An environmental protection equipment control method, system, device and medium

Abstract

The application provides an environmental protection equipment control method, system, device and medium, and belongs to the technical field of environmental protection equipment control. The method sets control parameters such as a working temperature threshold, a flow regulation range and a pressure safety limit, then collects running state data of each execution unit of the controlled environmental protection equipment, extracts and encodes features of the control parameters to obtain a low-dimensional feature encoding vector, performs hierarchical filtering processing on the running state data based on feature weights to obtain a key feature aggregation vector, combines historical data to perform correlation matching analysis on the two to obtain a correlation matching encoding vector, and finally inputs the correlation matching encoding vector into a decision maker to generate a control instruction to instruct equipment operation adjustment. Through accurate data processing and correlation analysis, the method improves the precision and adaptability of environmental protection equipment control, improves equipment operation efficiency and environmental protection effect, and has strong practicality and promotional value.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of environmental protection equipment control, and more particularly relates to an environmental protection equipment control method, system, device and medium. BACKGROUND

[0002] With the continuous improvement of environmental protection requirements, environmental protection equipment is increasingly widely used in industrial production, environmental governance and other fields. However, the existing environmental protection equipment control method has many deficiencies in actual application. The traditional control method usually adopts fixed control parameter setting, which is difficult to flexibly adjust according to the different running conditions of the environmental protection equipment and the changes of the external environment, resulting in low equipment running efficiency and inability to fully exert the environmental protection performance. For example, in the face of working temperature fluctuation, flow change or pressure anomaly, the fixed parameter setting may make the equipment in a non-optimal running state, which not only affects the treatment effect of pollutants, but also increases energy consumption.

[0003] The existing technology lacks effective data processing mechanism after collecting the running state data of the environmental protection equipment execution unit. A large amount of running state data contains a large amount of redundant information and irrelevant features. If these data are directly analyzed, not only the data processing workload and computational complexity will be increased, but also the analysis result may be inaccurate due to the interference of irrelevant information, which cannot accurately reflect the real running state of the equipment, and further affect the generation of control instructions and the regulation and control of the equipment.

[0004] When the traditional control method analyzes the correlation between the control parameters and the running state data, the value of the historical data is often not fully utilized. The correlation analysis lacking the support of historical data cannot accurately grasp the internal relationship between the control parameters and the running state, resulting in that the generated control instructions cannot accurately guide the running adjustment operation of the equipment, so that the adaptability of the equipment to complex working conditions is poor, and efficient and stable operation cannot be realized.

[0005] The existing control instruction generation method is relatively simple, usually based on a single decision model or rule, and cannot comprehensively consider the influence of multiple factors on equipment running. This simple generation method leads to insufficient flexibility and accuracy of the control instructions, which cannot meet the control requirements of the environmental protection equipment in different scenarios, and limits the further improvement of the performance of the environmental protection equipment. SUMMARY

[0006] The application aims to provide an environmental protection equipment control method, system, device and medium to solve the problems raised in the background.

[0007] To achieve the above-mentioned purpose, the application provides an environmental protection equipment control method, which comprises:

[0008] Setting control parameters, the control parameters including working temperature threshold, flow adjustment range and pressure safety limit;

[0009] Collecting running state data of each execution unit of the controlled environmental protection equipment under the constraints of the control parameters to obtain a set of execution unit running state data;

[0010] Feature extraction coding of the control parameters to obtain a control parameter low-dimensional feature coding vector;

[0011] Layered filtering processing of the set of execution unit running state data based on feature weights to obtain an execution unit running state key feature aggregation vector;

[0012] Correlation matching analysis of the execution unit running state key feature aggregation vector and the control parameter low-dimensional feature coding vector based on historical data enhancement to obtain a control parameter-running state correlation matching coding vector;

[0013] Based on the control parameter-running state correlation matching coding vector, generating an environmental protection equipment control instruction, specifically: inputting the control parameter-running state correlation matching coding vector into a decision-based control instruction generation module to obtain a device control instruction, the device control instruction being used to instruct the running adjustment operation of the environmental protection equipment.

[0014] In an embodiment of the present application, the layered filtering processing of the set of execution unit running state data based on feature weights to obtain an execution unit running state key feature aggregation vector comprises:

[0015] Respectively passing each execution unit running state data in the set of execution unit running state data through a filtering layer to obtain a set of execution unit running state feature filtering coding vectors;

[0016] Based on the feature distribution density of the set of execution unit running state feature filtering coding vectors, calculating an execution unit running state pseudo-reference filtering center representation vector;

[0017] Calculating the filtering adjustment direction of each execution unit running state feature filtering coding vector in the set of execution unit running state feature filtering coding vectors relative to the execution unit running state pseudo-reference filtering center representation vector to obtain a set of execution unit running state filtering adjustment directions;

[0018] Based on the set of execution unit running state filtering adjustment directions, performing layered filtering on the set of execution unit running state feature filtering coding vectors towards the execution unit running state pseudo-reference filtering center representation vector to obtain the execution unit running state key feature aggregation vector.

[0019] In an embodiment of the present application, the execution unit running state pseudo-benchmark filtering center representation vector is calculated based on the feature distribution density filtering of the set of execution unit running state feature filtering encoding vectors, comprising:

[0020] Each execution unit running state feature filtering encoding vector in the set of execution unit running state feature filtering encoding vectors is input into a feature distribution density measurement module to obtain a set of execution unit running state distribution density measurement coefficients;

[0021] The set of execution unit running state distribution density measurement coefficients is input into a feature density screening control unit to obtain a set of execution unit running state distribution density weight factors;

[0022] The weighted sum of the set of execution unit running state feature filtering encoding vectors is calculated based on the set of execution unit running state distribution density weight factors to obtain the execution unit running state pseudo-benchmark filtering center representation vector.

[0023] In an embodiment of the present application, each execution unit running state feature filtering encoding vector in the set of execution unit running state feature filtering encoding vectors is input into a feature distribution density measurement module to obtain a set of execution unit running state distribution density measurement coefficients, comprising:

[0024] The execution unit running state feature filtering encoding vector is normalized to obtain a normalized execution unit running state feature filtering encoding vector;

[0025] The sum of squares of each feature value in the normalized execution unit running state feature filtering encoding vector is divided by the feature dimension value of the normalized execution unit running state feature filtering encoding vector to obtain the execution unit running state distribution density measurement coefficient.

[0026] In an embodiment of the present application, the control parameter-running state association matching encoding vector is obtained by performing historical data enhancement-based association matching analysis on the execution unit running state key feature aggregation vector and the control parameter low-dimensional feature encoding vector, comprising:

[0027] Based on historical data, the execution unit running state key feature aggregation vector and the control parameter low-dimensional feature encoding vector are subjected to feature association optimization based on a matching mechanism to obtain a historical data optimized control parameter-running state association matching feature matrix;

[0028] optimizing the execution unit running state key feature aggregation vector and the control parameter low-dimensional feature coding vector based on the history data optimization control parameter-running state association matching feature matrix to obtain an optimized execution unit running state key feature aggregation vector and an optimized control parameter low-dimensional feature coding vector;

[0029] calculating the bitwise product between the optimized execution unit running state key feature aggregation vector and the optimized control parameter low-dimensional feature coding vector to obtain the control parameter-running state association matching coding vector.

[0030] In an embodiment of the present application, based on the history data, the execution unit running state key feature aggregation vector and the control parameter low-dimensional feature coding vector are optimized in feature association based on a matching mechanism to obtain a history data optimization control parameter-running state association matching feature matrix, including:

[0031] inputting the execution unit running state key feature aggregation vector and the control parameter low-dimensional feature coding vector into a fine-grained association analysis network to obtain a control parameter-running state association matching feature matrix;

[0032] inputting the control parameter-running state association matching feature matrix into a matching unit based on history data to obtain the history data optimization control parameter-running state association matching feature matrix.

[0033] In an embodiment of the present application, based on the history data optimization control parameter-running state association matching feature matrix, the execution unit running state key feature aggregation vector and the control parameter low-dimensional feature coding vector are respectively optimized in feature adjustment to obtain an optimized execution unit running state key feature aggregation vector and an optimized control parameter low-dimensional feature coding vector, including:

[0034] performing linear conversion on the execution unit running state key feature aggregation vector to obtain a first association feature vector and a first adjustment feature vector, and inputting the first association feature vector, the first adjustment feature vector and a reference matrix, which is the history data optimization control parameter-running state association matching feature matrix, into a fine-grained adjustment module based on a memory network to obtain the optimized execution unit running state key feature aggregation vector;

[0035] The low-dimensional feature encoding vector of the control parameters is linearly transformed to obtain a second associated feature vector and a second adjusted feature vector. The historical data is used to optimize the control parameter-running state association matching feature matrix as a reference matrix. The second associated feature vector, the second adjusted feature vector, and the reference matrix are input into the fine-grained adjustment module based on the memory network to obtain the optimized low-dimensional feature encoding vector of the control parameters.

[0036] The present invention also includes an environmental protection equipment control system for implementing the above-described environmental protection equipment control method, the system comprising:

[0037] The control parameter setting module is used to set control parameters, including operating temperature threshold, flow rate adjustment range, and pressure safety limit.

[0038] The operation status data acquisition module is used to collect the operation status data of each execution unit of the controlled environmental protection equipment under the constraints of the control parameters to obtain a set of execution unit operation status data;

[0039] The control parameter encoding module is used to perform feature extraction and encoding on the control parameters to obtain a low-dimensional feature encoding vector of the control parameters;

[0040] The feature hierarchical filtering processing module is used to perform hierarchical filtering processing based on feature weights on the set of execution unit running status data to obtain the key feature aggregation vector of the execution unit running status.

[0041] The association matching analysis processing module is used to perform association matching analysis based on historical data enhancement on the aggregated vector of key features of the execution unit's running state and the low-dimensional feature encoding vector of the control parameters to obtain the control parameter-running state association matching encoding vector.

[0042] The equipment control instruction generation module is used to generate environmental protection equipment control instructions based on the control parameter-operating status association matching encoding vector.

[0043] The present invention also includes a device comprising a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above-described environmental protection equipment control method.

[0044] The present invention also includes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the environmental protection equipment control method described above.

[0045] The beneficial effects of the environmental protection equipment control method, system, equipment, and medium provided by this invention are as follows:

[0046] The environmental protection equipment control method provided by this invention, by setting control parameters including operating temperature thresholds, flow regulation ranges, and pressure safety limits, can provide clear constraints for the operation of environmental protection equipment, ensuring that the equipment operates within a reasonable range and guaranteeing its safety and stability. The method collects and processes the operating status data of each execution unit of the controlled environmental protection equipment. By extracting and encoding features from the control parameters to obtain low-dimensional feature encoding vectors, and performing hierarchical filtering based on feature weights on the operating status data, key features can be effectively extracted, redundant information removed, and the efficiency and accuracy of data processing improved, enabling subsequent analysis to be based on more valuable information.

[0047] A correlation matching analysis based on historical data enhancement is performed on the aggregated vector of key features of the execution unit's operating status and the low-dimensional feature encoding vector of control parameters. This fully leverages the experience gained from historical data to deeply explore the correlation between control parameters and operating status, making the correlation matching analysis more accurate. This results in a control parameter-operating status correlation matching encoding vector that better reflects the actual operating situation. Based on this encoding vector, control commands for environmental protection equipment are generated. Through the control command generation module of the decision-maker, multiple factors are comprehensively considered to generate more targeted and accurate control commands, guiding the environmental protection equipment to perform precise operational adjustments.

[0048] This control method can adjust control parameters in real time based on the actual operating status and historical experience of the equipment, enabling environmental protection equipment to maintain efficient operation under different working conditions and external environments, and improving the equipment's adaptability to complex operating conditions. Precise control commands can effectively improve pollutant treatment efficiency while reducing energy consumption and increasing equipment operating efficiency, achieving the dual goals of environmental protection and energy conservation. Furthermore, through effective data processing and analysis, this method improves the system's intelligence level, reduces manual intervention, lowers operating costs and maintenance difficulty, and provides strong support for the automated and intelligent control of environmental protection equipment. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram illustrating the working principle of the environmental protection equipment control method described in this invention.

[0051] Figure 2 This is a flowchart of the hierarchical filtering process.

[0052] Figure 3Sub-flowcharts optimized for fine-grained tuning;

[0053] Figure 4 This is a block diagram of an environmental protection equipment control system. Detailed Implementation

[0054] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0056] Please see Figures 1-4 This invention provides a method, system, equipment, and medium for controlling environmental protection equipment, and the specific implementation steps are as follows:

[0057] Set control parameters, including operating temperature threshold, flow rate adjustment range, and pressure safety limit.

[0058] Under the constraints of the control parameters, the operating status data of each execution unit of the controlled environmental protection equipment are collected to obtain a set of execution unit operating status data.

[0059] The control parameters are feature extracted and encoded to obtain low-dimensional feature encoding vectors. Specifically, parameters such as operating temperature threshold, flow rate adjustment range, and pressure safety limit can be dimensionality reduced using a preset encoding model to remove redundant information, retain key feature dimensions, and form low-dimensional feature encoding vectors.

[0060] The set of execution unit running status data is subjected to hierarchical filtering based on feature weights to obtain the aggregated vector of key features of the execution unit running status.

[0061] A correlation matching analysis based on historical data enhancement is performed on the aggregated vector of key features of the execution unit's operating status and the low-dimensional feature encoding vector of the control parameters to obtain the correlation matching encoding vector of the control parameters and the operating status.

[0062] Based on the control parameter-operating status association matching encoding vector, environmental protection equipment control instructions are generated. Specifically, the control parameter-operating status association matching encoding vector is input into the decision-maker-based control instruction generation module to obtain equipment control instructions, which are used to instruct the operation adjustment of environmental protection equipment.

[0063] Example 1: When performing hierarchical filtering based on feature weights on a set of execution unit running status data to obtain aggregated vectors of key features of execution unit running status, the specific implementation method is as follows: The running status data of each execution unit in the set of execution unit running status data is passed through a filtering layer. This filtering layer can be a pre-set algorithm module with specific filtering rules, whose function is to perform preliminary feature screening and encoding processing on each execution unit running status data. In this process, each execution unit running status data undergoes the operation of the filtering layer, removing some redundant or unimportant information while retaining representative features, thereby obtaining a set of filtered and encoded vectors of execution unit running status features.

[0064] Based on the feature distribution density of the set of execution unit running state feature filtering encoded vectors, it is necessary to calculate the pseudo-benchmark filtering center representation vector of the execution unit running state. This process requires first inputting each execution unit running state feature filtering encoded vector in the set of execution unit running state feature filtering encoded vectors into the feature distribution density measurement module. In the feature distribution density measurement module, each execution unit running state feature filtering encoded vector is first normalized. The purpose of normalization is to transform feature values ​​at different scales to the same scale to facilitate subsequent calculations and comparisons. Specific normalization methods can use common normalization algorithms, such as dividing each element in the vector by the magnitude of the vector to obtain the normalized execution unit running state feature filtering encoded vector.

[0065] After obtaining the normalized execution unit running state feature filtering encoding vector, the sum of squares of each feature value in the vector is calculated. This sum is then divided by the feature dimension value of the normalized execution unit running state feature filtering encoding vector, thus obtaining the execution unit running state distribution density metric coefficient. In this way, each execution unit running state feature filtering encoding vector can yield a corresponding distribution density metric coefficient, and these coefficients constitute a set of execution unit running state distribution density metric coefficients.

[0066] The set of execution unit running state distribution density metric coefficients is input into the feature density filtering control unit. The feature density filtering control unit has specific filtering rules and weight calculation logic preset. It processes the input distribution density metric coefficients and assigns a corresponding execution unit running state distribution density weight factor to each execution unit running state feature filtering encoding vector according to the magnitude and distribution of each coefficient, thereby obtaining the set of execution unit running state distribution density weight factors.

[0067] After obtaining the set of weight factors for the distribution density of execution unit running states, a weighted sum of the set of execution unit running state feature filtering encoding vectors is calculated based on these weight factors. Each execution unit running state feature filtering encoding vector is multiplied by its corresponding weight factor, and then all multiplications are summed to finally obtain the pseudo-benchmark filtering center representation vector of the execution unit running states. This vector reflects, to some extent, the central tendency of the set of execution unit running state feature filtering encoding vectors.

[0068] The set of execution unit running state filtering adjustment directions is obtained by calculating the filtering adjustment direction of each execution unit running state feature filtering encoding vector in the set of execution unit running state feature filtering encoding vectors relative to the execution unit running state pseudo-reference filtering center representation vector. The filtering adjustment direction can be determined by calculating the difference between each feature filtering encoding vector and the pseudo-reference filtering center representation vector. For example, the filtering adjustment direction can be obtained by subtracting the pseudo-reference filtering center representation vector from the feature filtering encoding vector.

[0069] Based on the set of filtering adjustment directions for the execution unit's running state, the set of feature filtering encoding vectors for the execution unit's running state is subjected to hierarchical filtering towards the pseudo-reference filtering center representation vector of the execution unit's running state. The hierarchical filtering process can be divided into multiple layers. Each layer adjusts and filters the feature filtering encoding vectors according to the filtering adjustment direction, gradually removing unimportant features that deviate from the center and retaining key features close to the center. After multi-layer filtering, the final aggregated vector of key features of the execution unit's running state is obtained. This vector contains the most critical feature information in the execution unit's running state data, providing an important foundation for subsequent correlation matching analysis and control command generation. Throughout the entire process, each step strictly follows the preset algorithm and logic to ensure the accuracy and consistency of data processing.

[0070] Example 2: When performing historical data-enhanced association matching analysis on the aggregated vector of key features of the execution unit's running state and the low-dimensional feature encoding vector of control parameters to obtain the control parameter-running state association matching encoding vector, the specific implementation method is as follows. Based on historical data, feature association optimization based on a matching mechanism is performed on the aggregated vector of key features of the execution unit's running state and the low-dimensional feature encoding vector of control parameters to obtain a historical data-optimized control parameter-running state association matching feature matrix. Here, the aggregated vector of key features of the execution unit's running state and the low-dimensional feature encoding vector of control parameters are first input into a fine-grained association analysis network. This network can adopt a deep learning architecture, such as a multilayer perceptron or a convolutional neural network, whose internal nodes learn the association pattern between the two features through training data. The network receives two vectors as input, and after calculation and feature transformation by multiple layers of neurons, outputs the control parameter-running state association matching feature matrix. Each element of this matrix represents the association strength of the corresponding feature dimension of the two vectors, and fine-grained association analysis between features is realized through the network's operation.

[0071] The control parameter-operating status correlation matching feature matrix is ​​input into a matching unit based on historical data. This matching unit stores multiple sets of control parameter vectors, operating status vectors, and corresponding correlation matching matrices accumulated during historical operation. This historical data contains operating information of the equipment under different operating conditions and is a summary of long-term operating experience. The matching unit compares the currently input correlation matching feature matrix with the historically stored matrices. Specifically, it identifies the differences between the current and historical matrices by calculating similarity indices such as cosine similarity or Euclidean distance. Based on the difference analysis results, the matching unit adjusts the current matrix, for example, by correcting certain correlation strength values ​​and supplementing well-performing correlation patterns from historical data, thereby obtaining a historically optimized control parameter-operating status correlation matching feature matrix. This process is equivalent to integrating historical experience into the current correlation analysis, making the generated matrix more consistent with the actual operating patterns of the equipment.

[0072] After obtaining the historical data-optimized control parameter-operating state correlation matching feature matrix, it is necessary to perform feature adjustment and optimization on the execution unit operating state key feature aggregation vector and the control parameter low-dimensional feature encoding vector, respectively, to obtain the optimized execution unit operating state key feature aggregation vector and the optimized control parameter low-dimensional feature encoding vector. For the execution unit operating state key feature aggregation vector, a linear transformation is first performed. This linear transformation is achieved through a preset transformation matrix, constructed based on the structure of the historical data-optimized feature matrix, aiming to decompose the original vector into two parts: a first correlation feature vector and a first adjusted feature vector. The first correlation feature vector contains features with a high correlation to the control parameters, while the first adjusted feature vector contains features that require further optimization.

[0073] Using the historical data-optimized control parameter-running state correlation matching feature matrix as a reference matrix, the first correlation feature vector, the first adjustment feature vector, and the reference matrix are input together into a fine-grained adjustment module based on a memory network. The memory network stores successful historical feature adjustments, with each case recording the adjustment method and result under a specific feature combination. The fine-grained adjustment module uses an attention mechanism to retrieve the most relevant historical cases from the memory network based on the strength of feature correlations in the reference matrix. When processing the first correlation feature vector and the first adjustment feature vector, the module reallocates the weights of different features based on experience from historical cases, correcting feature values—for example, increasing the weights of certain key correlation features and reducing the influence of irrelevant features—thus obtaining an optimized aggregated vector of key features for the execution unit's running state.

[0074] Feature adjustment and optimization are performed on the low-dimensional feature encoding vector of the control parameters. First, a linear transformation is performed to obtain a second associated feature vector and a second adjusted feature vector. The linear transformation process is similar to the transformation of the aggregated vector of key features of the execution unit's running state; it also decomposes the original vector into parts highly correlated with the running state features and parts requiring adjustment using a preset matrix. Then, using the same historical data to optimize the control parameter-running state correlation matching feature matrix as a reference matrix, the second associated feature vector, the second adjusted feature vector, and the reference matrix are input into a fine-grained adjustment module based on a memory network. The module performs fine-grained adjustment of the control parameter features based on historical cases in the same manner, obtaining the optimized low-dimensional feature encoding vector of the control parameters.

[0075] The optimized execution unit operating state key feature aggregation vector and the optimized control parameter low-dimensional feature encoding vector are multiplied bitwise to obtain the control parameter-operating state correlation matching encoding vector. The bitwise multiplication rule involves multiplying corresponding elements of the two vectors to obtain new vector elements. Through this operation, the feature correlation information of the two vectors is integrated into the new vector. Each element contains information about both operating state features and control parameter features, and the value of the element reflects the correlation strength between the two in that feature dimension. The resulting control parameter-operating state correlation matching encoding vector integrates the results of historical data-enhanced correlation analysis and feature adjustment optimization, more accurately reflecting the intrinsic relationship between control parameters and operating state, providing a reliable basis for subsequent generation of environmental protection equipment control commands. Throughout the implementation process, the computational logic and data processing flow of each module are strictly based on preset algorithms and historical data, ensuring that each operation has clear logical support and data foundation, thereby guaranteeing the accuracy and effectiveness of the correlation matching analysis.

[0076] Example 3: When calculating the pseudo-benchmark filtering center representation vector of the execution unit's running state, it involves inputting each vector in the set of execution unit running state feature filtering encoding vectors into the feature distribution density measurement module to obtain the set of execution unit running state distribution density measurement coefficients. The specific implementation method is as follows: First, the execution unit running state feature filtering encoding vectors need to be normalized to obtain normalized execution unit running state feature filtering encoding vectors. Normalization is to eliminate the differences in scale between different feature vectors, so that subsequent calculations can be performed under a unified standard. Specifically, for each execution unit running state feature filtering encoding vector, the sum of squares of its various feature values ​​is calculated, and then the square root of the sum of squares is taken to obtain the magnitude of the vector. Then, each feature value in the vector is divided by the magnitude to obtain the normalized vector. After this processing, the magnitude of each normalized execution unit running state feature filtering encoding vector is 1, which facilitates the subsequent distribution density calculation.

[0077] After obtaining the normalized execution unit running state feature filtering encoding vector, it is necessary to calculate the execution unit running state distribution density metric coefficient of this vector. Specifically, the calculation method is as follows: first, calculate the sum of squares of each feature value in the normalized vector. This sum of squares reflects the degree of concentration of the feature values ​​in the vector space. Then, divide this sum of squares by the feature dimension value of the normalized execution unit running state feature filtering encoding vector. The feature dimension value refers to the number of features contained in the vector; for example, a vector containing n features has a feature dimension value of n. By dividing the sum of squares by the feature dimension value, the degree of concentration can be converted into a relative metric, making the distribution density comparable between vectors of different dimensions. In this way, each normalized feature filtering encoding vector can obtain a corresponding execution unit running state distribution density metric coefficient, and all such coefficients constitute the set of execution unit running state distribution density metric coefficients.

[0078] Next, the set of execution unit running state distribution density metric coefficients is input into the feature density filtering control unit. The feature density filtering control unit has pre-set specific filtering rules and weight allocation logic. This unit analyzes the input distribution density metric coefficients, determining the feature distribution density of the corresponding feature filtering encoding vector based on the magnitude of each coefficient. Generally, vectors with larger distribution density metric coefficients have more concentrated feature values ​​in the vector space, potentially containing more critical running state information; while vectors with smaller coefficients have relatively dispersed feature distributions, potentially containing more redundant information. Based on these coefficients, the feature density filtering control unit assigns an execution unit running state distribution density weight factor to each execution unit running state feature filtering encoding vector. The magnitude of the weight factor is positively correlated with the distribution density metric coefficient; that is, the higher the distribution density, the larger the corresponding weight factor, and vice versa. In this way, a set of execution unit running state distribution density weight factors is obtained, with each weight factor corresponding to a feature filtering encoding vector, used to represent the importance of that vector in subsequent calculations.

[0079] After obtaining the set of weighting factors for the distribution density of the execution unit's running state, a weighted sum of the set of feature filtering encoded vectors for the execution unit's running state is calculated based on these weighting factors to obtain the pseudo-benchmark filtering center representation vector of the execution unit's running state. Specifically, for each feature filtering encoded vector of the execution unit's running state, it is multiplied by its corresponding distribution density weighting factor to obtain a weighted vector. Then, all weighted vectors are summed to obtain the final weighted sum vector, which is the pseudo-benchmark filtering center representation vector of the execution unit's running state. This vector comprehensively considers the feature distribution density of all feature filtering encoded vectors; vectors with higher distribution densities have a larger weight in the weighted sum. Therefore, the pseudo-benchmark filtering center representation vector can better reflect the central trend of the entire set of feature filtering encoded vectors.

[0080] In this process, each step strictly adheres to pre-defined algorithms and logic. Normalization ensures the consistency of vector scale, the calculation of the distribution density metric quantifies the concentration of feature distribution, the weight allocation of the feature density filtering control unit reflects the differences in importance among different vectors, and the weighted sum calculation integrates these factors to obtain a pseudo-benchmark filtering center representation vector that represents the center of the overall feature distribution. This calculation method can effectively extract a representative center vector from the feature filtering encoding vector set, providing a reliable benchmark for subsequent calculations of the filtering adjustment direction of each feature filtering encoding vector relative to this center. Throughout the implementation, the data processing flow is clear, and each step has a clear purpose and basis, ensuring the accuracy and effectiveness of the calculation results. This lays the foundation for subsequent hierarchical filtering processing, enabling more accurate extraction of key feature information from the execution unit's operational status data.

[0081] Example 4: When performing correlation matching analysis on the aggregated vector of key features of the operating status of the execution unit and the low-dimensional feature encoding vector of the control parameters, feature correlation optimization based on historical data is achieved through a fine-grained correlation analysis network and a matching unit. The specific implementation method is as follows. Taking an industrial wastewater treatment environmental protection equipment as an example, the equipment includes multiple execution units, such as a filtration unit, an aeration unit, and a sedimentation unit. The operating status data of each unit includes parameters such as flow rate, pressure, and temperature, while the control parameters are set with operating temperature thresholds, flow rate adjustment ranges, and pressure safety limits.

[0082] The aggregated vector of key features of the execution unit's operating status and the low-dimensional feature encoding vector of control parameters are input into a fine-grained correlation analysis network. Assume the aggregated vector of key features of the execution unit's operating status includes real-time flow characteristics of the filtration unit and pressure characteristics of the aeration unit, while the low-dimensional feature encoding vector of control parameters includes set flow regulation range characteristics and temperature threshold characteristics. The fine-grained correlation analysis network can employ a multi-layer neural network structure. Its input layer receives these two vectors, and the hidden layers transform and perform correlation analysis on the features through the weighted connections of neurons. For example, the first hidden layer of the network might perform weighted calculations on the flow characteristics and flow regulation range characteristics, while the second hidden layer might combine pressure characteristics and temperature threshold characteristics for nonlinear transformation. After multiple layers of calculation, the output is a control parameter-operating status correlation matching feature matrix. Each element in this matrix represents the degree of correlation between the corresponding control parameter feature and the operating status feature. For example, the correlation strength between a certain element in the matrix and the flow characteristics of the filtration unit reflects the degree of correlation between the two.

[0083] The control parameter-operating status correlation matching feature matrix is ​​input into the matching unit based on historical data. The matching unit stores the operating data of the wastewater treatment equipment for the past year, including control parameter vectors, operating status vectors, and corresponding correlation matching matrices for different seasons and wastewater concentrations. For example, historical data records the correlation matching between the set temperature threshold and the aeration unit temperature characteristics during high-temperature summers with high wastewater concentrations, as well as the corresponding treatment effects. The matching unit compares the currently generated correlation matching feature matrix with the historically stored matrices, specifically identifying differences by calculating the similarity between the matrices. For instance, if the correlation strength value between the temperature threshold feature and the aeration unit temperature feature in the current matrix is ​​0.6, while in historical data, this correlation strength value under similar operating conditions is typically 0.8, the matching unit identifies this difference and adjusts the value of this element in the current matrix to 0.75 based on historical data. Simultaneously, it adjusts other differing elements accordingly, thus obtaining the historical data-optimized control parameter-operating status correlation matching feature matrix.

[0084] In this process, the fine-grained association analysis network is trained based on a large amount of data collected during normal equipment operation. Through supervised learning, using historical association matching matrices as labels, the network's weight parameters are adjusted to enable it to accurately learn the association patterns between control parameters and operating state characteristics. The historical data of the matching units comes from the accumulation of long-term equipment operation, covering operating information under various working conditions, ensuring the rationality and reliability of the current matrix adjustment.

[0085] Taking another scenario as an example, when the quality of the wastewater being treated by the equipment changes, such as the sudden introduction of oily wastewater, the pressure feature of the filter unit in the key feature aggregation vector of the execution unit's operating status will change significantly, and the flow regulation range feature in the low-dimensional feature encoding vector of the control parameters may need to be adjusted accordingly. In this case, the fine-grained correlation analysis network will generate a new correlation matching feature matrix based on the new input vector. The correlation strength between the filter unit pressure feature and the flow regulation range feature may differ from that in historical data under normal operating conditions. When comparing historical data, the matching unit will retrieve historical correlation matching matrices from similar water quality changes, identify the differences in the correlation strength in the current matrix, and optimize it based on historical experience to make the adjusted matrix more consistent with actual treatment needs.

[0086] Through the synergistic effect of a fine-grained correlation analysis network and a matching unit, feature correlation optimization based on historical data is achieved. The fine-grained correlation analysis network extracts the correlation relationships between features from the current input vector to generate an initial correlation matching matrix. The matching unit, on the other hand, uses experience accumulated from historical data to correct and optimize the initial matrix, compensating for any deficiencies in the current data. This ensures that the final historical data-optimized control parameter-operating state correlation matching feature matrix more accurately reflects the actual correlation between control parameters and operating states, providing a more reliable basis for subsequent feature adjustment optimization and control command generation.

[0087] Example 5: When optimizing the control parameters and operating status correlation feature matrix based on historical data for feature adjustment, a chemical waste gas treatment environmental protection equipment is used as an example to illustrate the implementation method. The equipment's execution units include a catalytic combustion unit and an adsorption unit. The control parameters are set as follows: catalytic temperature threshold of 280-320℃, adsorption flow rate adjustment range of 1500-2000 m³ / h, and pressure safety limit of 80 kPa. Assuming that during a certain operating period, the key feature aggregation vector of the execution unit's operating status includes the real-time temperature feature of the catalytic combustion unit (295℃) and the real-time flow rate feature of the adsorption unit (1800 m³ / h). The low-dimensional feature encoding vector of the control parameters includes the catalytic temperature threshold feature (mapped to the 280-320℃ range as the feature vector) and the flow rate adjustment range feature (mapped to the 1500-2000 m³ / h range as the feature vector).

[0088] A linear transformation is performed on the aggregated vector of key features of the execution unit's operating status. This linear transformation is achieved through a pre-defined transformation matrix, constructed based on the structure of a historical data-optimized control parameter-operating status correlation matching feature matrix. For example, the historical data-optimized feature matrix shows a high correlation between catalytic temperature features and catalytic temperature threshold features, followed by a lower correlation between adsorption flow rate features and flow rate adjustment range features. Based on this, the linear transformation decomposes the aggregated vector of key features of the execution unit's operating status into a first correlation feature vector and a first adjustment feature vector. The first correlation feature vector mainly contains features highly correlated with control parameters, such as the dimension corresponding to the real-time catalytic temperature feature (295℃), while the first adjustment feature vector contains features requiring further optimization, such as the dimension corresponding to the real-time adsorption flow rate feature (1800 m³ / h).

[0089] Using the historical data-optimized control parameter-operating status correlation matching feature matrix as a reference matrix, the first correlation feature vector, the first adjustment feature vector, and the reference matrix are input together into a fine-grained adjustment module based on a memory network. The memory network stores operating cases of the waste gas treatment equipment from the past two years. For example, when the real-time catalytic temperature feature is 290℃, historical adjustment experience indicates that the weight of the temperature threshold feature should be increased by 0.15 to ensure catalytic efficiency. When the real-time adsorption flow rate feature is 1900 m³ / h, historical cases show that flow rate fluctuations led to a decrease in adsorption efficiency; the corresponding adjustment method is to correct certain dimensions of the flow rate adjustment range feature. The fine-grained adjustment module, through an attention mechanism, retrieves the historical cases most relevant to the current catalytic temperature feature from the memory network based on the correlation strength between the catalytic temperature feature and the temperature threshold feature in the reference matrix (assuming the element value at that position in the reference matrix is ​​0.78), such as adjustment experience when the temperature is in the 290-300℃ range. Based on this historical case, the module adjusts the weight of the catalytic temperature feature dimension in the first associated feature vector, changing its weight from the initial 0.6 to 0.7. At the same time, it corrects the adsorption flow rate feature dimension in the first adjusted feature vector, referring to the adjustment method when the flow rate was 1900 m³ / h in history, and fine-tunes the feature value of this dimension, for example, adjusting its mapping value from 0.8 to 0.85, thereby obtaining the optimized aggregated vector of key features of the execution unit's operating status.

[0090] Feature adjustment and optimization are performed on the low-dimensional feature encoding vector of the control parameters. First, a linear transformation is performed to obtain a second correlated feature vector and a second adjusted feature vector. The linear transformation optimizes the feature matrix based on the same historical data, decomposing the low-dimensional feature encoding vector of the control parameters into parts highly correlated with the operating state features and parts requiring adjustment. For example, the second correlated feature vector contains dimensions in the catalytic temperature threshold feature that are highly correlated with the catalytic combustion unit temperature feature, while the second adjusted feature vector contains dimensions in the flow regulation range feature that require optimization in relation to the adsorption unit flow rate feature.

[0091] Using the same historical data to optimize the control parameter-operating status correlation matching feature matrix as a reference matrix, the second correlation feature vector, the second adjustment feature vector, and the reference matrix are input into a fine-grained adjustment module based on a memory network. The module retrieves historical adjustment cases from the memory network where the flow regulation range is 1800-2000 m³ / h, based on the correlation strength between the flow regulation range feature and the adsorption flow feature in the reference matrix (assuming an element value of 0.65). For example, when the correlation strength between the flow regulation range feature and the real-time flow feature is 0.6, the historical adjustment increases the weight of a certain sub-dimension of the flow regulation range feature by 0.1. Following this case, the module adjusts the weights of the relevant dimensions of the flow regulation range feature in the second correlation feature vector, and simultaneously corrects other dimensions in the second adjustment feature vector, obtaining the optimized low-dimensional feature encoding vector of the control parameters.

[0092] The optimized execution unit operating state key feature aggregation vector and the optimized control parameter low-dimensional feature encoding vector are multiplied bitwise to obtain the control parameter-operating state association matching encoding vector. For example, in the optimized operating state vector, the catalytic temperature feature dimension is 0.7, and in the optimized control parameter vector, the temperature threshold feature corresponding to the catalytic temperature feature dimension is 0.8; after bitwise multiplication, this dimension value is 0.56. Similarly, in the operating state vector, the adsorption flow rate feature dimension is 0.85, and in the control parameter vector, the flow rate adjustment range feature corresponding to the adsorption flow rate feature dimension is 0.7; after bitwise multiplication, this dimension value is 0.595. Through this operation, the value of each dimension of the generated new vector incorporates the association information between the operating state features and the control parameter features, reflecting the degree of matching between the two in that dimension.

[0093] In another scenario, if the equipment encounters a sudden operating condition, such as the catalytic combustion unit temperature suddenly rising to 310℃ due to changes in exhaust gas composition, the temperature feature dimension value in the key feature aggregation vector of the execution unit's operating status will change significantly. In this case, the linear transformation will re-decompose the vector, and the fine-grained adjustment module will retrieve historical adjustment cases under high-temperature conditions from the memory network. For example, if the weight of a certain dimension of the temperature threshold feature was reduced by 0.1 at 310℃ to trigger an early warning mechanism, the module will adjust the features based on this case, making the optimized vector more closely match the current operating condition. The correlation matching encoding vector generated by bitwise multiplication accurately reflects the correlation between control parameters and operating status under high-temperature conditions, providing a basis for subsequently generating control commands to adjust the catalytic temperature.

[0094] Throughout the implementation process, the matrix parameters for linear transformation are derived from historical data statistics of long-term equipment operation, and the case studies for the memory network are derived from equipment maintenance records and operation logs, ensuring that each adjustment step is supported by actual data. The attention mechanism of the fine-grained adjustment module can dynamically match the most relevant historical experience, avoiding the limitations of subjective parameter settings and making feature adjustments more consistent with the actual operating patterns of the equipment.

[0095] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for controlling environmental protection equipment, characterized in that, include: Set control parameters, including operating temperature threshold, flow rate adjustment range, and pressure safety limit; Under the constraints of the control parameters, the operating status data of each execution unit of the controlled environmental protection equipment are collected to obtain a set of execution unit operating status data; The control parameters are feature extracted and encoded to obtain a low-dimensional feature encoding vector of the control parameters; The set of execution unit running status data is subjected to hierarchical filtering based on feature weights to obtain an aggregated vector of key features of the execution unit running status; A correlation matching analysis based on historical data enhancement is performed on the aggregated vector of key features of the execution unit's operating status and the low-dimensional feature encoding vector of the control parameters to obtain the control parameter-operating status correlation matching encoding vector; Based on the control parameter-operation status association matching encoding vector, environmental protection equipment control instructions are generated. Specifically, the control parameter-operation status association matching encoding vector is input into the decision-maker-based control instruction generation module to obtain equipment control instructions, which are used to instruct the operation adjustment of environmental protection equipment. The set of execution unit running status data is subjected to hierarchical filtering based on feature weights to obtain an aggregated vector of key features of the execution unit running status, including: Each execution unit's running status data in the set of execution unit running status data is passed through a filtering layer to obtain a set of execution unit running status feature filtering encoding vectors; Based on the feature distribution density of the set of filter encoding vectors for the running state features of the execution unit, calculate the pseudo-reference filter center representation vector for the running state of the execution unit; The filtering adjustment direction of each execution unit running state feature filtering encoding vector in the set of execution unit running state feature filtering encoding vectors is calculated relative to the execution unit running state pseudo-reference filtering center representation vector to obtain the set of execution unit running state filtering adjustment directions; Based on the set of filtering adjustment directions for the execution unit's running state, the set of filtering encoding vectors for the execution unit's running state features is filtered hierarchically toward the pseudo-reference filtering center representation vector of the execution unit's running state to obtain the aggregated vector of the execution unit's running state key features.

2. The environmental protection equipment control method according to claim 1, characterized in that, Based on the feature distribution density of the set of filter encoding vectors for the execution unit's running state features, the pseudo-reference filter center representation vector for the execution unit's running state is calculated, including: Each execution unit running state feature filtering encoding vector in the set of execution unit running state feature filtering encoding vectors is input into the feature distribution density measurement module to obtain a set of execution unit running state distribution density measurement coefficients; The set of the execution unit running state distribution density metric coefficients is input into the feature density filtering control unit to obtain the set of execution unit running state distribution density weight factors; Based on the set of weight factors for the distribution density of the execution unit's running state, a weighted sum of the set of feature filtering encoding vectors for the execution unit's running state is calculated to obtain the pseudo-benchmark filtering center representation vector for the execution unit's running state.

3. The environmental protection equipment control method according to claim 2, characterized in that, Each execution unit running state feature filtering encoding vector in the set of execution unit running state feature filtering encoding vectors is input into the feature distribution density measurement module to obtain a set of execution unit running state distribution density measurement coefficients, including: The execution unit running state feature filtering encoding vector is normalized to obtain a normalized execution unit running state feature filtering encoding vector; The sum of the squares of each feature value in the normalized execution unit running state feature filtering encoding vector is calculated and divided by the feature dimension value of the normalized execution unit running state feature filtering encoding vector to obtain the execution unit running state distribution density metric coefficient.

4. The environmental protection equipment control method according to claim 3, characterized in that, A correlation matching analysis based on historical data enhancement is performed on the aggregated vector of key features of the execution unit's operating state and the low-dimensional feature encoding vector of the control parameters to obtain a control parameter-operating state correlation matching encoding vector, including: Based on historical data, feature association optimization based on a matching mechanism is performed on the aggregated vector of key features of the execution unit's running status and the low-dimensional feature encoding vector of the control parameters to obtain a historical data-optimized control parameter-running status association matching feature matrix. Based on the historical data, optimize the control parameter-operating state association matching feature matrix, and perform feature adjustment and optimization on the execution unit operating state key feature aggregation vector and the control parameter low-dimensional feature encoding vector to obtain the optimized execution unit operating state key feature aggregation vector and the optimized control parameter low-dimensional feature encoding vector. The bitwise product between the optimized execution unit running state key feature aggregation vector and the optimized control parameter low-dimensional feature encoding vector is calculated to obtain the control parameter-running state association matching encoding vector.

5. The environmental protection equipment control method according to claim 4, characterized in that, Based on historical data, feature association optimization based on a matching mechanism is performed on the aggregated vector of key features of the execution unit's operating state and the low-dimensional feature encoding vector of the control parameters to obtain a historical data-optimized control parameter-operating state association matching feature matrix, including: The aggregated vector of key features of the execution unit's running state and the low-dimensional feature encoding vector of the control parameters are input into a fine-grained correlation analysis network to obtain a control parameter-running state correlation matching feature matrix; The control parameter-operating state association matching feature matrix is ​​input into the historical data-based matching unit to obtain the historical data-optimized control parameter-operating state association matching feature matrix.

6. The environmental protection equipment control method according to claim 5, characterized in that, Based on the historical data, the control parameter-operating state correlation matching feature matrix is ​​optimized. Feature adjustment and optimization are then performed on the aggregated vector of key features of the execution unit's operating state and the low-dimensional feature encoding vector of the control parameters to obtain the optimized aggregated vector of key features of the execution unit's operating state and the optimized low-dimensional feature encoding vector of the control parameters, including: The execution unit's key feature aggregation vector is linearly transformed to obtain a first associated feature vector and a first adjusted feature vector. The historical data optimization control parameter-running state association matching feature matrix is ​​used as a reference matrix. The first associated feature vector, the first adjusted feature vector, and the reference matrix are input into a fine-grained adjustment module based on a memory network to obtain the optimized execution unit's key feature aggregation vector. The low-dimensional feature encoding vector of the control parameters is linearly transformed to obtain a second associated feature vector and a second adjusted feature vector. The historical data is used to optimize the control parameter-running state association matching feature matrix as a reference matrix. The second associated feature vector, the second adjusted feature vector, and the reference matrix are input into the fine-grained adjustment module based on the memory network to obtain the optimized low-dimensional feature encoding vector of the control parameters.

7. An environmental protection equipment control system, used to implement the environmental protection equipment control method as described in any one of claims 1 to 6, characterized in that, include: The control parameter setting module is used to set control parameters, including operating temperature threshold, flow rate adjustment range, and pressure safety limit. The operation status data acquisition module is used to collect the operation status data of each execution unit of the controlled environmental protection equipment under the constraints of the control parameters to obtain a set of execution unit operation status data; The control parameter encoding module is used to perform feature extraction and encoding on the control parameters to obtain a low-dimensional feature encoding vector of the control parameters; The feature hierarchical filtering processing module is used to perform hierarchical filtering processing based on feature weights on the set of execution unit running status data to obtain the key feature aggregation vector of the execution unit running status. The association matching analysis processing module is used to perform association matching analysis based on historical data enhancement on the aggregated vector of key features of the execution unit's running state and the low-dimensional feature encoding vector of the control parameters to obtain the control parameter-running state association matching encoding vector. The equipment control instruction generation module is used to generate environmental protection equipment control instructions based on the control parameter-operating status association matching encoding vector.

8. A device comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the environmental protection equipment control method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of an environmental protection equipment control method as described in any one of claims 1 to 6.

Images