CN-119292361-B - Pressure-adjustable pressure control system

CN119292361BCN 119292361 BCN119292361 BCN 119292361BCN-119292361-B

Abstract

The invention discloses an adjustable pressure type pressure control system which comprises a data acquisition module, a data preprocessing module, a pressure state detection model construction 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, in particular to an adjustable pressure type pressure control system, which uses a correlation coefficient and a multi-head attention mechanism, and adds space context information by residual connection to obtain a global space pressure characteristic matrix, respectively calculates local space and time pressure characteristics and splices the local space and time pressure characteristics to obtain local pressure characteristics, and introduces a position code and a multi-head self-attention mechanism to calculate global space-time pressure characteristics; based on the division probability, dividing the individuals into excellent individuals, general individuals and poor individuals, designing three different updating methods for updating the individual positions, determining the optimal model parameters, improving the accuracy of pressure state detection, and improving the overall effect and the intelligent level of the pressure control system.

Inventors

LIU FEI
LIN CONG
LI ZE

Assignees

星辰空间(重庆)航空航天装备智能制造有限公司

Dates

Publication Date: 20260512
Application Date: 20240927

Claims (9)

1. The pressure-adjustable pressure control system is characterized by 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; the data acquisition module acquires historical pressure data; The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction processing on the acquired data; The pressure state detection model building module builds a pressure map, uses a correlation coefficient and a multi-head attention mechanism, adds space context information by residual connection to obtain a global space pressure characteristic matrix, calculates local space and time pressure characteristics respectively, splices the local space and time pressure characteristics to obtain local pressure characteristics, introduces a position coding and multi-head self-attention mechanism to calculate global space-time pressure characteristics, and completes pressure state detection; the pressure state detection model parameter optimization module divides individuals into excellent individuals, general individuals and poor individuals based on division probability, introduces a position updating method for designing excellent individuals by random angles and dynamic sensitivities, introduces a position updating method for designing general individuals by control factors and dynamic convergence factors, introduces a position updating method for designing poor individuals by weighting coefficients and Gaussian functions, and determines optimal model parameters; The pressure control strategy generation module is used for knowing the current pressure state based on the output label of the pressure state detection model and making a pressure control strategy.
2. The pressure-adjustable pressure control system according to claim 1, wherein the pressure state detection model building module specifically comprises the following components: Generating a pressure map, regarding each data in the pressure detection data set as a node, taking the data information of each node as a characteristic vector of the node, presetting a similar threshold value, adding an edge between the nodes corresponding to the two data if the cosine similarity between the two data is larger than the similar threshold value, otherwise, not adding an edge between the nodes corresponding to the two data, and finally obtaining a node characteristic matrix G and a pressure map O= (F, delta, D), wherein F is a node set, delta is an edge set, D is an adjacent matrix, and elements in D represent connectivity between the nodes, if edges exist between the two nodes, the element value is 1, otherwise, the element value is 0; Extracting global space pressure characteristics; Extracting local pressure characteristics, respectively calculating a local space pressure characteristic matrix and a local time pressure characteristic matrix, and then splicing the two to obtain a local pressure characteristic matrix Comprising the following contents: Calculating a local space pressure characteristic matrix, wherein the formula is as follows: ; ; In the formula, Is a local spatial pressure feature matrix, sigmoid (.) is a Sigmoid activation function, E lap is a laplace matrix, W gc is a learnable matrix, I is an identity matrix, a is a degree matrix of d+i; calculating a local time pressure characteristic matrix, wherein the formula is as follows: ; ; ; ; Where u t 、m t and n t are the reset gate, update gate, and candidate hidden states for the t-th time step, respectively, t is the time step index, B u 、b m and B n are the offsets for u t 、m t and n t , respectively, B t-1 is the hidden state for the t-1 th time step, GC (·) is the graph convolution operation, tanh (·) is the hyperbolic tangent activation function, And The local time pressure characteristic matrixes extracted by the t-th time step and the t-1 th time step are respectively; Is the global space pressure characteristic matrix extracted by the t time step; Extracting global space-time pressure characteristics based on local pressure characteristic matrix Calculating a global space-time pressure characteristic matrix; And (3) detecting a pressure state, namely inputting the global space-time pressure characteristic matrix into a normalization layer, a feedforward neural layer and an output layer for processing to obtain the pressure state, and obtaining a loss function of a model based on a mean square error, wherein the formula is as follows: ; ; ; Wherein C and The outputs of the normalization layer and the feedforward neural layer respectively, Is the detection result output by the pressure state detection model, layerNorm is the normalization layer, W 1 and W 2 are the weight matrices of the first and second linear transformations of the feedforward neural layer, respectively, b 1 and b 2 are the offsets of the first and second linear transformations of the feedforward neural layer, respectively, and W y and b y are the weight matrices and offsets of the output layer, respectively.
3. The pressure-adjustable pressure control system of claim 2, wherein the extracting global space pressure features comprises the following specific steps: 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 the node indices, Is a modified linear element activation function with leakage, || is a stitching operation, W xb is a linear transformation weight matrix, g i and g j are feature vectors of the i-th and j-th nodes, respectively, And The characteristic vectors of g i and g j after linear transformation mapping are respectively, beta is a weight vector, and T is a vector transposition operation; the attention weight between nodes is calculated using the following formula: ; In the formula, Is the attention weight between the i-th and j-th nodes; Calculating a global space pressure characteristic matrix, embedding attention in a multi-head mode, and adding space context information by residual connection, wherein the formula is as follows: ; In the formula, Is a global spatial pressure feature matrix, G is a node feature matrix, head is the number of attention heads, head is the index of attention heads, and ReLU (·) is a ReLU activation function.
4. The pressure-adjustable pressure control system of claim 2, wherein the extracting global spatiotemporal pressure features comprises: Position coding in local pressure characteristic matrix The formula used for embedding the position coding information is as follows: ; ; In the formula, Is a local pressure characteristic matrix embedded with position coding information, x is a characteristic dimension index, and w is a local pressure characteristic matrix Is used for the number of feature dimensions of (a), Is position coding information; Computing a global spatiotemporal pressure feature matrix from using a multi-headed self-attention mechanism The global space-time pressure characteristic is calculated, and the formula is as follows: ; ; ; Where Q head 、K head and V head are the query vector, key vector and value vector, respectively, corresponding to the head attention header, 、 And The query weight matrix, the key weight matrix and the value weight matrix corresponding to the head attention head respectively, 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 pressure-adjustable pressure control system according to claim 1, wherein the pressure state detection model parameter optimization module specifically comprises the following components: Initializing individual positions based on a linear transformation weight matrix W xb , a weight vector beta, a leachable matrix W gc and a query weight matrix in a pressure state detection model Key weight matrix Value weight matrix Establishing a search space by using the weight matrix W y of the output layer, the weight matrix W 1 of the first linear transformation of the feedforward nerve layer and the weight matrix W 2 of the second linear transformation, taking the individual position as the representative of the pressure state detection model parameter, taking the loss function of the pressure state detection model established based on the model parameter as the fitness value of the individual, and randomly initializing N individual positions in the search space; Dividing individuals, calculating dividing probability P of each individual before each position update, dividing the individuals into excellent individuals if the dividing probability P is larger than 0.7, and updating the positions of the individuals by using a dynamic sensitivity position update method if the dividing probability P is larger than 0.7 And less than or equal to 0.7, dividing the individual into general individuals, and updating the individual position by using a dynamic convergence position updating method, wherein if the dividing probability P is less than or equal to Dividing the individuals into worse individuals, and updating the individual positions by using a distribution estimation updating method, wherein the formula for calculating the dividing probability of the individuals is as follows: ; In the formula, And The division probability and the position of the v-th individual at the H-th iteration, H is the current iteration number, H is the maximum iteration number, v is the individual index, Is the fitness value of the v-th individual at the h iteration, Is the relative ranking of the v th individual after the fitness value of the v th individual is added with 1 in the h iteration, and rank (·) is an ascending ranking function; Updating the individual position; The method comprises the steps of determining optimal model parameters, presetting an fitness threshold, updating an individual fitness value and a global optimal individual, when the fitness value corresponding to the global optimal individual is lower than the fitness threshold, setting the model parameters corresponding to the global optimal individual position as the optimal model parameters, constructing a pressure state detection model based on the optimal model parameters, otherwise, reinitializing the individual position if the maximum iteration number is reached, otherwise, adding 1 to the iteration number and reclassifying the individual to update the individual position.
6. The pressure-adjustable pressure control system of claim 5, wherein the individual location update comprises: the dynamic sensitivity position updating method is used for updating the position of an excellent individual by introducing a random angle theta and a dynamic sensitivity design position updating method, and the used formula is as follows: ; ; In the formula, Is that The updated position using the dynamic sensitivity position updating method, Is the position of the a1 st excellent individual at the h iteration, s (h) is the sensitivity at the h iteration, s max is the maximum sensitivity, The position of the global optimal individual in the h iteration, the global optimal individual is the individual with the minimum fitness value, the rand (·) is a random number generation function, and the value range of the random angle theta is 0 to 2 pi; The dynamic convergence position updating method is used for updating the position of a general individual by introducing a control factor and a dynamic convergence factor design position updating method, and the used formula is as follows: ; ; In the formula, Is that Updated positions using a dynamic convergence position updating method, Is the position of the a2 nd general individual at the h iteration, d min and d max are the minimum and maximum values of the convergence factor, respectively, d (h) is the convergence factor at the h iteration, r 1 and r 2 are the first and second random numbers within the (0, 1) range, respectively, and k is the control factor; the distribution estimation position updating method is used for updating the position of a worse individual by introducing a weighting coefficient and a Gaussian function design position updating method, and the used formula is as follows: ; ; ; ; In the formula, Is that The updated position using the distribution estimation position updating method, Is the position of the a3 rd worse individual at the h iteration, xmean (h) and cov (h) are the weighted average and weighted covariance at the h iteration, ω v is the weighting coefficient of the v th individual, and gauss (·) is a gaussian function, respectively.
7. The pressure-adjustable pressure control system of claim 1, wherein the pressure control strategy generation module is used for collecting real-time pressure data, the real-time pressure data comprises a time stamp, a pressure value, environment data and equipment operation state data, preprocessing the real-time pressure data, inputting the preprocessed real-time pressure data into a pressure state detection model constructed based on optimal model parameters, knowing the state of the current pressure based on an output label of the model, and formulating a pressure control strategy.
8. The pressure-adjustable pressure control system according to claim 1, wherein the data acquisition module acquires historical pressure data, wherein the historical pressure data comprises a time stamp, a pressure value, environment data, equipment operation state data and a pressure state, and the pressure state is used as a data tag.
9. The pressure-adjustable pressure control system according to claim 1, wherein the data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data, the data cleaning comprises processing missing values, abnormal values and repeated values, the data conversion is to convert the data into a vector form, the data normalization is to unify a data range based on a maximum and minimum normalization method, and the data set construction is to construct a pressure detection data set based on the processed data.

Description

Pressure-adjustable pressure control system Technical Field The invention belongs to the technical field of pressure control, and particularly relates to an adjustable pressure type pressure control system. Background The pressure control system utilizes an artificial intelligent algorithm to monitor the pressure state in real time, accurately adjusts and controls the pressure, and enhances the stability and safety of the system. However, the existing pressure control system has the problems that the pressure data is complex, the characteristic extraction is incomplete, the complex association and dynamic change in the pressure data cannot be accurately captured, and the pressure state detection is inaccurate, and the existing pressure control system has the problems that the model parameters are difficult to accurately optimize, the searching efficiency is low, and the pressure control strategy is inaccurate. Disclosure of Invention Aiming at the problems that pressure state detection is inaccurate due to complex association and dynamic change in pressure data, which are caused by incomplete feature extraction and incapability of accurately capturing the complex association and dynamic change in the pressure data, the scheme builds a pressure map, uses a correlation coefficient and a multi-head attention mechanism, adds spatial context information by residual connection, obtains a global spatial pressure feature matrix, calculates and splices local spatial and time pressure features respectively, obtains local pressure features, introduces a position coding and multi-head self-attention mechanism to calculate global space-time pressure features, completes pressure state detection, effectively integrates information of spatial and time dimensions, improves accuracy and reliability of pressure state detection, can better adapt to complex and changeable pressure environments and demands, and aims at solving the problems that a model parameter is difficult to accurately optimize and search efficiency is low and a pressure control strategy is inaccurate in the existing pressure control system, the scheme divides an individual into an excellent individual based on dividing probability, introduces a random angle and a poor individual, designs the position updating method of the individual, introduces a control factor and a poor individual dynamic updating method, introduces the excellent individual position updating factor and a poor individual dynamic updating method, and optimizes the overall position updating method, and improves the accuracy of the pressure state model, thereby improving the accuracy and the accuracy of the pressure state detection. The invention provides an adjustable pressure type pressure control system which 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, wherein the data acquisition module is used for acquiring data of a pressure state of a pressure sensor; the data acquisition module acquires historical pressure data; The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction processing on the acquired data; The pressure state detection model building module builds a pressure map, uses a correlation coefficient and a multi-head attention mechanism, adds space context information by residual connection to obtain a global space pressure characteristic matrix, calculates local space and time pressure characteristics respectively, splices the local space and time pressure characteristics to obtain local pressure characteristics, introduces a position coding and multi-head self-attention mechanism to calculate global space-time pressure characteristics, and completes pressure state detection; the pressure state detection model parameter optimization module divides individuals into excellent individuals, general individuals and poor individuals based on division probability, introduces a position updating method for designing excellent individuals by random angles and dynamic sensitivities, introduces a position updating method for designing general individuals by control factors and dynamic convergence factors, introduces a position updating method for designing poor individuals by weighting coefficients and Gaussian functions, and determines optimal model parameters; The pressure control strategy generation module is used for knowing the current pressure state based on the output label of the pressure state detection model and making a pressure control strategy. Further, the data acquisition module acquires historical pressure data, wherein the historical pressure data comprises a time stamp, a pressure value, environment data, equipment operation state data and a pressure state, and the pressure state is used as a data tag. The data preprocessing module i