An optimized control system and decision-making method for peak shaving in thermal power units

Abstract

This application provides an optimized control system and decision-making method for peak shaving of thermal power units, relating to the field of machine learning technology. The method includes: collecting and analyzing the operating characteristic values ​​and peak shaving performance data of thermal power units; training a neural network model capable of predicting the peak shaving performance of thermal power units using machine learning; clustering thermal power units with similar operating characteristics using a clustering algorithm; and using a data mining algorithm to analyze the correlation reasons for poor peak shaving performance of thermal power units, uncovering the implicit patterns and regularities between peak shaving performance and operation of thermal power units, thereby achieving control and management of the thermal power units. The system integrates a working condition data acquisition module, a peak shaving performance prediction module, a working condition pattern clustering module, and a peak shaving decision mining module to collect thermal power unit data in real time, quickly predict peak shaving performance, and analyze correlation reasons. This application uncovers the performance correlation reasons of thermal power units, providing insights for optimization decision-making.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, specifically designing an optimized control system and decision-making method for peak shaving of thermal power units. Background Technology

[0002] With the increasing complexity of power systems and the growing variety and number of thermal power units, real-time monitoring of equipment status and fault diagnosis face the challenge of processing massive amounts of data.

[0003] Chinese Patent Publication No. CN115438893A discloses a method and system for evaluating the operating status of power dispatching equipment. The method includes: acquiring the network operation status value, protocol transmission status value, and control service status value of the power dispatching system corresponding to a fault event occurring in the power dispatching equipment; taking the maximum value of the network operation status value, protocol transmission status value, and control service status value as the power dispatching system status value corresponding to the fault event; determining whether the power dispatching system status value meets a preset threshold condition; if the preset threshold condition is met, determining that the fault event has occurred; establishing an evaluation guideline, the evaluation guideline including the fault event and the weight corresponding to the fault event; subtracting the weight corresponding to the fault event from a preset score to obtain a score for the operating status of the power dispatching equipment. Currently, thermal power unit management systems mainly rely on traditional threshold alarms and periodic inspections, which make it difficult to detect the performance degradation trend and potential fault risks of equipment in a timely manner from a large amount of status data, leading to unplanned equipment outages and affecting the reliability of the power grid.

[0004] Therefore, there is an urgent need for an intelligent equipment management system that, based on machine learning technology, can uncover fault patterns in massive amounts of operational data to achieve online assessment of equipment status and predictive maintenance. Summary of the Invention

[0005] This application aims to at least partially address one of the technical problems in the related art. To this end, one objective of this application is to propose an optimized control system and decision-making method for peak shaving in thermal power units, thereby improving the availability of data resources for thermal power units.

[0006] One aspect of this application is to provide an optimized control decision-making method for peak shaving of thermal power units. This method predicts future peak shaving performance indicators by training a neural network model, clusters thermal power units with similar learning and operating characteristics using a clustering algorithm, and analyzes the correlations between peak shaving performance and other factors within each category using data mining algorithms. This allows power maintenance personnel to better address the individual differences among thermal power units based on their operating status and provide more targeted and effective maintenance measures. This method offers faster and more accurate assessment of thermal power units. Based on the assessment results, the system predicts the future peak shaving performance of the units and analyzes the causes of poor predicted peak shaving performance, allowing for human intervention based on the analysis results. Power maintenance personnel can proactively identify performance degradation and potential faults in thermal power units and take timely measures for corresponding maintenance and repairs.

[0007] It includes the following steps:

[0008] Step 11: Obtain the historical operating characteristic sequence set of thermal power units and the corresponding peak-shaving performance index sequence;

[0009] Step 12: Train a neural network model to predict future peak-shaving performance indicators by using the historical operating characteristic sequence set of thermal power units and the corresponding peak-shaving performance index sequence.

[0010] Step 13: Select the historical sub-peak performance thermal power unit dataset from all thermal power units, and use the historical sub-peak performance thermal power unit dataset to cluster the thermal power units through a clustering algorithm to obtain K thermal power unit clusters and train the clustering algorithm model.

[0011] Step 14: Based on the K thermal power unit clusters, obtain L thermal power unit datasets. Use data mining algorithms to mine the L thermal power unit datasets to obtain the frequent itemsets of each thermal power unit dataset; K=L;

[0012] Step 15: Set the frequent itemsets as the associated operation feature set, and set a decision label for each thermal power unit cluster based on the associated operation feature set;

[0013] Step 16: Obtain the actual peak-shaving performance index sequence and the actual operating characteristic sequence set, input them into the neural network model to predict the future peak-shaving performance of the thermal power unit, and obtain the future peak-shaving performance data of the thermal power unit.

[0014] Step 17: Based on the future peak-shaving performance data of thermal power units, select the sub-peak-shaving performance thermal power unit dataset, input each peak-shaving performance index and operating characteristic in the sub-peak-shaving performance thermal power unit dataset into the clustering algorithm model, and obtain the decision label of the thermal power unit cluster to which each thermal power unit belongs.

[0015] Each element in the historical operating feature sequence set is a historical operating feature sequence, wherein each historical operating feature sequence includes a sequence of operating feature values ​​of the corresponding historical operating feature, and the operating feature values ​​in the historical operating feature sequence are arranged in chronological order; the corresponding peak-shaving performance index sequence includes peak-shaving performance data of each thermal power unit for each test, and the peak-shaving performance data in the corresponding peak-shaving performance index sequence are arranged in chronological order.

[0016] The historical operating characteristics refer to relevant operating factors that affect the peak-shaving performance of thermal power units, including but not limited to historical operating duration, historical start-up and shutdown frequency, historical load rate, and historical overload frequency; the operating characteristic values ​​are the specific numerical values ​​of the corresponding historical operating characteristics.

[0017] The historical operating characteristic sequence is collected as follows: between every two tests of each thermal power unit, the corresponding historical operating characteristic value is collected once, and this operating characteristic value is added to the historical operating characteristic sequence in chronological order. For example, thermal power units are tested regularly, and the duration between every two tests is a data acquisition period T. The operating characteristic values ​​of the corresponding historical operating characteristics of the thermal power unit acquired in the t-th test within a data acquisition period are historical operating duration At, historical start-stop count Bt, historical load rate Ct, and historical overload count Dt, respectively. Here, t is the number of times the operating characteristic values ​​and peak-shaving performance data are acquired in chronological order, At is the operating characteristic value of the historical operating duration of the thermal power unit acquired in the t-th test within the data acquisition period, and Bt is the historical overload count Dt. The historical start-stop frequency of thermal power units acquired in the data acquisition period is denoted as {A1, A2, ..., At}, the historical start-stop frequency of thermal power units acquired in the data acquisition period is denoted as {B1, B2, ..., Bt}, the historical load rate of thermal power units acquired in the data acquisition period is denoted as {C1, C2, ..., Ct}, and the historical overload frequency of thermal power units acquired in the data acquisition period is denoted as {D1, D2, ..., Dt}.

[0018] The method for collecting the corresponding peak-shaving performance index sequence is as follows: there is one test within each data acquisition period T, and the peak-shaving performance data obtained from the test is used as the peak-shaving performance data corresponding to the historical operating characteristics within the corresponding data acquisition period T. The peak-shaving performance data obtained in the tth time within a data acquisition period T is Et, and the corresponding peak-shaving performance index sequence for collecting the peak-shaving performance indexes t times is {E1, E2, ..., Et}.

[0019] Each peak-shaving performance data in the corresponding peak-shaving performance index sequence corresponds one-to-one with each operating characteristic value in the historical operating characteristic sequence. For example, in collecting t historical operating characteristic sequences and corresponding peak-shaving performance index sequences, the first peak-shaving performance data E1 corresponds to the first operating characteristic values ​​A1, B1, C1, and D1; the second peak-shaving performance data E2 corresponds to the second operating characteristic values ​​A2, B2, C2, and D2; and the t-th peak-shaving performance data Et corresponds to the t-th operating characteristic values ​​At, Bt, Ct, and Dt.

[0020] The step of training a neural network model to predict future peak-shaving performance indicators by using a set of historical operating feature sequences of thermal power units and corresponding peak-shaving performance indicator sequences includes:

[0021] The prediction time step, sliding step, and sliding window length are preset. The historical operating feature sequence and the corresponding peak-shaving performance index sequence are divided into multiple groups of data samples according to the window sliding method. Each group of data samples is used as the input data of the neural network model for predicting the future peak-shaving performance index, and the corresponding peak-shaving performance index sequence of the prediction time step is used as the output. Each group of data samples includes a historical operating feature sequence of one sliding window length and a corresponding peak-shaving performance index sequence of one sliding window length.

[0022] Within the prediction time step of each peak-shaving performance index sequence, the peak-shaving performance index sequence is used as the prediction target, and the minimum error between the predicted peak-shaving performance index sequence and the actual peak-shaving performance index sequence is used as the training target to train the neural network model for predicting future peak-shaving performance indexes.

[0023] Preferably, the neural network model is an RNN neural network model or an LSTM neural network model;

[0024] The prediction time step, sliding step, and sliding window length are preset based on practical experience.

[0025] The method for selecting the historical sub-peak performance thermal power unit dataset from all thermal power units is as follows: input the historical operating feature sequence and peak performance index sequence into a trained neural network model for predicting future peak performance indicators, output the future peak performance data of the thermal power units, set the thermal power units with scores less than F as sub-peak performance thermal power units, use the operating feature values ​​of the sub-peak performance thermal power units as historical operating feature items, and obtain the historical sub-peak performance thermal power unit dataset from the peak performance index sequence and historical operating feature sequence of the sub-peak performance thermal power units; F is a preset value.

[0026] The specific method for clustering thermal power units using a clustering algorithm to obtain K clusters of thermal power units and training the clustering algorithm model is as follows:

[0027] Step 1301: Obtain thermal power unit data points by using the historical operating characteristic sequence and corresponding peak-shaving performance index sequence of thermal power units in the historical sub-peak-shaving performance thermal power unit dataset; wherein, each thermal power unit data point contains the historical operating characteristic sequence of the corresponding thermal power unit.

[0028] Step 1302: Randomly select K1 data points from thermal power units as initial cluster centers Xi; the initial cluster centers Xi are randomly selected data points from thermal power units.

[0029] Step 1303: Calculate the Euclidean distance from the data points of the thermal power units other than the initial cluster center to each initial cluster center, and assign each data point of the thermal power unit to the cluster of the thermal power units containing the nearest initial cluster center;

[0030] Step 1304: Calculate the average value of each operating characteristic value of the thermal power unit data points within each thermal power unit cluster, form a new thermal power unit data point from the average value of each operating characteristic value, and use the new thermal power unit data point as the new cluster center Xi' of the thermal power unit cluster;

[0031] Step 1305: Recalculate the Euclidean distance from each thermal power unit data point to the new cluster center, and reassign each thermal power unit data point to the thermal power unit cluster where the nearest new cluster center Xi' is located;

[0032] Step 1306: Repeat steps 1304-1305 until the clusters to which all thermal power unit data points are assigned no longer change, resulting in K thermal power unit clusters. The clustering algorithm model training is complete, where K=K1.

[0033] The method for obtaining L thermal power unit datasets based on K thermal power unit clusters is as follows: the thermal power unit data in the K thermal power unit clusters exist in the form of thermal power unit data points. The operating characteristic value and peak-shaving performance index of each thermal power unit data point constitute a thermal power unit sample data in the thermal power unit dataset. The operating characteristic value and peak-shaving performance data of the thermal power unit corresponding to each thermal power unit data point constitute a set of thermal power unit data, thus obtaining L thermal power unit datasets, where K=L.

[0034] The method for using data mining algorithms to mine L thermal power unit datasets and obtain the frequent itemsets of each thermal power unit dataset is as follows:

[0035] The data mining algorithm is the Apriori algorithm. The Apriori algorithm is used to mine frequent itemsets for each thermal power unit dataset. Preferably, the minimum support threshold M can be 0.4.

[0036] The Apriori algorithm involves mining frequent itemsets for each thermal power unit dataset and obtaining the associated runtime feature set. The steps are as follows:

[0037] Step 1401: Based on actual needs and the characteristics of the thermal power unit dataset, set a minimum support threshold M. Preferably, the minimum support threshold M can be 0.4.

[0038] Step 1402: Set counter j, with an initial value of 1, scan the thermal power unit dataset, calculate the frequency of each peak-shaving performance data and historical operating characteristic in the thermal power unit dataset, and generate the j-th candidate set G(j);

[0039] Step 1403: Scan the j-th candidate itemset G(j); label the support as S, filter out candidate items whose support S is greater than or equal to the minimum support threshold M, and generate the j-th frequent itemset H(j); it should be noted that the method of scanning the candidate itemset is a conventional technique in the art, and will not be described in detail here.

[0040] Step 1404: Perform join and pruning operations on the j-th frequent itemset H(j), update the value of j to j+1, and generate the (j+1)-th candidate itemset G(j+1); It should be noted that the method of joining and pruning the j-th frequent itemset is a conventional technique in the art, and will not be described in detail here.

[0041] Step 1405: Scan the (j+1)th candidate itemset G(j+1), count the number of candidates whose support S is greater than or equal to the minimum support threshold M, and generate the (j+1)th frequent itemset H(j+1).

[0042] Step 1406: Repeat steps 1404 to 1405 to continuously generate new candidate item sets. Statistically count the candidate items whose support is greater than or equal to the minimum support threshold to obtain new frequent itemsets until the generated candidate item sets are empty.

[0043] Step 1407: Set the obtained frequent itemsets as the associated running feature set, and sort the frequent itemsets in the associated running feature set from largest to smallest according to the weight of the frequent items in the frequent itemsets;

[0044] Step 1408: Number all associated running feature sets, and use the number of the associated running feature set as the decision label of the thermal power unit cluster in which the corresponding thermal power unit dataset belongs.

[0045] The actual peak-shaving performance index sequence refers to the actual peak-shaving performance data of the thermal power unit acquired t times according to the data acquisition period T, and the actual peak-shaving performance indexes are combined according to the acquisition time order to obtain the corresponding peak-shaving performance index sequence.

[0046] The actual operating feature sequence set refers to a set composed of multiple actual operating feature sequences. The actual operating feature sequence is obtained by combining actual operating feature values ​​in chronological order. The method for obtaining the actual operating feature values ​​is as follows: within a data acquisition period T, the operating feature value between two tests is obtained as an actual operating feature value. The actual operating feature value is obtained t times to obtain an actual operating feature sequence containing t actual operating feature values. The operating feature values ​​of multiple actual operating features are obtained to obtain multiple actual operating feature sequences, which together constitute the actual operating feature sequence set.

[0047] The steps for predicting the future peak-shaving performance of thermal power units are as follows: inputting the actual set of operating characteristic sequences of each thermal power unit and the actual sequence of peak-shaving performance indicators of each thermal power unit into a trained neural network model for predicting the future peak-shaving performance data of thermal power units, and thus predicting the future peak-shaving performance data of each thermal power unit.

[0048] The method for selecting the sub-peak performance thermal power unit dataset is as follows: the future peak performance data of thermal power units with predicted scores less than F are selected and used as the peak performance item of the sub-peak performance thermal power unit dataset, and the actual operating characteristic value is used as the operating characteristic item of the sub-peak performance thermal power unit dataset, thus forming the sub-peak performance thermal power unit dataset.

[0049] The decision label for each thermal power unit's cluster refers to the fact that each mined associated operational feature set has a corresponding associated operational feature set number as the decision label, and the decision label for the thermal power unit is obtained from the associated operational feature set number corresponding to the thermal power unit's cluster.

[0050] Another aspect of this application is to provide an optimized control system for peak shaving of thermal power units, which can operate an optimized control decision-making method for peak shaving of thermal power units.

[0051] The thermal power unit condition assessment system described in this application includes an operating condition data acquisition module, a peak-shaving performance prediction module, an operating condition pattern clustering module, and a peak-shaving decision mining module.

[0052] The operating condition data acquisition module is used to collect thermal power unit data, including historical operating characteristic sequence sets and corresponding peak-shaving performance index sequences, actual peak-shaving performance index sequences and actual operating characteristic sequence sets. It processes the operating characteristic values ​​in the historical operating characteristic sequence sets and actual operating characteristic sequence sets, as well as the peak-shaving performance data in the corresponding peak-shaving performance index sequences and actual peak-shaving performance index sequences. The operating characteristic values ​​and peak-shaving performance data are then transmitted to the peak-shaving performance prediction module, the operating condition pattern clustering module, and the peak-shaving decision mining module for training, mining, and application. Information collection is achieved through API interfaces and database connections.

[0053] The peak-shaving performance prediction module is used to train a neural network model to predict future peak-shaving performance indicators. The neural network model is used to predict the future peak-shaving performance data of thermal power units. The dataset of thermal power units with predicted peak-shaving performance less than a score F is selected and sent to the operating condition mode clustering module for analysis and clustering of thermal power units.

[0054] The operating condition clustering module clusters thermal power units according to their operating characteristics. Each category of the cluster has similar operating characteristics. The cluster information of each thermal power unit is transmitted to the peak shaving decision mining module.

[0055] The peak shaving decision mining module is used to mine the data of each thermal power unit cluster in the operating condition mode clustering module, find the frequent itemsets of each thermal power unit cluster and set them as the associated operating feature set, assign an associated operating feature set number to each associated operating feature set, and use the associated operating feature set number as the decision label of the thermal power unit cluster.

[0056] The beneficial effects of this invention are:

[0057] This invention discloses an optimized control system and decision-making method for peak shaving of thermal power units. It predicts future peak shaving performance data of thermal power units by training a neural network model, classifies thermal power units based on potential similar characteristics by training a clustering algorithm model, and utilizes factors related to peak shaving performance of thermal power units mined by the system to assist mutual understanding between thermal power units and power operation and maintenance personnel. It also assists thermal power units in learning knowledge online, extracts more useful value from large amounts of data, optimizes data processing and mining effects, extracts subtle internal connections from the data performance of thermal power units, and intervenes promptly in the identified problems.

[0058] To more clearly illustrate the implementation method of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Attached Figure Description

[0059] Figure 1 A flowchart of an optimized control decision-making method for peak shaving of thermal power units provided by the present invention;

[0060] Figure 2 A flowchart of a clustering algorithm for an optimized control decision-making method for peak shaving in thermal power units, provided by the present invention;

[0061] Figure 3 A flowchart illustrating the principle of the Apriori algorithm for an optimized control decision-making method for peak shaving in thermal power units, provided by this invention.

[0062] Figure 4 The present invention provides a system functional block diagram of an optimized control system for peak shaving of thermal power units. Detailed Implementation

[0063] To better understand this application, various aspects of this application will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are merely illustrative of exemplary embodiments of this application and are not intended to limit the scope of this application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0064] It should be understood that expressions such as "comprising," "including," "having," "containing," and / or "comprising" are open-ended rather than closed-ended expressions in this specification, indicating the presence of the stated features, elements, and / or components, but not excluding the presence of one or more other features, elements, components, and / or combinations thereof. Furthermore, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features, not just individual elements in the list. Additionally, when describing embodiments of this application, the word "may" is used to mean "one or more embodiments of this application." And the term "exemplary" is intended to refer to examples or illustrations.

[0065] Unless otherwise specified, all terms used herein (including engineering and technical terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that, unless expressly stated herein, terms defined in common dictionaries shall be interpreted as having the meaning consistent with their meaning in the context of the relevant art, and not as having an idealized or overly formalized meaning.

[0066] This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0067] Example 1

[0068] This application discloses an optimized control decision-making method for peak shaving in thermal power units, see [link to relevant documentation]. Figure 1 Specifically, it includes the following steps:

[0069] Step 11: Obtain the historical operating characteristic sequence set of thermal power units and the corresponding peak-shaving performance index sequence;

[0070] The method for obtaining the historical operating characteristic sequence set and the corresponding peak-shaving performance index sequence of thermal power units is as follows: Set a data acquisition period T, acquire the operating characteristic value of a historical operating characteristic within each data acquisition period as an operating characteristic value of the historical operating characteristic sequence, and acquire peak-shaving performance data once in each data acquisition period. Acquire the operating characteristic value and peak-shaving performance data for t data acquisition periods to obtain the historical operating characteristic sequence set and the corresponding peak-shaving performance index sequence of thermal power units, where t is the number of times the operating characteristic value and peak-shaving performance data are acquired in chronological order.

[0071] The acquisition of the historical operating characteristic sequence set and corresponding peak-shaving performance index sequence of thermal power units refers to:

[0072] Each element in the historical operating feature sequence set is a historical operating feature sequence, wherein each historical operating feature sequence includes a sequence of operating feature values ​​of the corresponding historical operating feature, and the operating feature values ​​in the historical operating feature sequence are arranged in chronological order; the corresponding peak-shaving performance index sequence includes peak-shaving performance data of each thermal power unit for each test, and the peak-shaving performance data in the corresponding peak-shaving performance index sequence are arranged in chronological order.

[0073] The historical operating characteristics refer to relevant operating factors that affect the peak-shaving performance of thermal power units, including but not limited to historical operating duration, historical start-up and shutdown frequency, historical load rate, and historical overload frequency; the operating characteristic values ​​are the specific numerical values ​​of the corresponding historical operating characteristics.

[0074] The historical operating characteristic sequence is collected as follows: between every two tests of each thermal power unit, the corresponding historical operating characteristic value is collected once, and this operating characteristic value is added to the historical operating characteristic sequence in chronological order. For example, thermal power units are tested regularly, and the duration between every two tests is a data acquisition period T. The operating characteristic values ​​of the corresponding historical operating characteristics of the thermal power unit acquired in the t-th test within a data acquisition period are historical operating duration At, historical start-stop count Bt, historical load rate Ct, and historical overload count Dt, respectively. Here, t is the number of times the operating characteristic values ​​and peak-shaving performance data are acquired in chronological order, At is the operating characteristic value of the historical operating duration of the thermal power unit acquired in the t-th test within the data acquisition period, and Bt is the historical overload count Dt. The historical start-stop frequency of thermal power units acquired in the t-th data acquisition period is defined as follows: Ct is the historical load rate of thermal power units acquired in the t-th data acquisition period; Dt is the historical overload frequency of thermal power units acquired in the t-th data acquisition period; the historical operating characteristic sequence of the historical operating duration of thermal power units acquired in the t-th data acquisition period is {A1, A2, ..., At}; the historical operating characteristic sequence of the historical start-stop frequency of thermal power units acquired in the t-th data acquisition period is {B1, B2, ..., Bt}; the historical operating characteristic sequence of the historical load rate of thermal power units acquired in the t-th data acquisition period is {C1, C2, ..., Ct}; and the historical operating characteristic sequence of the historical overload frequency of thermal power units acquired in the t-th data acquisition period is {D1, D2, ..., Dt}.

[0075] The method for collecting the corresponding peak-shaving performance index sequence is as follows: there is one test within each data acquisition period T, and the peak-shaving performance data obtained from the test is used as the peak-shaving performance data corresponding to the historical operating characteristics within the corresponding data acquisition period T. The peak-shaving performance data obtained in the tth time within a data acquisition period T is Et, and the corresponding peak-shaving performance index sequence for collecting the peak-shaving performance indexes t times is {E1, E2, ..., Et}.

[0076] Each peak-shaving performance data in the corresponding peak-shaving performance index sequence corresponds one-to-one with each operating characteristic value in the historical operating characteristic sequence. For example, in collecting t historical operating characteristic sequences and corresponding peak-shaving performance index sequences, the first peak-shaving performance data E1 corresponds to the first operating characteristic values ​​A1, B1, C1, and D1; the second peak-shaving performance data E2 corresponds to the second operating characteristic values ​​A2, B2, C2, and D2; and the t-th peak-shaving performance data Et corresponds to the t-th operating characteristic values ​​At, Bt, Ct, and Dt.

[0077] Step 12: Train a neural network model to predict future peak-shaving performance indicators by using the historical operating characteristic sequence set of thermal power units and the corresponding peak-shaving performance index sequence.

[0078] The step of training a neural network model to predict future peak-shaving performance indicators by using a set of historical operating feature sequences of thermal power units and corresponding peak-shaving performance indicator sequences includes:

[0079] The prediction time step, sliding step, and sliding window length are preset. The historical operating feature sequence and the corresponding peak-shaving performance index sequence are divided into multiple groups of data samples according to the window sliding method. Each group of data samples is used as the input data of the neural network model for predicting the future peak-shaving performance index, and the corresponding peak-shaving performance index sequence of the prediction time step is used as the output. Each group of data samples includes a historical operating feature sequence of one sliding window length and a corresponding peak-shaving performance index sequence of one sliding window length.

[0080] Within the prediction time step of each peak-shaving performance index sequence, the peak-shaving performance index sequence is used as the prediction target, and the minimum error between the predicted peak-shaving performance index sequence and the actual peak-shaving performance index sequence is used as the training target to train the neural network model for predicting future peak-shaving performance indexes.

[0081] Preferably, the neural network model is an RNN neural network model or an LSTM neural network model;

[0082] The prediction time step, sliding step, and sliding window length are preset based on practical experience.

[0083] The prediction time step, sliding step, and sliding window length are defined as follows: Assuming there is a historical operating characteristic sequence of a thermal power unit [1, 2, 3, 2, 4, 5] and a corresponding peak-shaving performance index sequence [10, 20, 15, 30, 25, 15], both sequences have six data acquisition periods; the prediction time step refers to the time interval for predicting the future based on historical data; the sliding step refers to defining a window in the sequence, and the length of the window sliding each time; the sliding window length refers to the size of a window defined in the sequence.

[0084] For example, in the historical operating feature sequence [1, 2, 3, 2, 4, 5] and the corresponding peak-shaving performance index sequence [10, 20, 15, 30, 25, 15], we define a prediction time step of 2, define a sliding window with a length of 3, and define a sliding step of 1. Then the first data sample is {[1, 2, 3], [10, 20, 15]}, and the second data sample is {[2, 3, 2], [20, 15, 30]}. Using the first data sample {[1, 2, 3], [10, 20, 15]}, we can predict the peak-shaving performance of a prediction time step as [30, 25], and using the second data sample {[2, 3, 2], [20, 15, 30]}, we can predict the peak-shaving performance of a prediction time step as [25, 15].

[0085] Step 13: Select the historical sub-peak performance thermal power unit dataset from all thermal power units, and use the historical sub-peak performance thermal power unit dataset to cluster the thermal power units through a clustering algorithm to obtain K thermal power unit clusters and train the clustering algorithm model.

[0086] The method for selecting the historical sub-peak performance thermal power unit dataset from all thermal power units is as follows: The historical operating characteristic sequence and peak-shaving performance index sequence are input into a trained neural network model predicting future peak-shaving performance indicators. The model outputs future peak-shaving performance data for the thermal power units. Thermal power units with scores less than F are designated as sub-peak performance thermal power units. The operating characteristic values ​​of these sub-peak performance thermal power units are used as historical operating characteristic items. The historical sub-peak performance thermal power unit dataset is obtained from the peak-shaving performance index sequence and the historical operating characteristic sequence of these sub-peak performance thermal power units. F is a preset value; it is set according to specific circumstances.

[0087] like Figure 2 As shown, the specific method for clustering thermal power units using a clustering algorithm to obtain K clusters of thermal power units and training the clustering algorithm model is as follows:

[0088] Step 1301: Obtain thermal power unit data points by using the historical operating characteristic sequence and corresponding peak-shaving performance index sequence of thermal power units in the historical sub-peak-shaving performance thermal power unit dataset; wherein, each thermal power unit data point contains the historical operating characteristic sequence of the corresponding thermal power unit.

[0089] Step 1302: Randomly select K1 data points from thermal power units as initial cluster centers Xi; the initial cluster centers Xi are randomly selected data points from thermal power units.

[0090] Step 1303: Calculate the Euclidean distance from the data points of the thermal power units other than the initial cluster center to each initial cluster center, and assign each data point of the thermal power unit to the cluster of the thermal power units containing the nearest initial cluster center;

[0091] Step 1304: Calculate the average value of each operating characteristic value of the thermal power unit data points within each thermal power unit cluster, form a new thermal power unit data point from the average value of each operating characteristic value, and use the new thermal power unit data point as the new cluster center Xi' of the thermal power unit cluster;

[0092] Step 1305: Recalculate the Euclidean distance from each thermal power unit data point to the new cluster center, and reassign each thermal power unit data point to the thermal power unit cluster where the nearest new cluster center Xi' is located;

[0093] Step 1306: Repeat steps 1304-1305 until the clusters to which all thermal power unit data points are assigned no longer change, resulting in K thermal power unit clusters. The clustering algorithm model training is complete, where K=K1.

[0094] Step 14: Based on the K thermal power unit clusters, obtain L thermal power unit datasets. Use data mining algorithms to mine the L thermal power unit datasets to obtain the frequent itemsets of each thermal power unit dataset; K=L;

[0095] The method for obtaining L thermal power unit datasets based on K thermal power unit clusters is as follows: the thermal power unit data in the K thermal power unit clusters exist in the form of thermal power unit data points. The operating characteristic value and peak-shaving performance index of each thermal power unit data point constitute a thermal power unit sample data in the thermal power unit dataset. The operating characteristic value and peak-shaving performance data of the thermal power unit corresponding to each thermal power unit data point constitute a set of thermal power unit data, thus obtaining L thermal power unit datasets, where K=L.

[0096] The method for using data mining algorithms to mine L thermal power unit datasets and obtain the frequent itemsets of each thermal power unit dataset is as follows:

[0097] The data mining algorithm is the Apriori algorithm. The Apriori algorithm is used to mine frequent itemsets for each thermal power unit dataset. Preferably, the minimum support threshold M can be 0.4.

[0098] The steps of the Apriori algorithm to mine frequent itemsets for each thermal power unit dataset and obtain the associated operational feature set are detailed below. Figure 3 The steps are as follows:

[0099] Step 1401: Based on actual needs and the characteristics of the thermal power unit dataset, set a minimum support threshold M. Preferably, the minimum support threshold M can be 0.4.

[0100] Step 1402: Set counter j, with an initial value of 1, scan the thermal power unit dataset, calculate the frequency of each peak-shaving performance data and historical operating characteristic in the thermal power unit dataset, and generate the j-th candidate set G(j);

[0101] Step 1403: Scan the j-th candidate itemset G(j); label the support as S, filter out candidate items whose support S is greater than or equal to the minimum support threshold M, and generate the j-th frequent itemset H(j); it should be noted that the method of scanning the candidate itemset is a conventional technique in the art, and will not be described in detail here.

[0102] Step 1404: Perform join and pruning operations on the j-th frequent itemset H(j), update the value of j to j+1, and generate the (j+1)-th candidate itemset G(j+1); It should be noted that the method of joining and pruning the j-th frequent itemset is a conventional technique in the art, and will not be described in detail here.

[0103] Step 1405: Scan the (j+1)th candidate itemset G(j+1), count the number of candidates whose support S is greater than or equal to the minimum support threshold M, and generate the (j+1)th frequent itemset H(j+1).

[0104] Step 1406: Repeat steps 1404 to 1405 to continuously generate new candidate item sets. Statistically count the candidate items whose support is greater than or equal to the minimum support threshold to obtain new frequent itemsets until the generated candidate item sets are empty.

[0105] j represents the number of the candidate itemset or frequent itemset. The j-th candidate itemset generates the j-th frequent itemset, the j-th frequent itemset generates the (j+1)-th candidate itemset, the (j+1)-th candidate itemset generates the (j+2)-th frequent itemset, and so on, until the generated candidate itemset is empty.

[0106] Step 1407: Set the obtained frequent itemsets as the associated running feature set, and sort the frequent itemsets in the associated running feature set from largest to smallest according to the weight of the frequent items in the frequent itemsets;

[0107] Step 1408: Number all associated running feature sets, and use the number of the associated running feature set as the decision label of the thermal power unit cluster in which the corresponding thermal power unit dataset belongs.

[0108] It should be noted that using the Apriori algorithm to mine frequent itemsets from L thermal power unit datasets is a conventional data mining technique. This invention will not elaborate on the underlying principles, but for ease of implementation, the following examples of frequent itemset mining are provided: As shown in Tables 1-7, assuming a thermal power unit dataset (see Table 1), frequent itemsets between peak-shaving performance data and operational characteristics are mined, with a minimum support threshold of 0.4.

[0109] Table 1 Dataset of Thermal Power Units

[0110]

[0111] Where A and B represent the standardized values ​​of the peak-shaving performance data of the thermal power unit, P and Q represent the standardized values ​​of the operating characteristic 1 of the thermal power unit, and E and f represent the standardized values ​​of the operating characteristic 2 of the thermal power unit.

[0112] The first candidate set is shown in Table 2;

[0113] Table 2 Candidate Set 1

[0114]

[0115] The first frequent itemset is obtained by filtering items whose support is greater than or equal to the minimum support threshold of 0.4, as shown in Table 3.

[0116] Table 3. First Frequent Itemset

[0117]

[0118] The second candidate set was obtained by performing connection and pruning, as shown in Table 4;

[0119] Table 4. Candidate Set 2

[0120]

[0121] Where {A, Q} represents the item with peak shaving performance data of A and operating characteristic 1 of Q, {A, E} represents the item with peak shaving performance data of A and operating characteristic 2 of E, {B, Q} represents the item with peak shaving performance data of B and operating characteristic 1 of Q, and {B, E} represents the item with peak shaving performance data of B and operating characteristic 2 of E.

[0122] The second frequent itemset is obtained by filtering items whose support is greater than or equal to the minimum support threshold of 0.4, as shown in Table 5.

[0123] Table 5. Second Frequent Itemset

[0124]

[0125] The third candidate set was obtained by performing connection and pruning, as shown in Table 6;

[0126] Table 6. Candidate Set 3

[0127]

[0128] Where {A, Q, E} represents a term with peak-shaving performance data of A, operating characteristic 1 of Q, and operating characteristic 2 of E;

[0129] The third frequent itemset is obtained by filtering items whose support is greater than or equal to the minimum support threshold of 0.4, as shown in Table 7.

[0130] Table 7. Frequent Itemset 3

[0131]

[0132] The mining process ends when no more candidate sets can be generated.

[0133] Step 15: Set the frequent itemsets as the associated operation feature set, and set a decision label for each thermal power unit cluster based on the associated operation feature set;

[0134] The decision label for each thermal power unit's cluster refers to the fact that each mined associated operational feature set has a corresponding associated operational feature set number as the decision label, and the decision label for the thermal power unit is obtained from the associated operational feature set number corresponding to the thermal power unit's cluster.

[0135] Step 16: Obtain the actual peak-shaving performance index sequence and the actual operating characteristic sequence set, input them into the neural network model to predict the future peak-shaving performance of the thermal power unit, and obtain the future peak-shaving performance data of the thermal power unit.

[0136] The actual peak-shaving performance index sequence refers to the actual peak-shaving performance data of the thermal power unit acquired t times according to the data acquisition period T, and the actual peak-shaving performance indexes are combined according to the acquisition time order to obtain the corresponding peak-shaving performance index sequence.

[0137] The actual operating feature sequence set refers to a set composed of multiple actual operating feature sequences. The actual operating feature sequence is obtained by combining actual operating feature values ​​in chronological order. The method for obtaining the actual operating feature values ​​is as follows: within a data acquisition period T, the operating feature value between two tests is obtained as an actual operating feature value. The actual operating feature value is obtained t times to obtain an actual operating feature sequence containing t actual operating feature values. The operating feature values ​​of multiple actual operating features are obtained to obtain multiple actual operating feature sequences, which together constitute the actual operating feature sequence set.

[0138] Step 17: Based on the future peak-shaving performance data of thermal power units, select the sub-peak-shaving performance thermal power unit dataset, input each peak-shaving performance index and operating characteristic in the sub-peak-shaving performance thermal power unit dataset into the clustering algorithm model, and obtain the decision label of the thermal power unit cluster to which each thermal power unit belongs.

[0139] The method for selecting the sub-peak performance thermal power unit dataset is as follows: the future peak performance data of thermal power units with predicted scores less than F are selected and used as the peak performance item of the sub-peak performance thermal power unit dataset, and the actual operating characteristic value is used as the operating characteristic item of the sub-peak performance thermal power unit dataset, thus forming the sub-peak performance thermal power unit dataset.

[0140] The steps for predicting the future peak-shaving performance of thermal power units are as follows: inputting the actual set of operating characteristic sequences of each thermal power unit and the actual sequence of peak-shaving performance indicators of each thermal power unit into a trained neural network model for predicting the future peak-shaving performance data of thermal power units, and thus predicting the future peak-shaving performance data of each thermal power unit.

[0141] Step 18: Based on the decision labels of the cluster of thermal power units where the sub-peak-shaving performance thermal power units are located, analyze the associated operating characteristics, and optimize and adjust the peak-shaving performance of the thermal power units accordingly based on the corresponding characteristics.

[0142] Example 2

[0143] For ease of understanding, the following provides a detailed description of an optimized control system for peak shaving in thermal power units, based on an embodiment of this application. It should be noted that this embodiment is an improvement upon Embodiment 1; the similarities to Embodiment 1 will not be elaborated here. See [link to relevant documentation]. Figure 4 An optimized control system for peak shaving in thermal power units includes an operating condition data acquisition module, a peak shaving performance prediction module, an operating condition pattern clustering module, and a peak shaving decision mining module.

[0144] The operating condition data acquisition module is used to collect thermal power unit data, including historical operating characteristic sequence sets and corresponding peak-shaving performance index sequences, actual peak-shaving performance index sequences and actual operating characteristic sequence sets. It processes the operating characteristic values ​​in the historical operating characteristic sequence sets and actual operating characteristic sequence sets, as well as the peak-shaving performance data in the corresponding peak-shaving performance index sequences and actual peak-shaving performance index sequences. The operating characteristic values ​​and peak-shaving performance data are then transmitted to the peak-shaving performance prediction module, the operating condition pattern clustering module, and the peak-shaving decision mining module for training, mining, and application. Information collection is achieved through API interfaces and database connections.

[0145] The peak-shaving performance prediction module is used to train a neural network model to predict future peak-shaving performance indicators. The neural network model is used to predict the future peak-shaving performance data of thermal power units. The dataset of thermal power units with predicted peak-shaving performance less than a score F is selected and sent to the operating condition mode clustering module for analysis and clustering of thermal power units.

[0146] The operating condition clustering module clusters thermal power units according to their operating characteristics. Each category of the cluster has similar operating characteristics. The cluster information of each thermal power unit is transmitted to the peak shaving decision mining module.

[0147] The peak shaving decision mining module is used to mine the data of each thermal power unit cluster in the operating condition mode clustering module, find the frequent itemsets of each thermal power unit cluster and set them as the associated operating feature set, assign an associated operating feature set number to each associated operating feature set, and use the associated operating feature set number as the decision label of the thermal power unit cluster.

[0148] For example, the historical peak-shaving performance index sequence and historical operating characteristic sequence set of all thermal power units are collected through the operating condition data acquisition module. The processed data is sent to the peak-shaving performance prediction module. The historical peak-shaving performance index sequence and historical operating characteristic sequence set are trained to obtain a neural network model that can predict the future peak-shaving performance data of thermal power units. The future peak-shaving performance data of thermal power units predicted by the neural network model are used to filter out the sub-peak-shaving performance thermal power unit dataset. The sub-peak-shaving performance thermal power unit dataset is sent to the clustering algorithm module to cluster the thermal power units in the sub-peak-shaving performance thermal power unit dataset, resulting in multiple thermal power unit clusters. The thermal power unit clusters are sent to the peak-shaving decision mining module. The frequent itemsets in each thermal power unit cluster are mined through the data mining algorithm to obtain the associated operating characteristic set. The associated operating characteristic set is numbered and the corresponding associated operating characteristic set number is used as the decision label of the thermal power unit cluster.

[0149] The information acquisition module obtains the current set of actual operating characteristic sequences and the corresponding peak-shaving performance index sequences of thermal power units. These are then input into a neural network model for training to predict future peak-shaving performance indicators, resulting in future peak-shaving performance data for thermal power units. A dataset of sub-peak-shaving performance thermal power units is then selected. The operating condition pattern clustering module categorizes each thermal power unit into its respective cluster. The decision labels of the thermal power unit clusters obtained by the peak-shaving decision mining module allow thermal power unit operators and power maintenance personnel to identify these labels, thereby obtaining the associated operating characteristic set of the thermal power units and enabling intervention and handling of related issues.

Claims

1. An optimal control decision method for peak shaving of a thermal power unit, characterized in that, The method comprises the following steps: obtaining a historical operation characteristic sequence set of a thermal power unit and a corresponding peak regulation performance index sequence; training the historical operation characteristic sequence set of the thermal power unit and the corresponding peak regulation performance index sequence to train a neural network model for predicting future peak regulation performance indexes; the neural network model is an RNN neural network model or an LSTM neural network model; selecting a historical sub-peak regulation performance thermal power unit dataset from all thermal power units, clustering the thermal power units by using the historical sub-peak regulation performance thermal power unit dataset through a clustering algorithm to obtain K thermal power unit clustering clusters, and training a clustering algorithm model; obtaining L thermal power unit datasets according to the K thermal power unit clustering clusters, mining the L thermal power unit datasets by using a data mining algorithm to obtain a frequent item set of each thermal power unit dataset; K = L; setting the frequent item set as a related operation characteristic set, and setting a decision label for each thermal power unit clustering cluster according to the related operation characteristic set; obtaining an actual peak regulation performance index sequence and an actual operation characteristic sequence set, inputting the neural network model to predict future peak regulation performance indexes, and obtaining future peak regulation performance data of the thermal power unit; selecting a sub-peak regulation performance thermal power unit dataset according to the future peak regulation performance data of the thermal power unit, inputting each peak regulation performance index and operation characteristic in the sub-peak regulation performance thermal power unit dataset into the clustering algorithm model, and obtaining a decision label of a thermal power unit clustering cluster where each thermal power unit is located.

2. The method for optimal control decision of peak regulation of thermal power units according to claim 1, characterized in that, Each element in the historical operation characteristic sequence set is a historical operation characteristic sequence, wherein each historical operation characteristic sequence comprises a sequence composed of operation characteristic values of corresponding historical operation characteristics, and the operation characteristic values in the historical operation characteristic sequence are arranged in chronological order; the corresponding peak regulation performance index sequence comprises peak regulation performance data of each test of each thermal power unit, and the peak regulation performance data in the corresponding peak regulation performance index sequence are arranged in chronological order.

3. The method for optimal control decision of peak regulation of thermal power units according to claim 1, characterized in that, The specific method for training the neural network model for predicting future peak regulation performance indexes by using the historical operation characteristic sequence set and the corresponding peak regulation performance index sequence of the thermal power unit is as follows: predetermining a prediction time step, a sliding step and a sliding window length, dividing the historical operation characteristic sequence and the corresponding peak regulation performance index sequence into multiple data samples according to a window sliding method, taking each data sample as input data of the neural network model for predicting future peak regulation performance indexes, and taking the corresponding peak regulation performance index sequence of the prediction time step as output, wherein each data sample comprises a historical operation characteristic sequence of the sliding window length and a corresponding peak regulation performance index sequence of the sliding window length; in the prediction time step of each peak regulation performance index sequence, taking the peak regulation performance index sequence as a prediction target, taking the minimum error between the predicted peak regulation performance index sequence and the actual peak regulation performance index sequence as a training target, and training the neural network model for predicting future peak regulation performance indexes.

4. The method of claim 1, wherein, The specific method of screening the historical sub-peak-regulation performance thermal power unit dataset from all thermal power units comprises: inputting the historical operation characteristic sequence and the peak-regulation performance index sequence into the trained neural network model for predicting future peak-regulation performance indexes, outputting the future peak-regulation performance data of the thermal power unit, setting the thermal power unit with a peak-regulation performance index less than F as a sub-peak-regulation performance thermal power unit, taking the operation characteristic value of the sub-peak-regulation performance thermal power unit as a historical operation characteristic item, and obtaining the historical sub-peak-regulation performance thermal power unit dataset from the peak-regulation performance index sequence and the historical operation characteristic sequence of the sub-peak-regulation performance thermal power unit; F is a preset value.

5. The method for optimal control decision of peak regulation of thermal power units according to claim 1, characterized in that, The specific method of clustering the thermal power units by using the clustering algorithm to obtain K thermal power unit clustering clusters and training the clustering algorithm model comprises the following steps: Step 1301: obtaining the thermal power unit data points from the historical operation characteristic sequence and the corresponding peak-regulation performance index sequence of the thermal power units in the historical sub-peak-regulation performance thermal power unit dataset; Step 1302: randomly selecting K1 thermal power unit data points as initial clustering centers Xi; Step 1303: calculating the Euclidean distance between the remaining thermal power unit data points except the initial clustering centers and each initial clustering center, and dividing each thermal power unit data point into the thermal power unit clustering cluster where the nearest initial clustering center is located; Step 1304: calculating the average value of each operation characteristic value of the thermal power unit data points in each thermal power unit clustering cluster, composing the average value of each operation characteristic value into a new thermal power unit data point, and taking the new thermal power unit data point as the new clustering center Xi' of the thermal power unit clustering cluster; Step 1305: recalculating the Euclidean distance between each thermal power unit data point and the new clustering center, and re-dividing each thermal power unit data point into the thermal power unit clustering cluster where the nearest new clustering center Xi' is located; Step 1306: repeating steps 1304-1305 until the thermal power unit clustering cluster to which all the thermal power unit data points are divided no longer changes, obtaining K thermal power unit clustering clusters, and completing the training of the clustering algorithm model, wherein K=K1.

6. The method for optimal control decision of peak regulation of thermal power units according to claim 1, characterized in that, The method of obtaining L thermal power unit datasets from the K thermal power unit clustering clusters comprises: The thermal power unit data in the K thermal power unit clustering clusters exist in the form of thermal power unit data points, the operation characteristic value in each thermal power unit data point and the peak-regulation performance index corresponding to each thermal power unit data point are a thermal power unit sample data in the thermal power unit dataset, and the operation characteristic value and the peak-regulation performance data of the thermal power unit corresponding to each thermal power unit data point are a group of thermal power unit data, thereby obtaining L thermal power unit datasets, and K=L.

7. The method for optimal control decision of peak regulation of thermal power units according to claim 1, characterized in that, The method of mining the L thermal power unit datasets by using the data mining algorithm to obtain the frequent item set of each thermal power unit dataset comprises the following steps: Step 1401: setting a minimum support threshold M according to actual requirements and the characteristics of the thermal power unit dataset; Step 1402: setting a counter j, the initial value of j is 1, scanning the thermal power unit dataset, calculating the frequency of each peak-regulation performance data and historical operation characteristic in the thermal power unit dataset, and generating the jth candidate item set G(j). Step 1403: scanning the jth candidate item set G(j); marking the support degree as S, screening the candidate items with the support degree S greater than or equal to the minimum support threshold M, and generating the jth frequent item set H(j); Step 1404: performing connection and pruning operations on the jth frequent item set H(j), updating the value of j to j+1, and generating the j+1th candidate item set G(j+1); Step 1405: scanning the j+1th candidate item set G(j+1), counting the candidate items with the support degree S greater than or equal to the minimum support threshold M, and generating the j+1th frequent item set H(j+1); Step 1406: repeating steps 1404-1405, constantly generating new candidate item sets, and obtaining new frequent item sets by counting the candidate items with the support degree greater than or equal to the minimum support threshold, until the generated candidate item set is empty.

8. The method for optimal control decision of peak regulation of thermal power units according to claim 1, characterized in that, The method for setting a decision label for each thermal power unit cluster is: According to the obtained frequent item set, set an associated operation feature set, and sort the frequent item sets in the associated operation feature set according to the weights of the frequent items in the frequent item sets from large to small; Assign an associated operation feature set number to all the associated operation feature sets, and use the associated operation feature set number as the decision label of the thermal power unit cluster where the corresponding thermal power unit data set is located.

9. The method for optimal control decision of peak regulation of thermal power units according to claim 1, characterized in that, The step of predicting the future peak shaving performance of the thermal power unit is: inputting the obtained actual operation feature sequence set of each thermal power unit and the actual peak shaving performance index sequence into the neural network model trained for predicting the future peak shaving performance data of the thermal power unit, and predicting the future peak shaving performance data of each thermal power unit.

10. An optimal control system for peak load regulation of a thermal power unit, the system being used to implement the optimal control decision-making method for peak load regulation of a thermal power unit according to any one of claims 1-9, characterized in that, Comprise: A working condition data acquisition module, a peak shaving performance prediction module, a working condition mode clustering module, and a peak shaving decision mining module; The working condition data acquisition module is used to collect thermal power unit data, wherein the thermal power unit data comprises a historical operation feature sequence set and a corresponding peak shaving performance index sequence, an actual peak shaving performance index sequence and an actual operation feature sequence set, and the operation feature values in the historical operation feature sequence set and the actual operation feature sequence set and the peak shaving performance data in the corresponding peak shaving performance index sequence and the actual peak shaving performance index sequence are processed, and the operation feature values and the peak shaving performance data are transmitted to the peak shaving performance prediction module, the working condition mode clustering module, and the peak shaving decision mining module for training, mining, and application; information collection is realized through an API interface and a connected database; The peak shaving performance prediction module is used to train a neural network model for predicting future peak shaving performance indexes, to predict the future peak shaving performance data of the thermal power unit by using the neural network model, to screen out sub-peak shaving performance thermal power unit data sets with a predicted peak shaving performance less than a score F, and to transmit the sub-peak shaving performance thermal power unit data sets to the working condition mode clustering module for analysis and clustering of the thermal power unit; The working condition mode clustering module is used to cluster the thermal power units according to the operation features of the thermal power units, and to transmit the information of each thermal power unit cluster in the clustering result to the peak shaving decision mining module; and The peak shaving decision mining module is used to set a decision label for each thermal power unit cluster according to the clustering result, to obtain an associated operation feature set according to the frequent item set, to sort the frequent item sets in the associated operation feature set according to the weights of the frequent items in the frequent item sets from large to small, and to assign an associated operation feature set number to all the associated operation feature sets, and use the associated operation feature set number as the decision label of the thermal power unit cluster where the corresponding thermal power unit data set is located. The peak regulation decision mining module is configured to mine data of each thermal power generating unit cluster in the working condition mode clustering module, find a frequent item set of each thermal power generating unit cluster, set the frequent item set as an associated operation feature set, assign a cluster number to each associated operation feature set, and use the cluster number as a decision tag of the thermal power generating unit cluster.

Images