W gc is a learnable matrix, I is the identity matrix, and A is the degree matrix of D+I;

[0108] Calculate the local time pressure feature matrix to accurately capture the pressure change characteristics in the time dimension. The formula used is as follows:

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] In the formula, u t 、m t and n t are the reset gate, update gate, and candidate hidden state of the tth time step, t is the time step index, and b u 、b m and b n u t 、m t and n t The bias, B t-1 is the hidden state at the t-1th time step, GC(·) is the graph convolution operation, tanh(·) is the hyperbolic tangent activation function, and are the local time pressure feature matrices extracted at the t-th and t-1-th time steps, respectively; is the global spatial pressure feature matrix extracted at the t-th time step;

[0114] Extract global spatiotemporal pressure features based on the local pressure feature matrix Calculate the global spatiotemporal pressure characteristic matrix, including the following:

[0115] Position encoding, in the local pressure feature matrix Embed position encoding information in the model to enhance the model's perception of time and space. The formula used is as follows:

[0116] ;

[0117] ;

[0118] In the formula, is the local pressure feature matrix after embedding the position encoding information, x is the feature dimension index, and w is the local pressure feature matrix The number of feature dimensions, is the position encoding information;

[0119] Calculate the global spatiotemporal pressure feature matrix and use the multi-head self-attention mechanism from The global spatiotemporal pressure characteristics are calculated in the following way; the characteristics of different spatiotemporal positions are comprehensively considered to improve the accuracy of the model; the formula used is as follows:

[0120] ;

[0121] ;

[0122] ;

[0123] In the formula, Q head , K head and V head are the query vector, key vector, and value vector corresponding to the head-th attention head, respectively. , and They are the query weight matrix, key weight matrix, and value weight matrix corresponding to the head-th attention head, is the self-attention score of the head, U 1 is the global spatiotemporal pressure feature matrix, Softmax(·) is the normalized exponential function;

[0124] Pressure state detection, the global spatiotemporal pressure feature matrix is ​​input into the normalization layer, the feedforward neural layer and the output layer for processing to obtain the pressure state, and the loss function of the model is obtained based on the mean square error. The formula used is as follows:

[0125] ;

[0126] ;

[0127] ;

[0128] In the formula, C and are the outputs of the normalization layer and the feedforward neural layer, is the detection result output by the pressure state detection model, LayerNorm is the normalization layer, W1 and W2 are the weight matrices of the first and second linear transformations of the feedforward neural layer, b1 and b2 are the biases of the first and second linear transformations of the feedforward neural layer, W y and b y are the weight matrix and bias of the output layer respectively.

[0129] By performing the above operations, in order to solve the problem that the existing pressure control system has complex pressure data, incomplete feature extraction, and cannot accurately capture the complex correlations and dynamic changes in the pressure data, resulting in inaccurate pressure state detection, this solution constructs a pressure map, uses correlation coefficients and multi-head attention mechanisms, and uses residual connections to add spatial context information to obtain a global spatial pressure feature matrix, calculates local spatial and temporal pressure features respectively and concatenates them to obtain local pressure features, introduces position encoding and multi-head self-attention mechanisms to calculate global spatiotemporal pressure features, completes pressure state detection, effectively integrates information in spatial and temporal dimensions, and more comprehensively and accurately captures the characteristics and relationships of pressure data, thereby improving the accuracy and reliability of pressure state detection, providing strong support for the precise regulation of the pressure control system, and being able to better adapt to complex and changing pressure environments and demands.

[0130] Example 5, see Figure 1 and Figure 3 Based on the above embodiment, this embodiment includes the following contents:

[0131] Initialize the individual position based on the linear transformation weight matrix W in the pressure state detection model xb , weight vector β, learnable matrix W gc , query weight matrix , key weight matrix , value weight matrix , the weight matrix W of the output layer y, the weight matrix W1 of the first linear transformation of the feedforward neural layer and the weight matrix W2 of the second linear transformation are used to establish a search space, the individual position is used as a representative of the pressure state detection model parameters, the loss function of the pressure state detection model established based on the model parameters is used as the individual fitness value, and N individual positions are randomly initialized in the search space;

[0132] Divide individuals. Before each position update, calculate the division probability P of each individual. If the division probability P is greater than 0.7, the individual is divided into excellent individuals and the dynamic sensitivity position update method is used to update the individual position. If the division probability P is greater than and is less than or equal to 0.7, the individual is divided into general individuals and the dynamic convergence position update method is used to update the individual position; if the division probability P is less than or equal to , the individuals are divided into poor individuals, and the distribution estimation update method is used to update the individual positions; through this detailed division, different update strategies can be adopted for different types of individuals to improve the optimization efficiency and effect; the formula used to calculate the individual division probability is as follows:

[0133] ;

[0134] In the formula, and are the partition probability and position of the vth individual at the hth iteration, h is the current iteration number, H is the maximum iteration number, v is the individual index, is the fitness value of the vth individual at the hth iteration, is the relative ranking of the fitness value of the vth individual among N individuals after adding 1 to it in the hth iteration, and rank(·) is the ascending ranking function;

[0135] Individual location updates, including the following:

[0136] Dynamic sensitivity position update method, introduces random angle θ and dynamic sensitivity to design position update method, and updates the position of excellent individuals; it enables excellent individuals to move more flexibly in the search space, which helps to find better model parameters faster; the formula used is as follows:

[0137] ;

[0138] ;

[0139] In the formula, yes The updated position using the dynamic sensitivity position update method, is the position of the a1th excellent individual at the hth iteration, s(h) is the sensitivity at the hth iteration, s maxis the maximum sensitivity, is the position of the global optimal individual at the hth iteration. The global optimal individual is the individual with the minimum fitness value. rand(·) is a random number generator function. The value range of the random angle θ is 0 to 2π.

[0140] The dynamic convergence position update method introduces the control factor and the dynamic convergence factor to design the position update method, and updates the position of general individuals; it makes the update of general individuals more targeted and promotes the overall optimization process; the formula used is as follows:

[0141] ;

[0142] ;

[0143] In the formula, yes The updated position using the dynamic convergence position update method, is the position of the a2th general individual at the hth iteration, d min and d max are the minimum and maximum values ​​of the convergence factor, d(h) is the convergence factor at the hth iteration, r1 and r2 are the first and second random numbers in the range of (0, 1), and k is the control factor;

[0144] The distribution estimation position update method introduces weighted coefficients and Gaussian functions to design a position update method to update the positions of poor individuals; guide poor individuals to move in the direction where the optimal individuals will appear, thereby improving the overall optimization effect; the formula used is as follows:

[0145] ;

[0146] ;

[0147] ;

[0148] ;

[0149] In the formula, yes The updated position using the distribution estimation position update method, is the position of the a3th poor individual at the hth iteration, xmean(h) and cov(h) are the weighted mean and weighted covariance at the hth iteration, respectively, ω v is the weight coefficient of the vth individual, Gaussion (·) is the Gaussian function;

[0150] The optimal model parameters are determined, the fitness threshold is set in advance, the individual fitness value and the global optimal individual are updated, when the fitness value corresponding to the global optimal individual is lower than the fitness threshold, the model parameters corresponding to the global optimal individual position are the optimal model parameters, and the pressure state detection model is constructed based on the optimal model parameters; otherwise, if the maximum number of iterations is reached, the individual position is reinitialized; otherwise, the number of iterations is increased by 1 and the individuals are re-divided to update the individual positions.

[0151] By performing the above operations, in order to solve the problem that model parameters in the existing pressure control system are difficult to accurately optimize and the search efficiency is low, resulting in inaccurate pressure control strategies, this scheme divides individuals into excellent, general and poor individuals based on the division probability, introduces random angles and dynamic sensitivity to design a position update method for excellent individuals, introduces control factors and dynamic convergence factors to design a position update method for general individuals, introduces weighting coefficients and Gaussian functions to design a position update method for poor individuals, determines the optimal model parameters, improves the efficiency and accuracy of model parameter optimization, enhances the pertinence and adaptability of optimization, can better adapt to different situations, and helps to achieve more accurate and reliable pressure state detection, thereby improving the overall effect and intelligence level of the pressure control system.

[0152] Example 6, see Figure 1 This embodiment is based on the above embodiment. The pressure control strategy generation module collects real-time pressure data; the real-time pressure data includes timestamp, pressure value, environmental data and equipment operation status data. After preprocessing the real-time pressure data, it is input into the pressure status detection model constructed based on the optimal model parameters. The current pressure status is understood based on the output label of the model, and a pressure control strategy is formulated.

[0153] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0154] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0155] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An adjustable pressure control system, characterized in that: It includes a data acquisition module, a data preprocessing module, a pressure state detection model building module, a pressure state detection model parameter optimization module and a pressure control strategy generation module; The data acquisition module acquires historical pressure data; The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data; The pressure state detection model building module builds a pressure map, uses the correlation coefficient and multi-head attention mechanism, and uses residual connection to add spatial context information to obtain a global spatial pressure feature matrix, calculates and concatenates local spatial and temporal pressure features to obtain local pressure features, introduces position encoding and multi-head self-attention mechanism to calculate global spatiotemporal pressure features, and completes pressure state detection; The stress state detection model parameter optimization module divides individuals into excellent, general and poor individuals based on the division probability, introduces random angles and dynamic sensitivity to design a position update method for excellent individuals, introduces control factors and dynamic convergence factors to design a position update method for general individuals, introduces weighted coefficients and Gaussian functions to design a position update method for poor individuals, and determines the optimal model parameters; The pressure control strategy generation module understands the current pressure state based on the output label of the pressure state detection model and formulates a pressure control strategy.

2. The adjustable pressure control system according to claim 1, characterized in that: The stress state detection model building module specifically includes the following contents: Generate a pressure map. Treat each data in the pressure detection data set as a node. Take the data information of each node as the feature vector of the node. Preset a similarity threshold. If the cosine similarity between two data is greater than the similarity threshold, add an edge between the nodes corresponding to the two data. Otherwise, no edge is added between the nodes corresponding to the two data. Finally, the node feature matrix G and the pressure map O=(F, δ, D) are obtained, where F is the node set, δ is the edge set, and D is the adjacency matrix. The elements in D represent the connectivity between nodes. If there is an edge between two nodes, the element value is 1, otherwise the element value is 0. Extract global spatial pressure features; Extract local pressure features, calculate the local spatial pressure feature matrix and the local temporal pressure feature matrix respectively, and then concatenate the two to obtain the local pressure feature matrix , including the following: Calculate the local space pressure characteristic matrix, the formula used is as follows: ； ； In the formula, is the local spatial pressure feature matrix, Sigmoid (•) is the Sigmoid activation function, E lap is the Laplacian matrix, W gc is a learnable matrix, I is the identity matrix, and A is the degree matrix of D+I; Calculate the local time pressure characteristic matrix using the following formula: ； ； ； ； In the formula, u t 、m t and n t are the reset gate, update gate, and candidate hidden state of the tth time step, t is the time step index, and b u 、b m and b n u t 、m t and n t The bias, B t-1 is the hidden state at the t-1th time step, GC(·) is the graph convolution operation, tanh(·) is the hyperbolic tangent activation function, and are the local time pressure feature matrices extracted at the t-th and t-1-th time steps, respectively; is the global spatial pressure feature matrix extracted at the t-th time step; Extract global spatiotemporal pressure features based on the local pressure feature matrix Calculate the global spatiotemporal pressure characteristic matrix; Pressure state detection, the global spatiotemporal pressure feature matrix is ​​input into the normalization layer, the feedforward neural layer and the output layer for processing to obtain the pressure state, and the loss function of the model is obtained based on the mean square error. The formula used is as follows: ； ； ； In the formula, C and are the outputs of the normalization layer and the feedforward neural layer, is the detection result output by the pressure state detection model, LayerNorm is the normalization layer, W1 and W2 are the weight matrices of the first and second linear transformations of the feedforward neural layer, b1 and b2 are the biases of the first and second linear transformations of the feedforward neural layer, W y and b y are the weight matrix and bias of the output layer respectively.

3. The adjustable pressure control system according to claim 2, characterized in that: The extraction of global spatial pressure features specifically includes the following contents: The correlation coefficient between nodes is calculated using the following formula: ； ； In the formula, is the correlation coefficient between the i-th and j-th nodes, i and j are node indices, is the leaky rectified linear unit activation function, || is the concatenation operation, and W xb is the linear transformation weight matrix, g i and g j are the feature vectors of the i-th and j-th nodes respectively, and They are g i and g j The feature vector after linear transformation mapping, β is the weight vector, and T is the vector transposition operation; To calculate the attention weight between nodes, the formula used is as follows: ； In the formula, is the attention weight between the i-th and j-th nodes; Calculate the global spatial pressure feature matrix, embed the attention in a multi-head manner, and use residual connections to add spatial context information; the formula used is as follows: ； In the formula, is the global spatial pressure feature matrix, G is the node feature matrix, Head is the number of attention heads, head is the index of the attention head, and ReLU(·) is the ReLU activation function.

4. The adjustable pressure control system according to claim 2, characterized in that: The method of extracting the global spatiotemporal pressure characteristics is specifically Includes the following: Position encoding, in the local pressure feature matrix The position encoding information is embedded in , and the formula used is as follows: ； ； In the formula, is the local pressure feature matrix after embedding the position encoding information, x is the feature dimension index, and w is the local pressure feature matrix The number of feature dimensions, is the position encoding information; Calculate the global spatiotemporal pressure feature matrix and use the multi-head self-attention mechanism from The global space-time pressure characteristics are calculated in the following formula: ； ； ； In the formula, Q head , K head and V head are the query vector, key vector, and value vector corresponding to the head-th attention head, respectively. , and They are the query weight matrix, key weight matrix, and value weight matrix corresponding to the head-th attention head, is the self-attention score of the head, U 1 is the global spatiotemporal pressure feature matrix, and Softmax(·) is the normalized exponential function.

5. The adjustable pressure control system according to claim 1, characterized in that: The pressure state detection model parameter optimization module specifically includes the following contents: Initialize the individual position based on the linear transformation weight matrix W in the pressure state detection model xb , weight vector β, learnable matrix W gc , query weight matrix , key weight matrix , value weight matrix , the weight matrix W of the output layer y , the weight matrix W1 of the first linear transformation of the feedforward neural layer and the weight matrix W2 of the second linear transformation are used to establish a search space, the individual position is used as a representative of the pressure state detection model parameters, the loss function of the pressure state detection model established based on the model parameters is used as the individual fitness value, and N individual positions are randomly initialized in the search space; Divide individuals. Before each position update, calculate the division probability P of each individual. If the division probability P is greater than 0.7, the individual is divided into excellent individuals and the dynamic sensitivity position update method is used to update the individual position. If the division probability P is greater than and is less than or equal to 0.7, the individual is divided into general individuals and the dynamic convergence position update method is used to update the individual position; if the division probability P is less than or equal to , the individual is classified as a poor individual, and the distribution estimation update method is used to update the individual position; the formula used to calculate the probability of individual division is as follows: ； In the formula, and are the partition probability and position of the vth individual at the hth iteration, h is the current iteration number, H is the maximum iteration number, v is the individual index, is the fitness value of the vth individual at the hth iteration, is the relative ranking of the fitness value of the vth individual among N individuals after adding 1 to it in the hth iteration, and rank(·) is the ascending ranking function; Individual location updates; The optimal model parameters are determined, the fitness threshold is pre-set, the individual fitness value and the global optimal individual are updated, and when the fitness value corresponding to the global optimal individual is lower than the fitness threshold, the model parameters corresponding to the global optimal individual position are the optimal model parameters, and the pressure state detection model is constructed based on the optimal model parameters; otherwise, if the maximum number of iterations is reached, the individual position is reinitialized; Otherwise, the number of iterations is increased by 1 and the individuals are re-divided to update the individual positions.

6. The adjustable pressure control system according to claim 5, characterized in that: The individual location update specifically includes the following contents: Dynamic sensitivity position update method, introduces random angle θ and dynamic sensitivity to design position update method, and updates the position of excellent individuals; the formula used is as follows: ； ； In the formula, yes The updated position using the dynamic sensitivity position update method, is the position of the a1th excellent individual at the hth iteration, s(h) is the sensitivity at the hth iteration, s max is the maximum sensitivity, is the position of the global optimal individual at the hth iteration. The global optimal individual is the individual with the minimum fitness value. rand(·) is a random number generator function. The value range of the random angle θ is 0 to 2π. Dynamic convergence position update method, introduces control factors and dynamic convergence factors to design a position update method, and performs position update of general individuals; the formula used is as follows: ； ； In the formula, yes The updated position using the dynamic convergence position update method, is the position of the a2th general individual at the hth iteration, d min and d max are the minimum and maximum values ​​of the convergence factor, d(h) is the convergence factor at the hth iteration, r1 and r2 are the first and second random numbers in the range of (0, 1), and k is the control factor; The distribution estimation position update method introduces weighted coefficients and Gaussian functions to design a position update method to update the position of poor individuals; the formula used is as follows: ； ； ； ； In the formula, yes The updated position using the distribution estimation position update method, is the position of the a3th poor individual at the hth iteration, xmean(h) and cov(h) are the weighted mean and weighted covariance at the hth iteration, respectively, ω v is the weight coefficient of the vth individual, and Gaussion (·) is the Gaussian function.

7. The adjustable pressure control system according to claim 1, characterized in that: The pressure control strategy generation module collects real-time pressure data; the real-time pressure data includes timestamp, pressure value, environmental data and equipment operation status data. After preprocessing the real-time pressure data, it is input into the pressure status detection model constructed based on the optimal model parameters. The current pressure status is understood based on the output label of the model, and a pressure control strategy is formulated.

8. The adjustable pressure control system according to claim 1, characterized in that: The data acquisition module collects historical pressure data; the historical pressure data includes timestamp, pressure value, environmental data, equipment operation status data and pressure status; and the pressure status is used as a data label.

9. The adjustable pressure control system according to claim 1, characterized in that: The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data; data cleaning includes processing missing values, outliers and duplicate values; Data conversion is to convert data into vector form; data normalization is to unify the data range based on the maximum and minimum normalization method; and data set construction is to construct a pressure detection data set based on the processed data.

Images