Power station key equipment temperature prediction method based on multi-source heterogeneous data fusion

Abstract

The invention relates to the technical field of power station equipment temperature prediction, and provides a power station key equipment temperature prediction method based on multi-source heterogeneous data fusion, and the method comprises the following steps: collecting multi-source heterogeneous data of a hydropower station, carrying out the preprocessing, feature extraction, feature selection and feature fusion of the multi-source heterogeneous data, and obtaining a temperature prediction fusion feature; performing hyper-parameter optimization on the initial back-propagation neural network through an improved underwater foraging optimization algorithm to obtain an optimized back-propagation neural network; inputting the temperature prediction fusion feature into an optimized back propagation neural network to obtain a temperature prediction value; and comparing the temperature prediction value with the actual temperature value, and when the error of the temperature prediction value is within a preset range, generating a temperature prediction table. According to the method, temperature abnormity can be found in time, early warning of equipment faults and generation of a temperature prediction table are facilitated, the method is suitable for processing complex multi-source heterogeneous data, and the accuracy of temperature prediction is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of power station equipment temperature prediction, and in particular to a method for predicting the temperature of key power station equipment based on multi-source heterogeneous data fusion. Background Art

[0002] Temperature prediction is crucial for online monitoring and fault diagnosis of equipment. Accurate temperature prediction can help detect equipment abnormalities in a timely manner, prevent potential failures, extend equipment service life, and improve the operating efficiency and safety of hydropower stations. The temperature changes of key equipment in hydropower stations are affected by many factors, such as equipment load, ambient temperature, cooling system efficiency, etc., and there is a complex nonlinear relationship between these factors.

[0003] Existing temperature prediction methods include model-based prediction algorithms, classical machine learning-based prediction algorithms, and deep learning-based prediction algorithms. Traditional models and machine learning methods have limitations when processing nonlinear and complex data. Although the prediction accuracy of deep learning methods is not low, there are problems such as high computational complexity and long training time, and it is impossible to predict the device temperature in a timely manner. Summary of the invention

[0004] In view of this, the present invention proposes a temperature prediction method for key power plant equipment based on multi-source heterogeneous data fusion, which solves the problems of high complexity and long training time in the prior art and inability to predict the equipment temperature in a timely manner.

[0005] The technical solution of the present invention is implemented as follows: The present invention provides a method for predicting the temperature of key equipment in a power station based on multi-source heterogeneous data fusion, comprising the following steps:

[0006] S1, collecting multi-source heterogeneous data of the hydropower station, preprocessing, feature extraction, feature selection and feature fusion of the multi-source heterogeneous data, and obtaining temperature prediction fusion features;

[0007] S2, optimizing the hyperparameters of the initial back propagation neural network through the improved underwater foraging optimization algorithm to obtain the optimized back propagation neural network;

[0008] S3, inputting the temperature prediction fusion feature into the optimized back propagation neural network to obtain a temperature prediction value;

[0009] S4, comparing the temperature prediction value with the actual temperature value, and generating a temperature prediction table when the error of the temperature prediction value is within a preset range.

[0010] Based on the above technical solution, preferably, step S2 includes:

[0011] S21, determining the structure of the initial back propagation neural network, the structure including the number of nodes in the input layer, the hidden layer and the output layer;

[0012] S22, setting the hyperparameters of the initial back propagation neural network, wherein the hyperparameters include a learning rate, a number of hidden layers, a number of neurons in each layer, an activation function type, a batch size, and a number of training rounds;

[0013] S23, define the search space of hyperparameters and set upper and lower bounds for each hyperparameter;

[0014] S24, initialize the underwater forager population, each underwater forager represents a candidate solution for a set of hyperparameters;

[0015] S25, uses the improved underwater forager algorithm for iterative search, and evaluates the performance of the current hyperparameter combination after each iterative search;

[0016] S26, updating the position of the underwater forager according to the evaluation result to update the value of the hyperparameter;

[0017] S27, repeating step S25 and step S26 until a preset number of iterations is reached or a preset convergence condition is met;

[0018] S28, based on the evaluation results, an optimal hyperparameter combination is obtained as the hyperparameter for optimizing the back propagation neural network.

[0019] On the basis of the above technical solution, preferably, the improved underwater forager algorithm includes:

[0020] In the initialization stage, the population position of underwater foragers is initialized to obtain the initial population position;

[0021] In the floating strategy stage, based on the initial population position, the potential optimal solution is searched in the search space;

[0022] In the deep dive behavior stage, the potential optimal solution is further developed and updated to obtain a better solution;

[0023] In the group gathering stage, a corresponding random number is generated for each underwater forager. When the random number is greater than the preset escape probability threshold, an escape operation is performed to update the position of the corresponding underwater forager; otherwise, the position of the corresponding underwater forager is kept unchanged to obtain the optimal hyperparameter combination.

[0024] On the basis of the above technical solution, preferably, the calculation formula of the improved underwater forager algorithm includes:

[0025] X i,j =LB+(UB-LB)×rand csi=1,2,...,N and j=1,2,...,Dim;

[0026]

[0027] Among them, X i,j is the position of the ith underwater forager in the jth dimension, UB and LB are the upper and lower boundaries of the search space, respectively, and rand cs is a random number between 0 and 1, is the tth number of the kth underwater forager in the floating strategy stage fy The position at +1 iteration, is the tth number of the kth underwater forager in the floating strategy stage fy The position at the iteration, α is a random parameter, T is the dynamic strategy parameter in the floating strategy stage, is the t of the lth underwater forager in the floating strategy stage fy The position at the iteration, N is the number of underwater foragers, is a random number between 1 and Dim, exp(·) is an exponential function, Max_Iter is the preset maximum number of iterations, BF is the beating factor, is a random number between 0 and 1, is the tth time of the ath underwater forager in the deep diving behavior stage sq The position at +1 iteration, is the tth time of the ath underwater forager in the deep diving behavior stage sq The position at the iteration, HA is the first hunting ability parameter, o is the second hunting ability parameter, b is the optimal offset, X best (t sq ) is the tth stage of deep diving behavior sq The optimal position of the iteration, is the tth time of the cth underwater forager in the group gathering stage qt The position at +1 iteration, is the tth time of the cth underwater forager in the group gathering stage qt The position at the iteration, PE is the predation efficiency of the underwater forager, X n (t qt ) is the tth number of the nth underwater forager in the group gathering stage qt The position at the iteration, is the tth number of the mth underwater forager in the group gathering stage qt The position at the iteration, rand is a random number between 0 and 1.

[0028] On the basis of the above technical solution, preferably, the deep diving behavior stage is introduced with a periodic function optimization idea based on a special ratio to obtain an improved deep diving behavior stage, and the calculation formula includes:

[0029]

[0030] x1=e*(1-t)+f*t;

[0031] x2=e*t+f*(1-t);

[0032] in, is the tth time step of the ath underwater forager in the improved deep diving behavior stage sq The position at +1 iteration, is the tth time step of the ath underwater forager in the improved deep diving behavior stage sq The position at the iteration, r1 and r2 are the first special ratio parameters and the second special ratio parameters respectively, x1 and x2 are the first special ratio segmentation parameters and the second special ratio segmentation parameters respectively, e and f are the initial values ​​of the special ratio search, and t is the segmentation ratio.

[0033] On the basis of the above technical solution, preferably, step S1 includes:

[0034] Preprocessing the multi-source heterogeneous data includes:

[0035] A temperature increment matrix is ​​formed based on the temperature increments at adjacent moments, and a temperature increment data set is obtained by integrating all temperature increment matrices;

[0036] The sample points in the temperature increment data set are randomly divided into multiple clusters, and the cluster centroids are randomly selected. The number of cluster centroids selected is equal to the number of clusters.

[0037] Calculate the distance from each sample point to the centroid of the cluster, and classify the sample points into a category according to the proximity principle;

[0038] Recalculate the cluster centroid by calculating the mean of all sample points in each cluster;

[0039] The clusters are evaluated using the internal evaluation index of the clustering algorithm, and the optimal number of clusters is jointly selected;

[0040] An optimal threshold is determined according to the clustering result, outliers in the multi-source heterogeneous data are judged according to the selected optimal threshold, and the outliers are adjusted to obtain first multi-source heterogeneous data.

[0041] On the basis of the above technical solution, preferably, the first multi-source heterogeneous data includes a one-dimensional time series signal and a two-dimensional image signal, and the feature extraction includes:

[0042] The one-dimensional time series signal is subjected to feature extraction through a transformer model, and the two-dimensional image signal is subjected to feature extraction through a sliding window model to obtain a multi-source heterogeneous feature data set.

[0043] On the basis of the above technical solution, preferably, the feature selection includes:

[0044] The multi-source heterogeneous feature data sets are input into the XGBoost model, and importance is sorted to obtain the optimal feature data set.

[0045] On the basis of the above technical solution, preferably, the feature fusion includes:

[0046] The feature vectors in the optimal feature data set are spliced ​​through an early fusion model, the spliced ​​feature vectors are normalized, the normalized feature vectors are nonlinearly changed using a convolutional layer, and the temperature prediction fusion features are output.

[0047] Based on the above technical solution, preferably, step S4 includes:

[0048] S41, obtaining a data set of temperature prediction values ​​and actual temperature values, and calculating the absolute error, mean absolute error, mean square error, and root mean square error of the temperature prediction value and the actual temperature value of each data point;

[0049] S42, setting a preset error threshold, comparing the calculated mean absolute error, mean square error and root mean square error with the preset threshold, and when all error indicators are within the preset range, executing step S45; otherwise, executing step S43;

[0050] S43, checking the hyperparameters of the optimized back-propagation neural network, re-optimizing the hyperparameters of the optimized back-propagation neural network using an improved underwater forager algorithm, re-evaluating the feature selection results, and adjusting the early fusion model;

[0051] S44, use the adjusted early fusion model to perform temperature prediction on the validation set, repeat the error calculation and evaluation process of step S41 and step S42, when the mean absolute error, mean square error and root mean square error still exceed the preset range, return to step S43 to adjust the early fusion model again, when the mean absolute error, mean square error and root mean square error still exceed the preset range after multiple consecutive adjustments, expand the preset error range;

[0052] S45, creating a temperature prediction table including a timestamp, an actual temperature value, a predicted temperature value, an absolute error, and a relative error, filling in corresponding data for each time point, calculating and adding summary statistical information, and generating a temperature change trend graph;

[0053] S46, save the temperature prediction table in a specific format, store the temperature prediction table in a specific database, record the generation time and related parameters, generate a report file containing prediction result analysis, error statistics and temperature change trend chart, set up an early warning reporting mechanism, send an alarm when an abnormal temperature trend is detected, and establish an archiving system for historical prediction results.

[0054] The temperature prediction method of key power plant equipment based on multi-source heterogeneous data fusion of the present invention has the following beneficial effects compared with the prior art:

[0055] (1) By integrating the multi-source heterogeneous data of the hydropower station, high-quality temperature prediction fusion features are obtained through preprocessing, feature extraction, feature selection and feature fusion. The improved underwater foraging optimization algorithm is used to optimize the hyperparameters of the back propagation neural network, which improves the performance and prediction accuracy of the model. The temperature is predicted by the optimized neural network and compared with the actual temperature value to ensure that the accuracy of the prediction result is within the preset range. It can detect temperature anomalies in time, which is helpful for early warning of equipment failure and generate temperature prediction tables. It provides an intuitive and effective tool for temperature monitoring and management of key equipment in the power station. It is suitable for processing complex multi-source heterogeneous data and enhances the accuracy of temperature prediction.

[0056] (2) Through the improved underwater forager algorithm, the position update mechanism of underwater foragers at different stages is described, including initialization, floating strategy, deep diving behavior, and group aggregation. The random parameters, dynamic strategy parameters, jump factors, and predation efficiency factors are comprehensively used to achieve a better balance between global search and local development, achieving faster convergence speed, higher optimization accuracy, and stronger ability to jump out of local optimality, thereby improving the reliability of the temperature prediction model of key equipment in power plants;

[0057] (3) By introducing the optimization idea of ​​periodic functions based on special proportions in the deep diving behavior stage and using a specific calculation formula to improve the deep diving behavior stage, the global search capability is enhanced, the local optimal solution is avoided, the optimization accuracy and convergence speed are improved, and the global exploration and local development are balanced, thereby improving the optimization performance of the underwater forager algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1The present invention is a flow chart of a method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion. DETAILED DESCRIPTION

[0060] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] See also Figure 1 The present invention provides a method for predicting the temperature of key equipment in a power station based on multi-source heterogeneous data fusion, comprising the following steps:

[0062] S1, collecting multi-source heterogeneous data of the hydropower station, preprocessing, feature extraction, feature selection and feature fusion of the multi-source heterogeneous data, and obtaining temperature prediction fusion features;

[0063] S2, optimizing the hyperparameters of the initial back propagation neural network through the improved underwater foraging optimization algorithm to obtain the optimized back propagation neural network;

[0064] S3, inputting the temperature prediction fusion feature into the optimized back propagation neural network to obtain a temperature prediction value;

[0065] S4, comparing the temperature prediction value with the actual temperature value, and generating a temperature prediction table when the error of the temperature prediction value is within a preset range.

[0066] Specifically, this embodiment obtains high-quality temperature prediction fusion features by fusing multi-source heterogeneous data of the hydropower station through preprocessing, feature extraction, feature selection and feature fusion steps, and uses an improved underwater foraging optimization algorithm to optimize the hyperparameters of the back propagation neural network, thereby improving the performance and prediction accuracy of the model. The temperature is predicted by the optimized neural network and compared with the actual temperature value to ensure that the accuracy of the prediction result is within a preset range. Temperature anomalies can be discovered in time, which is helpful for early warning of equipment failures and generating temperature prediction tables, thereby providing an intuitive and effective tool for temperature monitoring and management of key equipment in power stations. The tool is suitable for processing complex multi-source heterogeneous data and enhances the accuracy of temperature prediction.

[0067] Step S1 includes:

[0068] Preprocessing the multi-source heterogeneous data includes:

[0069] A temperature increment matrix is ​​formed based on the temperature increments at adjacent moments, and a temperature increment data set is obtained by integrating all temperature increment matrices;

[0070] The sample points in the temperature increment data set are randomly divided into multiple clusters, and the cluster centroids are randomly selected. The number of cluster centroids selected is equal to the number of clusters.

[0071] Calculate the distance from each sample point to the centroid of the cluster, and classify the sample points into a category according to the proximity principle;

[0072] Recalculate the cluster centroid by calculating the mean of all sample points in each cluster;

[0073] The clusters are evaluated using the internal evaluation index of the clustering algorithm, and the optimal number of clusters is jointly selected;

[0074] An optimal threshold is determined according to the clustering result, outliers in the multi-source heterogeneous data are judged according to the selected optimal threshold, and the outliers are adjusted to obtain first multi-source heterogeneous data.

[0075] In a specific embodiment, the temperature increments at adjacent moments are combined into a temperature increment matrix M I (Total m measuring points)

[0076] Let s∈(1,m) be a detection point in the optical fiber. Assume that the temperature curve of this detection point consists of n temperature values, then there are n-1 temperature increments at adjacent moments. I The sample points in are randomly divided into k clusters, and k cluster centroids are randomly selected

[0077] Calculate each sample M separately I，js The distance to the cluster centroid is calculated, and the sample points are divided into a category according to the proximity principle, which is called

[0078]

[0079] Recalculate cluster centroids

[0080]

[0081] The internal evaluation indicators CH and DB of the clustering algorithm are used to jointly select the optimal k value.

[0082]

[0083] The first category of the clustering results generally represents a normal temperature increase, so the mean method is used to determine the optimal threshold.

[0084]

[0085] The outliers are determined based on the optimal threshold selected. (NaN is replaced by the nearest non-missing value) An overall adjustment is made.

[0086] The first multi-source heterogeneous data includes a one-dimensional time series signal and a two-dimensional image signal, and the feature extraction includes:

[0087] The one-dimensional time series signal is subjected to feature extraction through a transformer model, and the two-dimensional image signal is subjected to feature extraction through a sliding window model to obtain a multi-source heterogeneous feature data set.

[0088] The feature selection includes:

[0089] The multi-source heterogeneous feature data sets are input into the XGBoost model, and importance is sorted to obtain the optimal feature data set.

[0090] The feature fusion includes:

[0091] The feature vectors in the optimal feature data set are spliced ​​through an early fusion model, the spliced ​​feature vectors are normalized, the normalized feature vectors are nonlinearly changed using a convolutional layer, and the temperature prediction fusion features are output.

[0092] Specifically, this embodiment enriches the data sources by simultaneously collecting and processing one-dimensional time series signals and two-dimensional image signals, avoids the problem of information loss that may be caused by a single data source, and uses the Transformer model to extract features of one-dimensional time series signals. It can effectively capture the long-term and short-term dependencies and complex dynamic changes in time series data, thereby improving the understanding of time series patterns.

[0093] Using a sliding window model to extract features from two-dimensional image signals can extract local and global image features and capture the spatial structure and change trend in image signals. By inputting multi-source heterogeneous feature data sets into the XGBoost model for importance sorting, the optimal feature data set is screened out, redundant and useless features are eliminated, the complexity of the model is reduced, overfitting is avoided, and the generalization ability of the model is improved.

[0094] By concatenating the feature vectors in the optimal feature data set, the key features of different data sources are integrated, and the concatenated feature vectors are normalized to eliminate the scale differences between different features and improve the comparability between features. The convolution layer is used to perform nonlinear changes on the normalized feature vectors to explore the deep associations between features and improve the nonlinear ability of feature expression.

[0095] Step S2 includes:

[0096] S21, determining the structure of the initial back propagation neural network, the structure including the number of nodes in the input layer, the hidden layer and the output layer;

[0097] S22, setting the hyperparameters of the initial back propagation neural network, wherein the hyperparameters include a learning rate, a number of hidden layers, a number of neurons in each layer, an activation function type, a batch size, and a number of training rounds;

[0098] S23, define the search space of hyperparameters and set upper and lower bounds for each hyperparameter;

[0099] S24, initialize the underwater forager population, each underwater forager represents a candidate solution for a set of hyperparameters;

[0100] S25, uses the improved underwater forager algorithm for iterative search, and evaluates the performance of the current hyperparameter combination after each iterative search;

[0101] S26, updating the position of the underwater forager according to the evaluation result to update the value of the hyperparameter;

[0102] S27, repeating step S25 and step S26 until a preset number of iterations is reached or a preset convergence condition is met;

[0103] S28, based on the evaluation results, an optimal hyperparameter combination is obtained as the hyperparameter for optimizing the back propagation neural network.

[0104] Specifically, this embodiment lays the foundation for hyperparameter optimization by determining the structure of the initial back propagation neural network, setting the initial hyperparameters, and defining the search space of the hyperparameters. The improved underwater forager algorithm is used to iteratively search and optimize the hyperparameters, which can automatically find the optimal hyperparameter combination and greatly improve the training effect of the neural network.

[0105] The improved underwater forager algorithm has good global optimization capabilities. During the iterative search process, the algorithm dynamically adjusts the position of the underwater forager through a bionic mechanism to avoid falling into the local optimum and ensure that the global optimal hyperparameter combination is found.

[0106] A population of underwater foragers is initialized, each forager represents a candidate solution for a set of hyperparameters, and swarm intelligence is used to accelerate the search process.

[0107] By setting upper and lower bounds for each hyperparameter, the search space of the hyperparameter is precisely defined, ensuring that the search process is carried out within a reasonable range, preventing model divergence caused by excessive learning rate or slow convergence caused by too small learning rate, as well as overfitting caused by too many hidden layers or neurons. During the optimization process, the performance of the current hyperparameter combination is evaluated at each iteration, and the position of the forager is updated based on the evaluation results. This optimization strategy based on performance feedback ensures that the final hyperparameter combination can achieve the best performance of the back-propagation neural network.

[0108] The improved underwater forager algorithm includes:

[0109] In the initialization stage, the population position of underwater foragers is initialized to obtain the initial population position;

[0110] In the floating strategy stage, based on the initial population position, the potential optimal solution is searched in the search space;

[0111] In the deep dive behavior stage, the potential optimal solution is further developed and updated to obtain a better solution;

[0112] In the group gathering stage, a corresponding random number is generated for each underwater forager. When the random number is greater than the preset escape probability threshold, an escape operation is performed to update the position of the corresponding underwater forager; otherwise, the position of the corresponding underwater forager is kept unchanged to obtain the optimal hyperparameter combination.

[0113] The calculation formula of the improved underwater forager algorithm includes:

[0114] X i,j =LB+(UB-LB)×rand cs i=1,2,...,N and j=1,2,...,Dim;

[0115]

[0116]

[0117] Among them, X i,j is the position of the ith underwater forager in the jth dimension, UB and LB are the upper and lower boundaries of the search space, respectively, and rand cs is a random number between 0 and 1, is the tth number of the kth underwater forager in the floating strategy stage fy The position at +1 iteration, is the tth number of the kth underwater forager in the floating strategy stage fy The position at the iteration, α is a random parameter, T is the dynamic strategy parameter in the floating strategy stage, is the t of the lth underwater forager in the floating strategy stage fy The position at the iteration, N is the number of underwater foragers, is a random number between 1 and Dim, exp(·) is an exponential function, Max_Iter is the preset maximum number of iterations, BF is the beating factor, is a random number between 0 and 1, is the tth time of the ath underwater forager in the deep diving behavior stage sq The position at +1 iteration, is the tth time of the ath underwater forager in the deep diving behavior stage sq The position at the iteration, HA is the first hunting ability parameter, o is the second hunting ability parameter, b is the optimal offset, X best (t sq ) is the tth stage of deep diving behavior sq The optimal position of the iteration, is the tth time of the cth underwater forager in the group gathering stage qt The position at +1 iteration, is the tth time of the cth underwater forager in the group gathering stage qt The position at the iteration, PE is the predation efficiency of the underwater forager, X n (t qt ) is the tth number of the nth underwater forager in the group gathering stage qt The position at the iteration, is the tth number of the mth underwater forager in the group gathering stage qt The position at the iteration, rand is a random number between 0 and 1.

[0118] Specifically, in the initialization stage, this embodiment initializes the population positions of underwater foragers so that the foragers are evenly distributed in the search space, thereby increasing the diversity of the population and improving the global search capability of the algorithm.

[0119] During the floating strategy phase: Based on the initial population position, the forager searches for the potential optimal solution in the search space. This phase uses randomness and dynamic strategy parameters T to promote the exploration ability of the population and expand the search range.

[0120] During the deep dive behavior phase: the potential optimal solution is deeply developed and updated. By introducing the first hunting ability parameter HA and the second hunting ability parameter o, the search ability is enhanced, which helps to finely search the area around the optimal solution and improve the accuracy of the solution.

[0121] In the group gathering stage, the random number generated by each underwater forager is compared with the preset escape probability threshold to decide whether to perform the escape operation. This mechanism gives the forager a probability to jump out of the current area and avoid falling into the local optimum.

[0122] By dynamically adjusting the strategy parameter T and the beating factor BF, the search process can adaptively adjust the search behavior according to the number of iterations and the quality of the current solution, thereby improving the convergence speed and accuracy of the algorithm.

[0123] The hyperparameters of the back propagation neural network are optimized by using the improved underwater forager algorithm. This method can efficiently search the hyperparameter space and find the optimal hyperparameter combination.

[0124] The deep diving behavior stage is improved by introducing the optimization idea of ​​periodic function based on special proportions, and the calculation formula includes:

[0125]

[0126] x1=e*(1-t)+f*t;

[0127] x2=e*t+f*(1-t);

[0128] in, is the tth time step of the ath underwater forager in the improved deep diving behavior stage sq The position at +1 iteration, is the tth time step of the ath underwater forager in the improved deep diving behavior stage sq The position at the iteration, r1 and r2 are the first special ratio parameters and the second special ratio parameters respectively, x1 and x2 are the first special ratio segmentation parameters and the second special ratio segmentation parameters respectively, e and f are the initial values ​​of the special ratio search, and t is the segmentation ratio.

[0129] Specifically, this embodiment uses a periodic function to adjust the position of the forager so that it has a wider exploration capability in the search space, thereby avoiding the situation where the algorithm falls into a local optimal solution.

[0130] By setting the first special ratio parameter r1 and the second special ratio parameter r2, as well as the first special ratio segmentation parameter x1 and the second special ratio segmentation parameter x2, the search range and step length of the algorithm are controlled, thereby improving the global search capability.

[0131] The introduction of a periodic function with a special proportion makes the position update of the underwater forager have a certain periodicity and regularity, which makes the algorithm converge to the global optimal solution faster.

[0132] The special ratio search initial values ​​e and f enable the algorithm to perform refined search during the optimization process to improve the accuracy of the solution. In this embodiment, e and f are respectively -π and and π, and t is

[0133] The introduced special scale parameter can be dynamically adjusted according to the number of iterations and the position of the current solution, which enhances the adaptability of the algorithm to different problems.

[0134] By introducing periodic functions of special proportions, the algorithm dynamically adjusts the position of the forager in each iteration and utilizes the periodic characteristics of sine and cosine functions to make the forager move along a periodic trajectory in the search space, thereby enhancing the diversity of the search process.

[0135] Step S4 includes:

[0136] S41, obtaining a data set of temperature prediction values ​​and actual temperature values, and calculating the absolute error, mean absolute error, mean square error, and root mean square error of the temperature prediction value and the actual temperature value of each data point;

[0137] S42, setting a preset error threshold, comparing the calculated mean absolute error, mean square error and root mean square error with the preset threshold, and when all error indicators are within the preset range, executing step S45; otherwise, executing step S43;

[0138] S43, checking the hyperparameters of the optimized back-propagation neural network, re-optimizing the hyperparameters of the optimized back-propagation neural network using an improved underwater forager algorithm, re-evaluating the feature selection results, and adjusting the early fusion model;

[0139] S44, use the adjusted early fusion model to perform temperature prediction on the validation set, repeat the error calculation and evaluation process of step S41 and step S42, when the mean absolute error, mean square error and root mean square error still exceed the preset range, return to step S43 to adjust the early fusion model again, when the mean absolute error, mean square error and root mean square error still exceed the preset range after multiple consecutive adjustments, expand the preset error range;

[0140] S45, creating a temperature prediction table including a timestamp, an actual temperature value, a predicted temperature value, an absolute error, and a relative error, filling in corresponding data for each time point, calculating and adding summary statistical information, and generating a temperature change trend graph;

[0141] S46, save the temperature prediction table in a specific format, store the temperature prediction table in a specific database, record the generation time and related parameters, generate a report file containing prediction result analysis, error statistics and temperature change trend chart, set up an early warning reporting mechanism, send an alarm when an abnormal temperature trend is detected, and establish an archiving system for historical prediction results.

[0142] Specifically, this embodiment obtains a data set of temperature prediction values ​​and actual temperature values, calculates the absolute error of each data point, as well as the overall mean absolute error, mean square error and root mean square error, sets a preset error threshold, and compares the calculated error index with the threshold to ensure that the prediction accuracy of the model meets the requirements. Through the evaluation of multiple error indicators, the prediction performance of the model is comprehensively measured.

[0143] When the error index exceeds the preset range, the hyperparameters of the back-propagation neural network are checked and re-optimized. The improved underwater forager algorithm is used to accelerate the optimization process. Combined with hyperparameter optimization, the feature selection results of the early fusion model are re-evaluated and adjusted to ensure the validity of the input features. Through multiple cycles of error calculation and model adjustment, the prediction error is gradually reduced, and the accuracy and robustness of the model are improved.

[0144] If the error still does not meet the preset range after multiple consecutive adjustments, the error threshold is expanded to avoid model overfitting caused by infinite loops. By properly adjusting the error threshold, the model accuracy and computational cost are balanced to improve the overall efficiency of the algorithm.

[0145] Create a temperature prediction table that includes timestamps, actual temperature values, predicted temperature values, absolute error, relative error, and other information. Comprehensively record the prediction results, calculate and add summary statistics, and generate a temperature trend chart to visually display how the temperature changes over time, which helps to identify potential problems.

[0146] The temperature prediction table is saved in a specific format and stored in a specific database to facilitate unified data management and subsequent calls, generate a report file containing prediction result analysis, error statistics and temperature change trend chart, improve the efficiency of information transmission, set up an early warning reporting mechanism, and automatically send an alarm when an abnormal temperature trend is detected, provide timely feedback, improve the timeliness of equipment safety monitoring, and establish an archiving system for historical prediction results to facilitate the query and analysis of historical data, which is conducive to the study of long-term trends and the continuous improvement of models.

[0147] From error assessment to model optimization and then to result application, a complete closed-loop optimization process is formed to ensure that the model is always in the best working state. Through dynamic adjustment of the model and parameters, the model's adaptability to different data characteristics is enhanced, and the accuracy and stability of the prediction are improved. The introduction of early warning mechanism and data storage management improves the response capability to abnormal situations and ensures the safe and reliable operation of the equipment.

[0148] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A temperature prediction method for key equipment in a power station based on multi-source heterogeneous data fusion, characterized in that: The following steps are involved: S1, collecting multi-source heterogeneous data of the hydropower station, preprocessing, feature extraction, feature selection and feature fusion of the multi-source heterogeneous data, and obtaining temperature prediction fusion features; S2, optimizing the hyperparameters of the initial back propagation neural network through the improved underwater foraging optimization algorithm to obtain the optimized back propagation neural network; S3, inputting the temperature prediction fusion feature into the optimized back propagation neural network to obtain a temperature prediction value; S4, comparing the temperature prediction value with the actual temperature value, and generating a temperature prediction table when the error of the temperature prediction value is within a preset range.

2. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion as claimed in claim 1, characterized in that: Step S2 includes: S21, determining the structure of the initial back propagation neural network, the structure including the number of nodes in the input layer, the hidden layer and the output layer; S22, setting the hyperparameters of the initial back propagation neural network, wherein the hyperparameters include a learning rate, a number of hidden layers, a number of neurons in each layer, an activation function type, a batch size, and a number of training rounds; S23, define the search space of hyperparameters and set upper and lower bounds for each hyperparameter; S24, initialize the underwater forager population, each underwater forager represents a candidate solution for a set of hyperparameters; S25, uses the improved underwater forager algorithm for iterative search, and evaluates the performance of the current hyperparameter combination after each iterative search; S26, updating the position of the underwater forager according to the evaluation result to update the value of the hyperparameter; S27, repeating step S25 and step S26 until a preset number of iterations is reached or a preset convergence condition is met; S28, based on the evaluation results, an optimal hyperparameter combination is obtained as the hyperparameter for optimizing the back propagation neural network.

3. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion as claimed in claim 2, characterized in that: The improved underwater forager algorithm includes: In the initialization stage, the population position of underwater foragers is initialized to obtain the initial population position; In the floating strategy stage, based on the initial population position, the potential optimal solution is searched in the search space; In the deep dive behavior stage, the potential optimal solution is further developed and updated to obtain a better solution; In the group gathering stage, a corresponding random number is generated for each underwater forager. When the random number is greater than the preset escape probability threshold, an escape operation is performed to update the position of the corresponding underwater forager; otherwise, the position of the corresponding underwater forager is kept unchanged to obtain the optimal hyperparameter combination.

4. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion as claimed in claim 3, characterized in that: The calculation formula of the improved underwater forager algorithm includes: X i,j =LB+(UB-LB)×rand cs i=1,2,...,N and j=1,2,...,Dim; Among them, X i,j is the position of the ith underwater forager in the jth dimension, UB and LB are the upper and lower boundaries of the search space, respectively, and rand cs is a random number between 0 and 1, is the tth number of the kth underwater forager in the floating strategy stage fy The position at +1 iteration, is the tth number of the kth underwater forager in the floating strategy stage fy The position at the iteration, α is a random parameter, T is the dynamic strategy parameter in the floating strategy stage, is the t of the lth underwater forager in the floating strategy stage fy The position at the iteration, N is the number of underwater foragers, is a random number between 1 and Dim, exp(·) is an exponential function, Max_Iter is the preset maximum number of iterations, BF is the beating factor, is a random number between 0 and 1, is the tth time of the ath underwater forager in the deep diving behavior stage sq The position at +1 iteration, is the tth time of the ath underwater forager in the deep diving behavior stage sq The position at the iteration, HA is the first hunting ability parameter, o is the second hunting ability parameter, b is the optimal offset, X best (t sq ) is the tth stage of deep diving behavior sq The optimal position of the iteration, is the tth time of the cth underwater forager in the group gathering stage qt The position at +1 iteration, is the tth time of the cth underwater forager in the group gathering stage qt The position at the iteration, PE is the predation efficiency of the underwater forager, X n (t qt ) is the tth number of the nth underwater forager in the group gathering stage qt The position at the iteration, is the tth number of the mth underwater forager in the group gathering stage qt The position at the iteration, rand is a random number between 0 and 1.

5. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion as claimed in claim 4, characterized in that: The deep diving behavior stage is improved by introducing the optimization idea of ​​periodic function based on special proportions, and the calculation formula includes: x1=e*(1-t)+f*t; x2=e*t+f*(1-t); in, is the tth time step of the ath underwater forager in the improved deep diving behavior stage sq The position at +1 iteration, is the tth time step of the ath underwater forager in the improved deep diving behavior stage sq The position at the iteration, r1 and r2 are the first special ratio parameters and the second special ratio parameters respectively, x1 and x2 are the first special ratio segmentation parameters and the second special ratio segmentation parameters respectively, e and f are the initial values ​​of the special ratio search, and t is the segmentation ratio.

6. The method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion according to claim 1, characterized in that: Step S1 includes: Preprocessing the multi-source heterogeneous data includes: A temperature increment matrix is ​​formed based on the temperature increments at adjacent moments, and a temperature increment data set is obtained by integrating all temperature increment matrices; The sample points in the temperature increment data set are randomly divided into multiple clusters, and the cluster centroids are randomly selected. The number of cluster centroids selected is equal to the number of clusters. Calculate the distance from each sample point to the centroid of the cluster, and classify the sample points into a category according to the proximity principle; Recalculate the cluster centroid by calculating the mean of all sample points in each cluster; The clusters are evaluated using the internal evaluation index of the clustering algorithm, and the optimal number of clusters is jointly selected; An optimal threshold is determined according to the clustering result, outliers in the multi-source heterogeneous data are judged according to the selected optimal threshold, and the outliers are adjusted to obtain first multi-source heterogeneous data.

7. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion as claimed in claim 6, characterized in that: The first multi-source heterogeneous data includes a one-dimensional time series signal and a two-dimensional image signal, and the feature extraction includes: The one-dimensional time series signal is subjected to feature extraction through a transformer model, and the two-dimensional image signal is subjected to feature extraction through a sliding window model to obtain a multi-source heterogeneous feature data set.

8. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion as claimed in claim 7, characterized in that: The feature selection includes: The multi-source heterogeneous feature data sets are input into the XGBoost model, and importance is sorted to obtain the optimal feature data set.

9. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion as claimed in claim 8, characterized in that: The feature fusion includes: The feature vectors in the optimal feature data set are spliced ​​through an early fusion model, the spliced ​​feature vectors are normalized, the normalized feature vectors are nonlinearly changed using a convolutional layer, and the temperature prediction fusion features are output.

10. A method for predicting temperature of key equipment in a power station based on multi-source heterogeneous data fusion according to claim 9, characterized in that: Step S4 includes: S41, obtaining a data set of temperature prediction values ​​and actual temperature values, and calculating the absolute error, mean absolute error, mean square error, and root mean square error of the temperature prediction value and the actual temperature value of each data point; S42, setting a preset error threshold, comparing the calculated mean absolute error, mean square error and root mean square error with the preset threshold, and when all error indicators are within the preset range, executing step S45; otherwise, executing step S43; S43, checking the hyperparameters of the optimized back-propagation neural network, re-optimizing the hyperparameters of the optimized back-propagation neural network using an improved underwater forager algorithm, re-evaluating the feature selection results, and adjusting the early fusion model; S44, use the adjusted early fusion model to perform temperature prediction on the validation set, repeat the error calculation and evaluation process of step S41 and step S42, when the mean absolute error, mean square error and root mean square error still exceed the preset range, return to step S43 to adjust the early fusion model again, when the mean absolute error, mean square error and root mean square error still exceed the preset range after multiple consecutive adjustments, expand the preset error range; S45, creating a temperature prediction table including timestamp, actual temperature value, predicted temperature value, absolute error, and relative error, filling in corresponding data for each time point, calculating and adding summary statistical information, and generating a temperature change trend graph; S46, save the temperature prediction table in a specific format, store the temperature prediction table in a specific database, record the generation time and related parameters, generate a report file containing prediction result analysis, error statistics and temperature change trend chart, set up an early warning reporting mechanism, send an alarm when an abnormal temperature trend is detected, and establish an archiving system for historical prediction results.

Images