A unit configuration adjustment method and system based on multi-algorithm optimization

Abstract

The present invention discloses a method and system for adjusting unit configuration based on multi-algorithm optimization, belonging to the technical field of power system optimization. Specifically, it includes: acquiring and preprocessing unit operation data, classifying operating conditions, identifying the optimal operating parameters under each condition and conducting modeling analysis; by constructing a knowledge graph of unit operation, real-time matching of historical optimal condition parameters, and setting a joint optimization space in combination with constraint conditions; using a pre-trained machine learning model and a hybrid optimization algorithm to obtain the optimal parameter combination within the optimization space; adjusting the unit configuration, real-time monitoring the operating status, and optimizing the model parameters according to real-time feedback; the present invention improves the operating efficiency and stability of the unit, and realizes intelligent and adaptive adjustment of unit configuration.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system optimization, and specifically relates to a unit configuration adjustment method and system based on multi-algorithm optimization. Background Art

[0002] In industrial production, especially in the power system, unit configuration adjustment is a crucial link. With the continuous expansion of the scale and the increasing complexity of the power system, the importance of unit configuration adjustment has become increasingly prominent. Unit configuration adjustment methods include traditional optimization methods and intelligent optimization algorithms. Among them, traditional optimization methods mainly use traditional optimization methods such as nonlinear programming method, heuristic algorithm and priority sequence method to solve unit configuration adjustment problems. Although the problems are solved to a certain extent, there are problems such as large computational amount, slow convergence speed and easy to fall into local optimum; intelligent optimization algorithms include genetic algorithm, ant colony algorithm and particle swarm algorithm, which mainly find the optimal solution of the problem by simulating natural evolution or swarm intelligence behavior. However, although intelligent optimization algorithms have achieved remarkable results in unit configuration adjustment, a single algorithm often has difficulty in dealing with all complex situations, and there are problems such as slow convergence speed and easy to fall into local optimum.

[0003] For example, the patent with the authorization announcement number CN116632931A discloses a unit grid-connected control parameter configuration method and system, including: for local faults of the unit, configuring the unit grid-connected control parameters through simulation. If the effect after configuration meets the set conditions, further check the adaptability of the configured parameters to other key faults of the system, and adjust the unit grid-connected control parameters. If the effect after configuration does not meet the set conditions, re-simulate the local fault and re-configure the parameters. Therefore, this technical solution can optimize the configuration of unit parameters in multiple types of units, can consider the performance of the unit itself and its impact on the system performance, can also consider the adaptability of the unit grid-connected control parameters to multiple faults, and can consider the coordinated configuration of multiple unit control parameters and its impact on the overall performance of the system.

[0004] The above existing technologies all have the following problems: relying on the simulation environment to simulate the effects of unit faults and parameter configurations, resulting in a deviation between the simulation results and the actual situation; the need for manual judgment on whether the effect after configuration meets the set conditions and adjusting the parameters accordingly; lacking the ability of real-time monitoring and dynamic adjustment of the unit operation state. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention proposes a method and system for adjusting unit configuration based on multi-algorithm optimization, which acquires and preprocesses unit operation data, classifies operating conditions, identifies the optimal operating parameters under each operating condition and conducts modeling analysis; by constructing a knowledge graph of unit operation, it matches the real-time operating condition parameter values with the nodes in the knowledge graph in real time, combines with the constraint conditions to set a joint optimization space; uses a pre-trained machine learning model and a hybrid optimization algorithm to obtain the optimal parameter combination within the optimization space; adjusts the unit configuration, monitors the operating status in real time, and optimizes the model parameters according to the real-time feedback; the present invention improves the operating efficiency and stability of the unit and realizes intelligent and adaptive adjustment of unit configuration.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for adjusting unit configuration based on multi-algorithm optimization, comprising:

[0008] Step S1: Acquire unit operation data and perform preprocessing. Based on the preprocessed unit operation data, divide the unit operating status into different operating condition categories;

[0009] Step S2: In each operating condition category, screen out the optimal operating parameter combination, generate an optimal operating condition table for the unit operation, and conduct modeling analysis on the optimal operating parameters;

[0010] Step S3: Construct a knowledge graph of unit operation, and match the real-time operating condition parameter values with the nodes in the knowledge graph through a similarity measurement strategy to obtain the historical optimal operating condition parameters. At the same time, based on the historical optimal operating condition parameters, combine with the constraint conditions of unit operation to set a joint optimization space;

[0011] Step S4: Load the pre-trained machine learning model. At the same time, input the parameter combinations within the joint optimization space into the machine learning model, and obtain the optimal parameter combination through a hybrid optimization algorithm;

[0012] Step S5: Based on the optimal parameter combination, adjust the unit configuration, and monitor the operating status of the adjusted unit in real time. At the same time, according to the real-time feedback information, optimize the machine learning model parameters;

[0013] The specific steps of the said Step S3 include:

[0014] S3.1: Acquire environmental parameters and historical unit maintenance records, and generate a multi-source data set in combination with the optimal operating condition table and modeling results generated in Step S2;

[0015] S3.2: Define the entities, relationships and attributes of the knowledge graph, and use graph neural networks and graph embedding methods to automatically extract and construct the nodes and edges of the knowledge graph from the multi-source data set to form a dynamic knowledge graph of unit operation;

[0016] S3.3: Real-time collect the unit operating condition parameters, and through the fusion of fuzzy logic, combine with the deep learning algorithm to measure the similarity between the real-time unit operating condition parameters and the nodes in the knowledge graph, and comprehensively evaluate the similarity between the real-time operating condition and the historical optimal operating condition ;

[0017] S3.4: According to the similarity result, take the corresponding unit operating condition parameters as the historical optimal operating condition parameters, where denotes the maximum value of;

[0018] S3.5: Based on the historical optimal operating condition parameters, combined with the constraint conditions of the unit operation, construct a multi-objective optimization model;

[0019] S3.6: Use the multi-objective optimization algorithm to solve the multi-objective optimization model, generate a multi-dimensional parameter adjustment scheme, and at the same time, according to the real-time operating condition feedback, dynamically adjust the parameter fluctuation range and step size to form a joint optimization space;

[0020] The specific steps of the said S3.2 include:

[0021] S3.21: Define the entities, relationships and attributes of the knowledge graph, and obtain multi-source data sets;

[0022] S3.22: Use natural language processing methods to automatically extract entities from multi-source data, and use pattern matching methods to automatically extract the relationships between entities from multi-source data;

[0023] S3.23: Load the pre-trained graph neural network model, and use the graph neural network model to learn the interaction and dependency relationships between entities and relationships;

[0024] S3.24: Map the nodes and edges in the graph structure data to the low-dimensional vector space through the graph embedding method, and construct the nodes and edges of the knowledge graph according to the extracted entities, relationships and the embedding vectors generated by the graph embedding;

[0025] S3.25: Dynamically update the nodes and edges in the knowledge graph according to the real-time collected unit operating condition parameters;

[0026] The specific steps of the said S3.3 include:

[0027] S3.31: Use sensors to collect the unit operating condition parameters in real time, and load the dynamic unit operation knowledge graph;

[0028] S3.32: Integrate the real-time collected unit operating condition parameters with the data in the knowledge graph to obtain integrated data, and use fuzzy logic to convert the integrated data set into a fuzzy set;

[0029] S3.33: Use the neural network model to extract features from the fused fuzzy data set, perform representation learning on the nodes in the knowledge graph, and convert the nodes into vectors;

[0030] The specific steps of S3.3 also include:

[0031] S3.34: According to the output result of the neural network model, combined with the result of fuzzy logic processing, perform similarity measurement on the real-time working condition parameters and the nodes in the knowledge graph. The formula is:

[0032] ;

[0033] where, represents the similarity measurement, represents the weight on dimension i, represents the value of the real-time working condition parameter on dimension i, represents the value of the knowledge graph node on dimension i, represents the exponent of the non-linear transformation on dimension i, q represents the adjustment factor, and n represents the dimension of the node vector;

[0034] S3.35: According to the result of the similarity measurement, screen out the N historical working condition nodes with the highest similarity to the real-time working condition;

[0035] If the identifier of the historical optimal working condition is in the screening result of the nodes, calculate the similarity between the real-time working condition and the historical optimal working condition;

[0036] If the identifier of the historical optimal working condition is not in the screening result of the nodes, then comprehensively evaluate the similarity between the real-time working condition and the historical optimal working condition by comparing the size of the similarity scores in step S3.34;

[0037] The steps of the hybrid optimization algorithm in step S4 include:

[0038] S4.1: Load the pre-trained machine learning model and the parameter range in the joint optimization space in step S3, and randomly generate M parameter combinations in the joint optimization space as the initial evaluation points;

[0039] S4.2: Use the prediction performance index based on the machine learning model as the objective function, and use the objective function to evaluate each initial evaluation point, and record the evaluation results;

[0040] S4.3: Use the initial evaluation point data and the corresponding objective function values obtained in step S4.2 to train the Bayesian optimization model, and in each iteration, use the prediction result of the Bayesian optimization model to calculate the acquisition function;

[0041] S4.4: Select the next evaluation point according to the value of the acquisition function, evaluate the objective function for the selected evaluation point, record the results, and at the same time, update the Bayesian optimization model according to the new evaluation results;

[0042] S4.5: Set the number of iterations. When the preset number of iterations is reached, stop the iteration and output the optimal parameter combination found during the iteration and its corresponding objective function value as the final optimization result.

[0043] Specifically, the entities in the knowledge graph in S3.2 refer to the unit, components, operating conditions, and environmental parameters, and the attributes of the knowledge graph include temperature values, pressure ranges, and maintenance times.

[0044] A unit configuration adjustment system based on multi-algorithm optimization includes: a data processing module, an operating condition identification module, a matching module, a hybrid optimization module, and a unit adjustment module;

[0045] The data processing module is used to obtain the unit operation data and perform preprocessing;

[0046] The operating condition identification module is used to divide the unit operation state into different operating condition categories according to the preprocessed unit operation data, and identify the optimal operating parameter combination under each operating condition;

[0047] The matching module is used to match the real-time operating condition parameters with the nodes in the constructed unit operation knowledge graph through similarity measurement to obtain the historical optimal operating condition parameters;

[0048] The hybrid optimization module is used to load the pre-trained machine learning model, combine the parameter combinations in the joint optimization space, and obtain the optimal parameter combination through the hybrid optimization algorithm;

[0049] The unit adjustment module adjusts the unit configuration based on the optimal parameter combination, and monitors the operation state of the adjusted unit in real time. At the same time, it optimizes the parameters of the machine learning model according to the real-time feedback information.

[0050] Specifically, the operating condition identification module includes: an operating condition classification unit and an operating condition identification unit;

[0051] The operating condition classification unit is used to divide the unit operation state into different operating conditions using the clustering algorithm;

[0052] The operating condition identification unit screens out the optimal operating parameter combination through the optimization algorithm in each operating condition category and generates an optimal operating condition table for the unit operation.

[0053] Specifically, the matching module includes: a knowledge graph construction unit and a real-time matching unit;

[0054] The knowledge graph construction unit is used to construct a knowledge graph containing operating condition characteristics, parameter combinations, and effect evaluation information according to the optimal operating parameters;

[0055] The real-time matching unit is used to match the real-time operating condition parameters with the nodes in the knowledge graph using a similarity measurement strategy to locate the historical optimal operating condition parameters.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] 1. The present invention proposes a unit configuration adjustment system based on multi-algorithm optimization, and has optimized improvements in the architecture, operating steps, and processes. The system has the advantages of simple processes, low investment and operating costs, and low production work costs.

[0058] 2. The present invention proposes a unit configuration adjustment method based on multi-algorithm optimization. Through systematic data preprocessing and operating condition division, it realizes the refined management and analysis of unit operation data, effectively improves the recognition accuracy of unit operation states, and screens out the optimal operating parameter combinations in each operating condition category. This not only provides a scientific guiding basis for unit operation, but also deepens the understanding of unit performance through modeling analysis, helps optimize the operation strategy of the unit, improves operation efficiency, and reduces energy consumption and failure rates.

[0059] 3. The present invention proposes a unit configuration adjustment method based on multi-algorithm optimization, constructs a unit operation knowledge graph and introduces a similarity measurement strategy, realizes the rapid matching of real-time operating conditions and historical optimal operating conditions, and provides an accurate reference for unit adjustment; combines the constraint conditions of unit operation to set a joint optimization space, and uses a pre-trained machine learning model and a hybrid optimization algorithm for parameter optimization, improving the accuracy and efficiency of parameter adjustment. This not only ensures that the unit always operates in the optimal state, but also continuously optimizes the parameters of the machine learning model through real-time feedback information, improving the adaptability and prediction accuracy of the machine learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of a unit configuration adjustment method based on multi-algorithm optimization of the present invention;

[0061] Figure 2 It is a principle flowchart of a unit configuration adjustment method based on multi-algorithm optimization of the present invention;

[0062] Figure 3 It is a flowchart for realizing the optimal parameter combination of a unit configuration adjustment method based on multi-algorithm optimization of the present invention;

[0063] Figure 4 It is an architecture diagram of a unit configuration adjustment system based on multi-algorithm optimization of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0064] Example 1

[0065] Please refer to Figure 1 and Figure 2 , an embodiment provided by the present invention: a unit configuration adjustment method based on multi - algorithm optimization, comprising the following steps:

[0066] Step S1: Obtain the unit operation data and perform pre - processing. Based on the pre - processed unit operation data, divide the unit operation status into different working condition categories;

[0067] In the present invention, dividing the unit operation status into different working condition categories adopts a dynamic clustering analysis strategy. The dynamic clustering analysis strategy means first using the dynamic time warping algorithm for pre - processing to measure the distance between data, and then using the adaptive clustering analysis method to perform clustering according to the Euclidean distance result, so as to dynamically divide the unit operation status into different working condition categories. Among them, the working condition categories reflect the operation status of the unit under different working conditions.

[0068] It should be noted that the unit operation data is a series of information about the unit working status obtained from the unit operation system, which records the operation status of the unit at different time points and is the basis for evaluating the unit performance, monitoring the operation status, and performing fault diagnosis, etc.; while the unit operation status refers to the working condition or mode of the unit in the current or past period of time, including various working condition categories such as normal status, fault status, overload status, and low - load status; at the same time, the unit operation data is the basis for analyzing the unit operation status, and the unit operation status is identified by analyzing the change trend and characteristics of the unit operation data. Exemplarily, by monitoring the changes in power output, temperature, and pressure parameters, it can be judged whether the unit is in a normal or abnormal state. Therefore, the unit operation data is the basis for analyzing the unit operation status, and the unit operation status is the result of the operation data analysis and processing.

[0069] Further, the specific steps of step S1 include:

[0070] S1.1: Obtain the unit operation data from the unit operation monitoring system, including but not limited to temperature, pressure, flow rate, and vibration parameters, and perform data cleaning, data transformation, and feature selection on the unit operation data;

[0071] S1.2: Select the hierarchical clustering method according to the data characteristics and analysis requirements, and determine the optimal number of clusters, that is, the number of working condition categories, through the silhouette coefficient;

[0072] S1.3: Use the dynamic time warping algorithm to calculate the Euclidean distance for the preprocessed unit operation data, and use the hierarchical clustering method to cluster the Euclidean distance calculation results to generate clustering results, and evaluate the clustering results to ensure that the clustering effect meets expectations. At the same time, analyze the characteristics of each working condition category and explain the unit operation states they represent. Among them, the calculation formulas of the hierarchical clustering method and the dynamic time warping algorithm are the existing technical content in the field and are not the creative solutions of this application, so they will not be elaborated here;

[0073] S1.4: Apply the clustering results to the actual scenarios of unit operation optimization and fault diagnosis, collect feedback, and continuously optimize the clustering parameters.

[0074] Step S2: In each working condition category, screen out the optimal operation parameter combinations, generate the optimal working condition table for unit operation, and conduct modeling analysis on the optimal operation parameters;

[0075] Furthermore, the specific steps of step S2 include:

[0076] S2.1: For each working condition category, screen out all the unit operation data belonging to this category from the preprocessed unit operation data to form different parameter combinations;

[0077] S2.2: Set maximizing efficiency, minimizing energy consumption, and improving stability as optimization goals, and calculate the performance index values corresponding to each parameter combination according to the optimization goals. The calculation formula of the performance index value is:

[0078] ;

[0079] Among them, represents the comprehensive performance index value, , , represent the weights of the optimization goals, E represents the efficiency value under each parameter combination, c represents the total energy consumption under each parameter combination, represents the maximum value of the total energy consumption under each parameter combination, and S represents the stability of each parameter combination;

[0080] S2.3: According to the performance index values, screen out the optimal operation parameter combinations in each working condition category;

[0081] S2.4: Organize the optimal operation parameter combinations in each working condition category into a table form, including the working condition category identifier, the optimal operation parameter combination and its corresponding performance index value;

[0082] S2.5: Select the modeling method of the machine learning model for modeling, and use the optimal operation parameters as the training set to train the machine learning model;

[0083] S2.6: Analyze the results of the machine learning model to understand the relationships between parameters and how they affect performance metrics.

[0084] Step S3: Construct a knowledge graph of unit operation, and match the real-time operating condition parameter values with the nodes in the knowledge graph through a similarity measurement strategy to obtain the historical optimal operating condition parameters. At the same time, based on the historical optimal operating condition parameters and combined with the constraint conditions of unit operation, set up a joint optimization space;

[0085] Among them, the constraint conditions include: safety thresholds and energy efficiency indicators.

[0086] Step S4: Load the pre-trained machine learning model. At the same time, input the parameter combinations within the joint optimization space into the machine learning model, and obtain the optimal parameter combination through a hybrid optimization algorithm;

[0087] Step S5: Based on the optimal parameter combination, adjust the unit configuration and monitor the real-time operating state of the adjusted unit. At the same time, according to the real-time feedback information, optimize the parameters of the machine learning model.

[0088] Furthermore, the specific steps of step S5 include:

[0089] S5.1: According to the optimal parameter combination obtained by Bayesian optimization, adjust the configuration parameters of the unit, such as temperature, pressure, and flow control parameters, to ensure the safety and smoothness of the adjustment process and avoid unnecessary damage to the unit;

[0090] S5.2: Real-time monitor the unit operation data, and preprocess the real-time monitored unit operation data, including cleaning, denoising, and normalization operations, analyze the unit operation state, and identify the characteristics and trends of the unit operation state, such as whether it is in a stable operation state, whether there are signs of abnormalities or faults;

[0091] S5.3: Use real-time data to evaluate the accuracy of the current machine learning model , if , then use the Bayesian optimization method to optimize the parameters of the machine learning model according to the real-time feedback information. Among them, the calculation process of Bayesian optimization is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here;

[0092] S5.4: Redeploy the optimized machine learning model to the monitoring system, continue to collect data and evaluate the model performance, and according to the new evaluation results and feedback information, optimize the model parameters again, and repeat this process until the machine learning model's .

[0093] Embodiment 2

[0094] Please refer to Figure 3, in this embodiment, the specific steps of step S3 include:

[0095] S3.1: Obtain environmental parameters and historical unit maintenance records, and generate a multi-source dataset by combining the optimal working condition table and the modeling results generated in step S2;

[0096] Among them, the environmental parameters include temperature, humidity, and air pressure; the unit maintenance records include failure types, repair times, and replaced parts.

[0097] S3.2: Define the entities, relationships, and attributes of the knowledge graph, and use graph neural networks and graph embedding methods to automatically extract and construct the nodes and edges of the knowledge graph from the multi-source dataset to form a dynamic unit operation knowledge graph;

[0098] The entities of the knowledge graph in S3.2 refer to units, components, working conditions, and environmental parameters. The attributes of the knowledge graph include temperature values, pressure ranges, and maintenance times. The relationships include containment, influence, and causation.

[0099] Exemplarily, there is a containment relationship between the unit and the unit equipment components, and there is a causation relationship between the failure type and the maintenance record.

[0100] S3.3: Real-time collect the unit working condition parameters, and through the fusion of fuzzy logic, combine the deep learning algorithm to measure the similarity between the real-time unit working condition parameters and the nodes in the knowledge graph, and comprehensively evaluate the similarity between the real-time working condition and the historical optimal working condition ;

[0101] S3.4: According to the similarity result, take the corresponding unit working condition parameters as the historical optimal working condition parameters, where denotes the maximum value of;

[0102] S3.5: Based on the historical optimal working condition parameters, combine the constraint conditions of the unit operation to construct a multi-objective optimization model;

[0103] S3.6: Use the multi-objective optimization algorithm to solve the multi-objective optimization model to generate a multi-dimensional parameter adjustment plan. At the same time, according to the real-time working condition feedback, dynamically adjust the parameter fluctuation range and step size to form a joint optimization space.

[0104] The specific steps of S3.2 include:

[0105] S3.21: Define the entities, relationships, and attributes of the knowledge graph, and obtain the multi-source dataset;

[0106] S3.22: Automatically extract entities from multi-source data using natural language processing methods, and automatically extract the relationships between entities from multi-source data through pattern matching methods. The calculation formulas of the natural language processing method and the pattern matching method are the prior art content in this field and are not the creative solutions of this application, so they will not be elaborated here;

[0107] S3.23: Load a pre-trained graph neural network model, and use the graph neural network model to learn the interaction and dependency relationships between entities and relationships;

[0108] Further, the specific steps of S3.23 include:

[0109] (1) Load a pre-trained graph convolutional network model;

[0110] (2) Convert the extracted entities and relationships into graph structure data. Usually, entities are used as nodes and relationships are used as edges, and construct the corresponding adjacency matrix or graph object;

[0111] (3) Configure the parameters of the graph convolutional network model according to the characteristics of the specific task and dataset, including the learning rate, the number of iterations, the number of layers, and the size of the hidden layer;

[0112] (4) Use the trained graph convolutional network model to perform inference on the graph structure data, and learn the interaction and dependency relationships between entities and relationships. Among them, the graph convolutional network model updates the representation of each node by aggregating the information of neighbor nodes, so as to capture the complex relationships between nodes. The inference process and result output process of the graph convolutional network model are the prior art content in this field and are not the creative solutions of this application, so they will not be elaborated here;

[0113] (5) Output the node representation or relationship prediction result learned by the graph convolutional network model.

[0114] S3.24: Map the nodes and edges in the graph structure data to a low-dimensional vector space through graph embedding methods, and construct the nodes and edges of the knowledge graph according to the extracted entities, relationships, and the embedding vectors generated by graph embedding;

[0115] Further, the specific steps of S3.24 include:

[0116] (1) Collect graph structure data containing entity and relationship information, and clean the data to remove noise and outliers. At the same time, extract the nodes and edges in the graph;

[0117] (2) Use the graph embedding method of DeepWalk to train the graph structure data. During the training process, the graph convolutional network model maps the nodes and edges in the graph to a low-dimensional vector space. Among them, the training process of the graph embedding method of DeepWalk is the prior art content in the field and not the creative solution of this application, so it will not be elaborated here. At the same time, the mapping process of the graph convolutional network model is the prior art content in the field and not the creative solution of this application, so it will not be elaborated here;

[0118] (3) Evaluate the quality of the embedding vectors through node classification and adjust the parameters of the graph convolutional network model according to the evaluation results. Among them, the calculation formula of the node classification method is the prior art content in the field and not the creative solution of this application, so it will not be elaborated here;

[0119] (4) Construct the nodes and edges of the knowledge graph according to the extracted entities and relationships, and the embedding vectors generated by graph embedding, where the embedding vectors are used as the features of the nodes;

[0120] (5) Optimize the constructed knowledge graph, such as removing redundant information, merging similar nodes, and deploy the knowledge graph in the actual scenario.

[0121] S3.25: Dynamically update the nodes and edges in the knowledge graph according to the real-time collected unit operating conditions parameters, including adding new nodes, updating node attributes, modifying or deleting edges.

[0122] The specific steps of S3.3 include:

[0123] S3.31: Use sensors to collect unit operating conditions parameters in real time and load the dynamic unit operation knowledge graph;

[0124] Among them, the dynamic unit operation knowledge graph contains the historical operating conditions, operation rules, and constraint condition information of the unit.

[0125] S3.32: Integrate the real-time collected unit operating conditions parameters with the data in the knowledge graph to obtain integrated data, and use fuzzy logic to convert the integrated data set into a fuzzy set to handle the uncertainty and ambiguity in the data;

[0126] S3.33: Use a neural network model to extract features from the fused fuzzy data set and perform representation learning on the nodes in the knowledge graph, converting the nodes into vectors or points in a high-dimensional space for similarity measurement. Among them, the process and steps of feature extraction are the prior art content in the field and not the creative solution of this application, so it will not be elaborated here;

[0127] S3.34: According to the output result of the neural network model and combined with the result of fuzzy logic processing, perform similarity measurement on the real-time working condition parameters and the nodes in the knowledge graph. The formula is as follows:

[0128] ;

[0129] where, represents the similarity measurement, represents the weight on dimension i, which is used to represent the importance of different dimensions to the similarity measurement, represents the value of the real-time working condition parameter on dimension i, represents the value of the knowledge graph node on dimension i, represents the exponent of the non-linear transformation on dimension i, which is used to adjust the influence degree of the differences on different dimensions. If when, the influence of the difference is linear. If when, the larger difference will be amplified. If when, the larger difference will be reduced. q represents the adjustment factor, which is used to control the rate of difference accumulation on different dimensions. If when, the formula degenerates into the normalized form of the weighted Manhattan distance. If when, the overall difference will be amplified. If when, the overall difference will be reduced. n represents the dimension of the node vector;

[0130] S3.35: According to the result of the similarity measurement, screen out the N historical working condition nodes with the highest similarity to the real-time working condition;

[0131] If the identifier of the historical optimal working condition is in the screening result of the nodes, calculate the similarity between the real-time working condition and the historical optimal working condition;

[0132] If the identifier of the historical optimal working condition is not in the screening result of the nodes, comprehensively evaluate the similarity between the real-time working condition and the historical optimal working condition by comparing the similarity scores in step S3.34.

[0133] The steps of the hybrid optimization algorithm in step S4 include:

[0134] S4.1: Load the pre-trained machine learning model and the parameter range in the joint optimization space in step S3, and randomly generate M parameter combinations as the initial evaluation points in the joint optimization space;

[0135] S4.2: Use the prediction performance index based on the machine learning model as the objective function, and evaluate each initial evaluation point using the objective function, and record the evaluation results;

[0136] Furthermore, the specific steps of S4.2 include: ​

[0137] S4.21: Load the pre-trained machine learning model and determine the mapping relationship between the input parameters and the output prediction results of the machine learning model;

[0138] S4.22: Select the accuracy metric as the objective function;

[0139] S4.23: For each initial evaluation point, i.e., a set of parameter combinations, use this set of parameters as the input of the machine learning model, run the machine learning model, and obtain the corresponding prediction results;

[0140] S4.24: Calculate the accuracy metric based on the prediction results and the true labels, and record the values of the prediction performance metrics corresponding to each initial evaluation point.

[0141] S4.3: Use the initial evaluation point data and the corresponding objective function values obtained in step S4.2 to train the Bayesian optimization model, and in each iteration, calculate the acquisition function using the prediction results of the Bayesian optimization model;

[0142] Further, the specific steps of S4.3 include:

[0143] S4.31: Obtain the initial evaluation point data and the corresponding objective function values, i.e., the accuracy metric, from step S4.1;

[0144] S4.32: Select the RBF kernel function to describe the covariance relationship between the parameter combinations and the objective function values, and use the initial evaluation point data as the training data to initialize the Gaussian process model, including setting the mean function, covariance matrix, and noise term of the Gaussian process model;

[0145] S4.33: Use the initial evaluation point data to train the Gaussian process model to learn the mapping relationship between the parameter combinations and the objective function values. The training process of the Gaussian process model is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here;

[0146] S4.34: In each iteration, use the current Gaussian process model to predict the objective function values of the unevaluated parameter combinations, i.e., the predicted mean and variance. The prediction process of the Gaussian process model is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here;

[0147] S4.35: Calculate the EI acquisition function based on the prediction results. The EI acquisition function is used to evaluate the expected benefits of different parameter combinations, and the calculation process of the EI acquisition function is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here;

[0148] S4.36: Select the parameter combination that maximizes the acquisition function value as the next evaluation point, evaluate the objective function at the selected evaluation point, and record the new evaluation result;

[0149] S4.37: Add the data of the new evaluation point and the objective function value to the training data, and retrain the Gaussian process model;

[0150] S4.38: Repeat the iterative process of steps S4.34 to S4.37 until the preset number of iterations is reached.

[0151] S4.4: Select the next evaluation point according to the value of the acquisition function, evaluate the objective function at the selected evaluation point, and record the result. At the same time, update the Bayesian optimization model according to the new evaluation result;

[0152] S4.5: Set the number of iterations. When the preset number of iterations is reached, stop the iteration, and output the optimal parameter combination found during the iteration and its corresponding objective function value as the final optimization result.

[0153] Embodiment 3

[0154] Please refer to Figure 4 , another embodiment provided by the present invention: A unit configuration adjustment system based on multi-algorithm optimization, including:

[0155] A data processing module, a working condition identification module, a matching module, a hybrid optimization module, and a unit adjustment module;

[0156] The data processing module is used to obtain the unit operation data and perform preprocessing;

[0157] The working condition identification module is used to divide the unit operation state into different working condition categories according to the preprocessed unit operation data, and identify the optimal operation parameter combination under each working condition;

[0158] The matching module is used to match the real-time working condition parameters with the nodes in the constructed unit operation knowledge graph through similarity measurement to obtain the historical optimal working condition parameters;

[0159] The hybrid optimization module is used to load the pre-trained machine learning model, combine the parameter combinations in the joint search space, and obtain the optimal parameter combination through the hybrid optimization algorithm;

[0160] The unit adjustment module adjusts the unit configuration based on the optimal parameter combination, and monitors the operation state of the adjusted unit in real time. At the same time, it optimizes the parameters of the machine learning model according to the real-time feedback information.

[0161] The working condition identification module includes: a working condition classification unit and a working condition identification unit;

[0162] The operating condition classification unit is used to divide the operating state of the unit into different operating conditions by using a clustering algorithm;

[0163] The operating condition identification unit screens out the optimal operating parameter combinations through an optimization algorithm in each operating condition category and generates an optimal operating condition table for the unit operation.

[0164] The matching module includes: a knowledge graph construction unit and a real-time matching unit;

[0165] The knowledge graph construction unit is used to construct a knowledge graph containing operating condition characteristics, parameter combinations, and effect evaluation information according to the optimal operating parameters;

[0166] The real-time matching unit is used to match the real-time operating condition parameters with the nodes in the knowledge graph by using a similarity measurement strategy to quickly locate the historical optimal operating condition parameters.

[0167] The unit adjustment module includes: a configuration adjustment unit, a real-time monitoring unit, and a model optimization unit;

[0168] The configuration adjustment unit is used to send adjustment instructions to the unit control system according to the optimal parameter combination to realize the automatic adjustment of the unit configuration;

[0169] The real-time monitoring unit is used to monitor the operating state of the unit in real time through sensors and the control system and collect feedback data;

[0170] The model optimization unit is used to optimize the parameters of the machine learning model by using an online learning method according to the real-time feedback information and the unit operation effect to improve the accuracy and adaptability of the machine learning model.

[0171] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make changes, modifications, substitutions, and variations to the above embodiments without departing from the spirit and scope of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A unit configuration adjustment method based on multi-algorithm optimization, characterized in that: include: Step S1: Acquire the unit operation data and perform preprocessing, and divide the unit operation status into different operating condition categories based on the preprocessed unit operation data; Step S2: In each operating condition category, select the optimal operating parameter combination, generate the unit operating optimal operating condition table, and perform modeling analysis on the optimal operating parameters; Step S3: Construct a knowledge graph of unit operation, and match the real-time operating parameter values ​​with the nodes in the knowledge graph through a similarity measurement strategy to obtain the historical optimal operating parameters. At the same time, based on the historical optimal operating parameters and combined with the constraints of unit operation, set a joint optimization space; Step S4: loading the pre-trained machine learning model, and at the same time, inputting the parameter combination in the joint optimization space into the machine learning model, and obtaining the optimal parameter combination through the hybrid optimization algorithm; Step S5: Based on the optimal parameter combination, the unit configuration is adjusted, and the adjusted unit operation status is monitored in real time. At the same time, the machine learning model parameters are optimized according to the real-time feedback information; The specific steps of step S3 include: S3.1: Obtain environmental parameters and historical unit maintenance records, combine the optimal operating condition table and modeling results generated in step S2, and generate a multi-source data set; S3.2: Define the entities, relationships and attributes of the knowledge graph, and use graph neural networks and graph embedding methods to automatically extract and construct nodes and edges of the knowledge graph from multi-source data sets to form a dynamic unit operation knowledge graph; S3.3: Collect unit operating parameters in real time, measure the similarity between real-time unit operating parameters and nodes in the knowledge graph by integrating fuzzy logic and combining deep learning algorithms, and comprehensively evaluate the similarity between real-time operating conditions and historical optimal operating conditions. ; S3.4: According to the similarity results, The corresponding unit operating parameters are taken as the historical optimal operating parameters, where: express The maximum value of S3.5: Based on the historical optimal operating parameters and combined with the constraints of unit operation, a multi-objective optimization model is constructed; S3.6: Use the multi-objective optimization algorithm to solve the multi-objective optimization model and generate a multi-dimensional parameter adjustment plan. At the same time, according to the real-time working condition feedback, dynamically adjust the parameter fluctuation range and step size to form a joint optimization space; The specific steps of S3.2 include: S3.21: Define entities, relationships, and attributes of the knowledge graph and obtain multi-source datasets; S3.22: Automatically extract entities from multi-source data using natural language processing methods, and automatically extract relationships between entities from multi-source data using pattern matching methods; S3.23: Load the pre-trained graph neural network model and use the graph neural network model to learn the interactions and dependencies between entities and relationships; S3.24: Map the nodes and edges in the graph structure data into a low-dimensional vector space through graph embedding methods, and construct the nodes and edges of the knowledge graph based on the embedding vectors generated by the extracted entities, relationships, and graph embedding; S3.25: Dynamically update the nodes and edges in the knowledge graph based on the unit operating parameters collected in real time; The specific steps of S3.3 include: S3.31: Use sensors to collect unit operating parameters in real time and load dynamic unit operation knowledge graph; S3.32: Fuse the unit operating parameters collected in real time with the data in the knowledge graph to obtain fused data, and convert the fused data set into a fuzzy set using fuzzy logic; S3.33: Use the neural network model to extract features from the fused fuzzy data set, perform representation learning on the nodes in the knowledge graph, and convert the nodes into vectors; The specific steps of S3.3 also include: S3.34: Based on the output results of the neural network model and the results of fuzzy logic processing, the similarity between the real-time operating parameters and the nodes in the knowledge graph is measured. The formula is: ； in, represents the similarity measure, represents the weight on dimension i, represents the value of the real-time operating condition parameter in dimension i, represents the value of the knowledge graph node on dimension i, represents the exponent of the nonlinear transformation on dimension i, q represents the adjustment factor, and n represents the dimension of the node vector; S3.35: based on the similarity measurement result, select N historical operating condition nodes with the highest similarity to the real-time operating condition; If the identifier of the historical optimal working condition is in the screening result of the node, the similarity between the real-time working condition and the historical optimal working condition is calculated; If the identifier of the historical optimal working condition is not in the screening result of the node, the similarity between the real-time working condition and the historical optimal working condition is comprehensively evaluated by comparing the similarity scores in step S3.34; The steps of the hybrid optimization algorithm in step S4 include: S4.1: Load the pre-trained machine learning model and the parameter range in the joint optimization space in step S3, and randomly generate M parameter combinations in the joint optimization space as initial evaluation points; S4.2: Use the prediction performance indicator based on the machine learning model as the objective function, and use the objective function to evaluate each initial evaluation point and record the evaluation results; S4.3: training the Bayesian optimization model using the initial evaluation point data and the corresponding objective function value obtained in step S4.2, and calculating the acquisition function using the prediction result of the Bayesian optimization model in each iteration; S4.4: Select the next evaluation point according to the value of the acquisition function, evaluate the objective function at the selected evaluation point, and record the result. At the same time, update the Bayesian optimization model according to the new evaluation result; S4.5: Set the number of iterations. When the preset number of iterations is reached, stop the iteration and output the optimal parameter combination found in the iteration process and its corresponding objective function value as the final optimization result.

2. A unit configuration adjustment method based on multi-algorithm optimization as claimed in claim 1, characterized in that: The entities of the knowledge graph in S3.2 refer to units, components, operating conditions, and environmental parameters, and the attributes of the knowledge graph include temperature values, pressure ranges, and maintenance time.

3. A unit configuration adjustment system based on multi-algorithm optimization, which is used to implement a unit configuration adjustment method based on multi-algorithm optimization as described in any one of claims 1-2, characterized in that: include: Data processing module, operating condition identification module, matching module, hybrid optimization module, unit adjustment module; The data processing module is used to obtain the unit operation data and perform preprocessing; The operating condition identification module is used to classify the unit operating status into different operating condition categories according to the preprocessed unit operating data, and identify the optimal operating parameter combination under each operating condition; The matching module is used to match the real-time operating condition parameters with the nodes in the constructed unit operation knowledge graph through similarity measurement to obtain the historical optimal operating condition parameters; The hybrid optimization module is used to load the pre-trained machine learning model, combine the parameter combination in the joint optimization space, and obtain the optimal parameter combination through the hybrid optimization algorithm; The unit adjustment module adjusts the unit configuration based on the optimal parameter combination, monitors the adjusted unit operating status in real time, and optimizes the machine learning model parameters according to real-time feedback information.

4. A unit configuration adjustment system based on multi-algorithm optimization as claimed in claim 3, characterized in that: The working condition identification module includes: a working condition classification unit and a working condition identification unit; The operating condition classification unit is used to classify the unit operating state into different operating conditions using a clustering algorithm; The operating condition identification unit selects the optimal operating parameter combination in each operating condition category through an optimization algorithm and generates an optimal operating condition table for the unit.

5. The unit configuration adjustment system based on multi-algorithm optimization according to claim 4, characterized in that: The matching module includes: a knowledge graph construction unit and a real-time matching unit; The knowledge graph construction unit is used to construct a knowledge graph including operating condition characteristics, parameter combinations and effect evaluation information according to the optimal operating parameters; The real-time matching unit is used to match the real-time operating parameters with the nodes in the knowledge graph using a similarity measurement strategy to locate the historical optimal operating parameters.

Images