Identification method of factors affecting power generation capacity of renewable energy farms based on dimensionality reduction analysis

Abstract

The present invention discloses a method for identifying factors affecting the power generation capacity of a new energy field group based on dimensionality reduction analysis, and relates to fields related to new energy power generation. The method includes: establishing an intelligent agent; collecting data of power generation units to establish a time series data set; performing data preprocessing, performing data cycle evaluation, and establishing long and short time scale windows; generating a backtracking instruction, performing historical data backtracking reading, updating the time series data preprocessing results, and performing time scale feature extraction; performing spatial scale feature extraction, performing feature fusion on the time scale feature extraction results and the spatial scale feature extraction results, and establishing a feature vector; performing feature dimensionality reduction, generating a collaborative dimensionality reduction strategy, performing feature dimensionality reduction optimization, establishing feature dimensionality reduction results, and establishing key influencing factors. The method solves the technical problems of low recognition efficiency and accuracy in the identification of factors affecting the power generation capacity of existing new energy fields, and achieves the technical effect of realizing efficient and accurate identification of influencing factors.

Description

Technical Field

[0001] The present application relates to the field of renewable energy power generation, and in particular to a method for identifying factors affecting the power generation capacity of renewable energy field groups based on dimensionality reduction analysis. Background Art

[0002] With the continuous growth of global demand for clean energy, the prediction and management of the power generation capacity of new energy clusters (such as wind farms, photovoltaic power stations, etc.) as important renewable energy supply bases has become the key to ensure the stable operation of the power system. The power generation capacity of new energy clusters is affected by many factors, including meteorological conditions (such as wind speed, wind direction, solar radiation intensity, etc.), equipment status (such as equipment health, operating efficiency, etc.) and grid dispatching requirements. However, the complex interactions and highly nonlinear relationships between these factors make the prediction of the power generation capacity of new energy clusters face huge challenges. At present, the identification of factors affecting the power generation capacity of new energy clusters mainly adopts methods based on complex statistical models and machine learning algorithms. These methods attempt to reveal the potential impact of multiple factors such as meteorological conditions, equipment status and grid operation status on the power generation capacity of new energy by analyzing a large amount of historical data. However, due to the multi-dimensional and high-frequency characteristics of the data generated by new energy clusters, these methods are prone to fall into the "dimensional disaster" in practical applications, resulting in low computational efficiency and even affecting the accuracy of the prediction results.

[0003] In the current relevant technologies, the identification of factors affecting the power generation capacity of renewable energy farms has technical problems such as low recognition efficiency and accuracy. Summary of the invention

[0004] The present application provides a method for identifying factors affecting the power generation capacity of a new energy farm group based on dimensionality reduction analysis. It collects multi-dimensional data through an intelligent agent system, and uses time series data analysis technology to pre-process and periodically evaluate the data. Then, it reduces the data dimension through an adaptive dimensionality reduction strategy to improve computing efficiency. Finally, it uses collaborative dimensionality reduction optimization technology to achieve information sharing and collaborative optimization between different power generation units, thereby directly outputting key factors affecting the power generation capacity of a new energy farm group and other technical means, thereby achieving the technical effect of achieving efficient and accurate identification of factors affecting the power generation capacity of a new energy farm group.

[0005] This application provides a method for identifying factors affecting the power generation capacity of a new energy farm group based on dimensionality reduction analysis, including:

[0006] An intelligent agent mapped to a power generation unit is established, the intelligent agent includes a perception module, an analysis module, and a decision module, and the intelligent agent is configured with a central server for common communication and interaction; data of the power generation unit is collected through the perception module to establish a time series data set, the time series data set includes meteorological data, equipment status data, power generation output data, and the time series data set is the time series data of each power generation unit in the new energy field group; the time series data set collected by the perception module is received through the analysis module, data preprocessing is performed, a time series data preprocessing result is established, and a data cycle evaluation is performed on the time series data preprocessing result, and a long and short time scale window is established based on the data cycle evaluation result; a backtracking instruction is generated based on the long and short time scale windows, and a historical data backtracking of the power generation unit is performed based on the backtracking instruction Retrospective reading: update the time series data preprocessing results according to the retrospective reading results of historical data, and extract the time scale features of the updated time series data preprocessing results based on the long and short time scale windows to establish the time scale feature extraction results; perform spatial scale feature extraction on the time series data preprocessing results to establish the spatial scale feature extraction results, send the time scale feature extraction results and the spatial scale feature extraction results to the decision module, perform feature fusion, and establish a feature vector; configure an adaptive dimensionality reduction strategy, perform feature dimensionality reduction based on the adaptive dimensionality reduction strategy and the feature vector, generate a collaborative dimensionality reduction strategy for shared dimensionality reduction based on the central server, optimize the feature dimensionality reduction based on the collaborative dimensionality reduction strategy, establish the feature dimensionality reduction results, and establish key influencing factors according to the feature dimensionality reduction results.

[0007] In a possible implementation, the preprocessing result of the time series data is evaluated for a data period, and a long and short time scale window is established based on the data period evaluation result, and the following processing is performed:

[0008] The time series data preprocessing results are subjected to detrending and differential processing to generate a processed data set; the processed data set is subjected to autocorrelation function calculation as follows: ;in, Characterize the time interval, Characterize the hysteresis time interval The autocorrelation function value under Characterization time The data value below, Representation in time The data value below, Characterize the data mean of the processed data set, The total length of the time series in the processing data set is determined according to the calculation result of the autocorrelation function, a periodic peak hysteresis point is established with the periodic peak hysteresis point, and a short time scale window is established with the reference period.

[0009] In a possible implementation, the short time scale window is established based on the reference period, and the following processing is performed:

[0010] The local coefficient of variation of the data is calculated based on the reference period; a dynamic short-time window is established through the local coefficient of variation and the reference period, and the dynamic short-time window is used as a short-time scale window.

[0011] In a possible implementation, after the short time scale window is established with the reference period, the following processing is performed:

[0012] The time series data preprocessing result is converted into frequency domain, and the main frequency component is analyzed based on the frequency domain conversion result, as follows: ;in, is the frequency domain signal, is the time domain signal, is the time variable, Characterize the frequency, is an imaginary unit; the reciprocal of the main frequency component is taken as the dominant period, and the dominant period is periodically expanded to establish a long time scale window, and a long and short time scale window is established according to the long time scale window and the short time scale window.

[0013] In a possible implementation, the collaborative dimensionality reduction strategy based on the central server for generating shared dimensionality reduction is used, feature dimensionality reduction optimization is performed based on the collaborative dimensionality reduction strategy, a feature dimensionality reduction result is established, and the following processing is performed:

[0014] After each intelligent agent performs preliminary dimensionality reduction, the corresponding dimensionality reduction result is sent to the central server; the evaluation weights of the dimensionality reduction effect are configured on the central server, and the evaluation weights include reconstruction error weights, information entropy weights, and prediction performance weights; the dimensionality reduction effect of the dimensionality reduction result is evaluated based on the evaluation weights, and a collaborative dimensionality reduction strategy for the intelligent agents that have established shared dimensionality reduction is configured according to the dimensionality reduction effect evaluation results.

[0015] In a possible implementation, the collaborative dimension reduction strategy of the intelligent agent for shared dimension reduction is configured according to the dimension reduction effect evaluation result, and the following processing is performed:

[0016] Acquire the equipment information of the power generation unit corresponding to the intelligent agent, perform similarity evaluation on the equipment parameters according to the equipment information, and establish a first similarity evaluation result; perform meteorological similarity evaluation according to the meteorological data of the power generation unit, and establish a second similarity evaluation result; read the equipment working data of the power generation unit, perform similarity evaluation according to the equipment working data, and establish a third similarity evaluation result; complete the establishment of shared dimensionality reduction of the intelligent agent according to the first similarity evaluation result, the second similarity evaluation result, and the third similarity evaluation result.

[0017] In a possible implementation, the data preprocessing is performed by performing the following processing:

[0018] An isolation forest model is constructed, and random data segmentation is performed in a time series data set based on the isolation forest model; outliers are identified based on the random data segmentation result, and data are removed according to the outlier scores to complete data preprocessing.

[0019] In a possible implementation, the following processing is performed:

[0020] Establish a fallback node for dimensionality reduction; when performing feature dimensionality reduction optimization based on the collaborative dimensionality reduction strategy, if the next round of dimensionality reduction results fail to meet the expected dimensionality reduction effect, generate a fallback instruction; perform feature dimensionality reduction reconstruction according to the fallback instruction and the fallback node.

[0021] The method for identifying factors affecting the power generation capacity of a new energy field group based on dimensionality reduction analysis proposed in the present application first establishes an intelligent agent mapped to the power generation unit, the intelligent agent includes a perception module, an analysis module, and a decision module, the intelligent agent is configured with a central server for common communication and interaction, and then collects data from the power generation unit through the perception module to establish a time series data set, the time series data set includes meteorological data, equipment status data, and power generation output data, the time series data set is the time series data of each power generation unit in the new energy field group, then receives the time series data set collected by the perception module through the analysis module, performs data preprocessing, establishes a time series data preprocessing result, and performs data cycle evaluation on the time series data preprocessing result, establishes a long and short time scale window based on the data cycle evaluation result, and then generates a backtracking instruction based on the long and short time scale window, and performs a backtracking instruction based on the historical data of the power generation unit. Retrospective reading, update the time series data preprocessing results according to the retrospective reading results of historical data, and extract the time scale features of the updated time series data preprocessing results based on the long and short time scale windows to establish the time scale feature extraction results, and perform spatial scale feature extraction on the time series data preprocessing results to establish the spatial scale feature extraction results, send the time scale feature extraction results and the spatial scale feature extraction results to the decision module, perform feature fusion, establish feature vectors, and finally configure an adaptive dimensionality reduction strategy, perform feature dimensionality reduction based on the adaptive dimensionality reduction strategy and feature vectors, and generate a collaborative dimensionality reduction strategy for shared dimensionality reduction based on the central server, optimize feature dimensionality reduction based on the collaborative dimensionality reduction strategy, establish feature dimensionality reduction results, and establish key influencing factors based on the feature dimensionality reduction results, thereby achieving the technical effect of realizing efficient and accurate identification of factors affecting the power generation capacity of new energy field groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the method according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0023] Figure 1 A flow chart of a method for identifying factors affecting the power generation capacity of a new energy farm group based on dimensionality reduction analysis provided in an embodiment of the present application.

[0024] Figure 2 A schematic diagram of a process for establishing intelligent agent shared dimensionality reduction in a method for identifying factors affecting power generation capacity of a new energy farm group based on dimensionality reduction analysis provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0026] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0027] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0028] The present application embodiment provides a method for identifying factors affecting the power generation capacity of a new energy farm group based on dimensionality reduction analysis, such as Figure 1 As shown, the method includes:

[0029] Step S100, establish an intelligent agent mapped to the power generation unit, the intelligent agent includes a perception module, an analysis module, and a decision module, and the intelligent agent is configured with a central server for common communication and interaction. Specifically, the intelligent agent is a software entity that can autonomously perform tasks, and is used to monitor, analyze and optimize the power generation capacity of the new energy power generation unit. Determine the functions and interfaces of the three modules of the intelligent agent (perception module, analysis module, and decision module), wherein the perception module is used to collect various data of the power generation unit; the analysis module is used to process and analyze the collected data; and the decision module is used to make decisions based on the analysis results. Deploy the intelligent agent software to each power generation unit of the new energy field group to ensure that each power generation unit independently runs an intelligent agent instance. Set up a central server to receive data and instructions from the intelligent agents of each power generation unit to achieve unified management and communication interaction.

[0030] Step S200, collect data of the power generation unit through the perception module to establish a time series data set, the time series data set includes meteorological data, equipment status data, power generation output data, and the time series data set is the time series data of each power generation unit in the new energy field group. Specifically, according to the characteristics of the new energy power generation unit, it is determined that the type of data to be collected includes meteorological data, equipment status data and power generation output data, and the data collection frequency is set according to the real-time requirements of the data. Collect data in real time through sensors, database interfaces, etc., arrange the collected data in chronological order, and obtain a time series data set, which is used to analyze the trend and law of data changes over time.

[0031] Step S300, receiving the time series data set collected by the perception module through the analysis module, performing data preprocessing, establishing the time series data preprocessing results, and performing data cycle evaluation on the time series data preprocessing results, and establishing long and short time scale windows based on the data cycle evaluation results. Specifically, data preprocessing is performed on the time series data set, including removing noise data such as outliers and missing values, improving data quality, and converting data of different dimensions into the same dimension for data standardization. Statistical analysis is used to identify the periodic characteristics in the preprocessed data, that is, to evaluate whether there are periodic changes in the data, and to determine the division criteria of long and short time scales based on the length of the cycle.

[0032] In a possible implementation, the data preprocessing is performed, and step S300 further includes step S310, constructing an isolation forest model, and performing random data segmentation in the time series data set based on the isolation forest model. Specifically, the isolation forest algorithm is used for data preprocessing, the isolation forest model is initialized, and the isolation forest model is trained using the time series data set. The model simulates the distribution of data by constructing multiple isolated trees. Each isolated tree recursively segments the data by randomly selecting features and segmentation values ​​until a preset stop condition is reached. Step S320, outliers are identified based on the random data segmentation results, and data are removed according to the outlier score to complete data preprocessing. Specifically, the trained isolation forest model is used to calculate the outlier score of each data point in the time series data set. The higher the score, the more likely the data point is an outlier. According to business requirements and data characteristics, a threshold is set to distinguish between normal points and outliers, and the outlier score is compared with the threshold. Data points with scores higher than the threshold are regarded as outliers. The identified outliers are removed from the time series data set to complete the data preprocessing process. This implementation method uses algorithms such as isolation forests to eliminate data, which can better adapt to the complexity and uncertainty of new energy fields and improve the accuracy and reliability of data by eliminating outliers.

[0033] In one possible implementation, the time series data preprocessing results are subjected to data period evaluation, and long and short time scale windows are established based on the data period evaluation results. Step S300 further includes step S330, which performs detrending and differential processing on the time series data preprocessing results to generate a processed data set. Specifically, by fitting a nonlinear trend model, and then subtracting this trend model from the original data to obtain detrended data, the time series data preprocessing results are subjected to a detrending operation to eliminate long-term trend components in the data. Next, the detrended data are subjected to differential processing to eliminate seasonal or cyclical components in the data, while retaining the changing trend and short-term fluctuations of the data. Step S340, the processed data set is subjected to autocorrelation function calculation, as follows:

[0034] ;

[0035] in, Characterize the time interval, Characterize the hysteresis time interval The autocorrelation function value under Characterization time The data value below, Representation in time The data value below, Characterize the data mean of the processed data set, The total length of the time series in the processing data set. Specifically, the autocorrelation function is a statistic that measures the correlation between data values ​​at different time points in the time series. The lag refers to the time interval between the current time point and a certain time point in the past, and the autocorrelation function value under each lag is recorded. Step S350, determine the periodic peak lag point according to the autocorrelation function calculation result, establish a reference period with the periodic peak lag point, and establish a short time scale window with the reference period. Specifically, in the autocorrelation function calculation result, find the peaks that appear periodically. The lag points corresponding to these peaks represent the period length of the data. Select the most significant periodic peak lag point as the reference period, and establish a short time scale window based on the reference period. The short time scale window is used to capture short-term fluctuations and periodic changes in the data. This implementation method eliminates the long-term trend and seasonal components in the data through detrending and differential processing, making short-term fluctuations more significant, and identifies the periodic components in the data through autocorrelation function analysis, accurately analyzes the periodic behavior of the data, and improves the technical effect of the accuracy of the establishment of the short time scale window.

[0036] In a possible implementation, the short time scale window is established with the reference period, and step S350 further includes step S351, calculating the local coefficient of variation of the data based on the reference period. Specifically, based on the reference period, a segment of data is selected from the processing data set as the analysis object, and for each selected data segment, its standard deviation is calculated, and the standard deviation measures the degree of discreteness of the data points in the data segment. Similarly, the mean of each data segment is calculated, that is, the average value of all data points in the data segment. The local coefficient of variation is the ratio of the standard deviation to the mean, which is used to measure the relative discreteness of the data points in the data segment. Step S352, establish a dynamic short time window through the local coefficient of variation and the reference period, and use the dynamic short time window as a short time scale window. Specifically, according to actual needs and data characteristics, a threshold value of the local coefficient of variation is set. This threshold value is used to determine whether the fluctuation in the data segment is large enough to require a more refined analysis window. Compare the local coefficient of variation of each data segment with the set threshold. If the local coefficient of variation exceeds the threshold, it means that the fluctuation in the data segment is large, and a shorter dynamic short-time window needs to be established to capture these fluctuations. On the contrary, if the local coefficient of variation is small, a too short window is not needed. According to the size of the local coefficient of variation and the reference period, the length of the short-time window corresponding to each data segment is dynamically determined. The calculation formula is as follows:

[0037] W short =T base × (1-α·CV);

[0038] Among them, W short To dynamically adjust the short time window length, Tbase is the reference period, α is the adjustment coefficient, and CV is the local coefficient of variation. Based on the determined window length, a dynamic short-time window is established on the processed data set. This implementation method uses a longer window on the data segment with smaller fluctuations to reduce the amount of calculation and improve the analysis efficiency; using a shorter window on the data segment with larger fluctuations ensures the precision of the analysis. This method of dynamically adjusting the window length achieves the technical effect of maintaining the analysis accuracy while improving the overall efficiency.

[0039] In a possible implementation, after the short time scale window is established with the reference period, step S300 further includes step S360, performing frequency domain conversion on the time series data preprocessing result, and analyzing the main frequency component based on the frequency domain conversion result, as follows:

[0040] ;

[0041] in, is the frequency domain signal, is the time domain signal, is the time variable, Characterize the frequency, is an imaginary unit; step S370, taking the inverse of the main frequency component as the dominant period, and performing period expansion on the dominant period, establishing a long time scale window, and establishing a long and short time scale window according to the long time scale window and the short time scale window. Specifically, by finding the frequency point with the largest amplitude of the frequency domain signal, the main frequency component of the signal is identified, and the main frequency component reflects the main fluctuation period in the data. The inverse of the main frequency component is calculated as the dominant period, and the dominant period represents the period of the main fluctuation or trend in the data. The dominant period is periodically expanded, that is, according to the length of the dominant period, the length of the long time scale window is determined by directly multiplying it by a coefficient greater than 1. The long time scale window is used to capture long-term trends or low-frequency information in the data. Period expansion ensures that the long time scale window can cover the long-term fluctuations in the data, while avoiding information loss or redundancy caused by being too short or too long. Combining the short time scale window and the long time scale window, a combination of long and short time scale windows is formed. This combination is used to comprehensively analyze data at different time scales and capture short-term fluctuations and long-term trends. This implementation method determines the long time scale window by analyzing the frequency domain, identifying the dominant period, and expanding based on the dominant period, thereby achieving the technical effect of improving the accuracy and reliability of establishing the long time scale window.

[0042] Step S400, based on the long and short time scale windows, a backtracking instruction is generated, and the historical data of the power generation unit is backtracked and read based on the backtracking instruction, and the time series data preprocessing result is updated according to the result of the backtracking reading of the historical data, and the time scale feature extraction of the updated time series data preprocessing result is performed based on the long and short time scale windows, and the time scale feature extraction result is established. Specifically, the backtracking reading of historical data is used to obtain the detailed operation data of the power generation unit in the past period of time. These data contain the cycle information that is missing or incomplete in the current time series data preprocessing result. Through the backtracking reading, the integrity and continuity of the data on the time scale are ensured, thereby providing a more comprehensive and accurate basis for the time scale feature extraction. Specifically, according to the long and short time scale window settings, an instruction to read historical data is generated, and the historical data within the specified time period is read from the database or storage medium, and the read historical data is integrated with the current time series data to form a more complete and continuous data set. According to the settings of the long and short time scale windows, different time scale levels (such as short-term, medium-term, long-term, etc.) are defined. At each time scale level, corresponding features are extracted, including mean, standard deviation, extreme value, periodicity index, trend index, etc., to describe the statistical characteristics and dynamic changes of the data at this time scale. The extracted features are sorted and organized to form the time scale feature extraction results.

[0043] Step S500, extract spatial scale features from the time series data preprocessing results, establish spatial scale feature extraction results, send the time scale feature extraction results and spatial scale feature extraction results to the decision module, perform feature fusion, and establish feature vectors. Specifically, the time series data preprocessing results contain information related to spatial position, such as the geographical location coordinates corresponding to each data point, analyze the spatial relationship between different power generation units, extract spatial distribution, correlation and other features, for example, the average distance between each power generation unit and its neighboring units, the center position of the spatial cluster, etc., and sort and organize all calculated spatial scale features to form spatial scale feature extraction results. The time scale feature extraction results and the spatial scale feature extraction results are spliced ​​in a certain order to form a new feature vector, which contains important information about the data in both time scale and spatial scale.

[0044] Step S600, configure an adaptive dimensionality reduction strategy, perform feature dimensionality reduction based on the adaptive dimensionality reduction strategy and feature vectors, generate a collaborative dimensionality reduction strategy for shared dimensionality reduction based on the central server, optimize feature dimensionality reduction based on the collaborative dimensionality reduction strategy, establish feature dimensionality reduction results, and establish key influencing factors based on the feature dimensionality reduction results. Specifically, according to data characteristics and analysis requirements, select a dimensionality reduction method (such as PCA) and configure parameters. Utilize the computing power of the central server to achieve collaborative dimensionality reduction between multiple power generation units and improve dimensionality reduction effect and efficiency. Optimize the dimensionality reduction results through an iterative optimization algorithm to ensure that key information is not lost. Based on the feature vector after dimensionality reduction, use data mining to identify key factors that affect power generation capacity. The embodiment of the present application collects multi-dimensional data through an intelligent agent system, and uses time series data analysis technology to pre-process and periodically evaluate the data. Then, an adaptive dimensionality reduction strategy is used to reduce the data dimension and improve computing efficiency. Finally, collaborative dimensionality reduction optimization technology is used to achieve information sharing and collaborative optimization between different power generation units, thereby directly outputting key factors that affect the power generation capacity of new energy farms and other technical means, thereby achieving the technical effect of achieving efficient and accurate identification of factors affecting the power generation capacity of new energy farms.

[0045] In a possible implementation, the collaborative dimension reduction strategy based on the central server generates shared dimension reduction, performs feature dimension reduction optimization based on the collaborative dimension reduction strategy, and establishes feature dimension reduction results. Step S600 further includes step S610, after each intelligent agent performs preliminary dimension reduction, the corresponding dimension reduction result is sent to the central server. Specifically, after each intelligent agent receives the feature vector, it applies a dimension reduction algorithm (such as PCA, LDA, t-SNE, etc.) to perform preliminary dimension reduction processing on the data according to its own computing power and resources. After completing the preliminary dimension reduction, the intelligent agent sends the dimension-reduced data to the central server through the network. Step S620, configure the evaluation weight of the dimension reduction effect on the central server, and the evaluation weight includes a reconstruction error weight, an information entropy weight, and a prediction performance weight. Specifically, the central server determines multiple dimensions for evaluating the dimension reduction effect according to actual needs, wherein the reconstruction error represents the degree of difference between the dimension-reduced data and the original data when reconstructing back to the original dimension; information entropy is an indicator to measure the amount of information in the data set, and information loss should be minimized during the dimension reduction process; prediction performance reflects the performance of the dimension-reduced data in the power generation capacity prediction task. Assign corresponding weights to each evaluation dimension, and these weights reflect the importance of different dimensions in the overall evaluation. Step S630, evaluate the dimensionality reduction effect of the dimensionality reduction result based on the evaluation weight, and configure the collaborative dimensionality reduction strategy of the intelligent agent that has established shared dimensionality reduction according to the evaluation result of the dimensionality reduction effect. Specifically, the central server receives the dimensionality reduction results from each intelligent agent, and calculates the score of each dimensionality reduction result according to the preset evaluation weight and evaluation index. According to the evaluation result of the dimensionality reduction effect, the central server can adjust the dimensionality reduction algorithm parameters of each intelligent agent, select a more suitable dimensionality reduction algorithm, or adjust the data allocation and task allocation between intelligent agents to optimize the overall dimensionality reduction effect. The configured collaborative dimensionality reduction strategy is sent to each intelligent agent to guide them to perform subsequent dimensionality reduction tasks. This implementation method comprehensively evaluates the dimensionality reduction effect by configuring the weights of multiple evaluation dimensions, ensuring that the data after dimensionality reduction retains key information and is suitable for subsequent analysis and prediction tasks, and configures the collaborative dimensionality reduction strategy based on the evaluation result of the dimensionality reduction effect, thereby improving the accuracy of the generation of the collaborative dimensionality reduction strategy and thus improving the overall dimensionality reduction effect.

[0046] like Figure 2As shown, in a possible implementation, the collaborative dimensionality reduction strategy of the intelligent agent for shared dimensionality reduction is configured according to the dimensionality reduction effect evaluation result, and step S630 further includes step S631, obtaining the equipment information of the power generation unit corresponding to the intelligent agent, performing similarity evaluation of the equipment parameters according to the equipment information, and establishing a first similarity evaluation result. Specifically, the equipment information is collected from each power generation unit or the corresponding database, including but not limited to the equipment model, manufacturing year, rated power, technical specifications, etc. Statistical methods are used to compare the equipment parameters of different power generation units, and the similarity between them is calculated, such as calculating the Euclidean distance between the parameters. According to the similarity calculation result, the power generation units are grouped to form a first similarity evaluation result. Step S632, meteorological similarity evaluation is performed according to the meteorological data of the power generation unit, and a second similarity evaluation result is established. Specifically, the meteorological data of the location of the power generation unit is collected from the meteorological station, meteorological database or online service, including temperature, humidity, wind speed, wind direction, rainfall, etc. The clustering algorithm is used to evaluate the similarity of meteorological conditions at the locations of different power generation units. According to the meteorological similarity calculation result, the power generation units are grouped to form a second similarity evaluation result. Step S633, read the equipment working data of the power generation unit, perform similarity evaluation based on the equipment working data, and establish a third similarity evaluation result. Specifically, obtain real-time working data from the monitoring system of the power generation unit, including power generation, operating efficiency, fault records, etc. Use data mining methods to analyze the similarities of different power generation units in terms of working performance, operating stability, etc. According to the calculation results of the equipment working similarity, the power generation units are grouped to form a third similarity evaluation result. Step S634, complete the establishment of shared dimensionality reduction of intelligent agents based on the first similarity evaluation result, the second similarity evaluation result, and the third similarity evaluation result. Specifically, the first, second and third similarity evaluation results are integrated, and the final similarity evaluation result is obtained by weighted average. According to the comprehensive similarity evaluation results, the intelligent agents corresponding to the power generation units with higher similarity are configured as partners for shared dimensionality reduction, that is, these intelligent agents can share data, algorithms or dimensionality reduction results during the dimensionality reduction process to improve the efficiency and effect of dimensionality reduction. Establish a shared dimensionality reduction mechanism between intelligent agents to ensure that they can exchange data and work together according to the collaborative dimensionality reduction strategy. This implementation method comprehensively evaluates the similarity between power generation units by integrating the similarities of multiple dimensions such as equipment information, meteorological data and equipment working data, and achieves the technical effect of accurately establishing shared dimensionality reduction between intelligent agents.

[0047] In a possible implementation, step S600 further includes step S640, establishing a fallback node for dimensionality reduction. Specifically, in the dimensionality reduction process, several key dimensionality reduction steps are selected as fallback nodes, and the fallback nodes are key points set in the dimensionality reduction process for restoring the dimensionality reduction state when needed, which can be the key iteration step of the dimensionality reduction algorithm, a specific number of dimensionality reduction dimensions, or the state when a specific dimensionality reduction effect is achieved. At each fallback node, the state, parameters, and intermediate data of the current dimensionality reduction model are saved for subsequent fallback operations. A fallback mechanism is set to quickly restore to the state of any fallback node when needed. Step S650, when feature dimensionality reduction optimization is performed based on the collaborative dimensionality reduction strategy, if the next round of dimensionality reduction results fail to meet the expected dimensionality reduction effect, a fallback instruction is generated. Specifically, after each dimensionality reduction iteration, the dimensionality reduction effect is evaluated according to a preset evaluation standard (such as information retention rate, computational efficiency, etc.). The evaluation result is compared with the expected dimensionality reduction effect. If the evaluation result fails to meet the expected effect, a fallback instruction is generated, which contains the fallback node information to be fallen back to. Step S660, perform feature dimensionality reduction reconstruction according to the fallback instruction and the fallback node. Specifically, read the fallback instruction and determine the fallback node to which you need to fall back. According to the fallback information saved by the fallback node, restore the dimensionality reduction model to the state of the fallback node. After the state is restored, re-plan the dimensionality reduction strategy, including adjusting the dimensionality reduction algorithm parameters, changing the number of dimensionality reduction dimensions, etc. Re-execute the dimensionality reduction process according to the adjusted dimensionality reduction strategy. In the dimensionality reduction process, due to the influence of various factors such as data characteristics, algorithm selection or parameter settings, the dimensionality reduction effect may be unsatisfactory. At this time, if the dimensionality reduction process is continued directly, computing resources will be wasted and the quality of the final result will be reduced. This implementation method can flexibly respond to these challenges by establishing a fallback node for dimensionality reduction. When the dimensionality reduction result fails to meet the expected dimensionality reduction effect, it will fall back to a previous state and re-plan the dimensionality reduction strategy. It can find a dimensionality reduction method that is more suitable for the current data and improve the efficiency and effect of dimensionality reduction. In addition, the feature dimensionality reduction reconstruction also enhances the robustness and stability of the dimensionality reduction process, making it more adaptable to complex and changeable data environments.

[0048] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for identifying factors affecting the power generation capacity of new energy farms based on dimensionality reduction analysis, characterized in that: The method comprises: Establishing an intelligent agent mapped with the power generation unit, the intelligent agent includes a perception module, an analysis module, and a decision module, and the intelligent agent is configured with a central server for common communication and interaction; The sensing module is used to collect data of the power generation unit to establish a time series data set, wherein the time series data set includes meteorological data, equipment status data, and power generation output data. The time series data set is the time series data of each power generation unit in the new energy field group; The analysis module receives the time series data set collected by the perception module, performs data preprocessing, establishes the time series data preprocessing result, and performs data cycle evaluation on the time series data preprocessing result, and establishes the long and short time scale windows based on the data cycle evaluation result; Generate a backtracking instruction based on the long and short time scale windows, perform backtracking reading of historical data of the power generation unit based on the backtracking instruction, update the time series data preprocessing result according to the result of the backtracking reading of the historical data, and extract the time scale feature of the updated time series data preprocessing result based on the long and short time scale windows to establish the time scale feature extraction result; Performing spatial scale feature extraction on the time series data preprocessing result to establish a spatial scale feature extraction result, sending the time scale feature extraction result and the spatial scale feature extraction result to the decision module, performing feature fusion, and establishing a feature vector; Configure an adaptive dimensionality reduction strategy, perform feature dimensionality reduction based on the adaptive dimensionality reduction strategy and feature vectors, generate a collaborative dimensionality reduction strategy for shared dimensionality reduction based on the central server, perform feature dimensionality reduction optimization based on the collaborative dimensionality reduction strategy, establish feature dimensionality reduction results, and establish key influencing factors based on the feature dimensionality reduction results; The performing of data period evaluation on the time series data preprocessing result and establishing long and short time scale windows based on the data period evaluation result also includes: Detrending and differencing the time series data preprocessing results to generate a processed data set; The autocorrelation function of the processed data set is calculated as follows: ; in, Characterize the time interval, Characterize the hysteresis time interval The autocorrelation function value under Characterization time The data value below, Representation in time The data value below, Characterize the data mean of the processed data set, is the total length of the time series in the processed dataset; Determine a periodic peak hysteresis point according to the calculation result of the autocorrelation function, establish a reference period with the periodic peak hysteresis point, and establish a short time scale window with the reference period; The collaborative dimensionality reduction strategy for shared dimensionality reduction generated by the central server, feature dimensionality reduction optimization based on the collaborative dimensionality reduction strategy, and establishment of feature dimensionality reduction results also include: After each intelligent agent performs preliminary dimensionality reduction, the corresponding dimensionality reduction result is sent to the central server; Configuring evaluation weights of dimensionality reduction effects on a central server, wherein the evaluation weights include reconstruction error weights, information entropy weights, and prediction performance weights; The dimensionality reduction effect of the dimensionality reduction result is evaluated based on the evaluation weight, and a collaborative dimensionality reduction strategy of the intelligent agent for shared dimensionality reduction is configured according to the dimensionality reduction effect evaluation result.

2. The method for identifying factors affecting power generation capacity of a new energy farm group based on dimensionality reduction analysis according to claim 1, characterized in that: The short time scale window is established by the reference period, and further includes: Calculating a local coefficient of variation of the data based on the reference period; A dynamic short time window is established by using the local coefficient of variation and the reference period, and the dynamic short time window is used as a short time scale window.

3. The method for identifying factors affecting power generation capacity of a new energy farm group based on dimensionality reduction analysis according to claim 1, characterized in that: After establishing the short time scale window with the reference period, the method further includes: The time series data preprocessing result is converted into frequency domain, and the main frequency component is analyzed based on the frequency domain conversion result, as follows: ； in, is the frequency domain signal, is the time domain signal, is the time variable, Characterize the frequency, is an imaginary unit; The reciprocal of the main frequency component is taken as the dominant period, and the dominant period is periodically expanded to establish a long time scale window, and a long and short time scale window is established according to the long time scale window and the short time scale window.

4. The method for identifying factors affecting power generation capacity of a new energy farm group based on dimensionality reduction analysis according to claim 1, characterized in that: The collaborative dimension reduction strategy of configuring the intelligent agent for shared dimension reduction according to the dimension reduction effect evaluation result also includes: Acquire equipment information of a power generation unit corresponding to the intelligent agent, perform similarity evaluation of equipment parameters according to the equipment information, and establish a first similarity evaluation result; Performing a meteorological similarity evaluation based on the meteorological data of the power generation unit and establishing a second similarity evaluation result; reading device operating data of the power generation unit, performing similarity evaluation according to the device operating data, and establishing a third similarity evaluation result; The shared dimensionality reduction of the intelligent agent is established based on the first similarity evaluation result, the second similarity evaluation result, and the third similarity evaluation result.

5. The method for identifying factors affecting power generation capacity of a new energy farm group based on dimensionality reduction analysis according to claim 1, characterized in that: The execution data preprocessing also includes: Constructing an isolation forest model, and performing random data segmentation in a time series data set based on the isolation forest model; Outliers are identified based on the random data segmentation results, and data are removed according to the outlier scores to complete data preprocessing.

6. The method for identifying factors affecting power generation capacity of a new energy farm group based on dimensionality reduction analysis according to claim 1, characterized in that: The method further comprises: Establish a fallback node for dimensionality reduction; When feature dimensionality reduction optimization is performed based on the collaborative dimensionality reduction strategy, if the next round of dimensionality reduction results fail to meet the expected dimensionality reduction effect, a rollback instruction is generated; Feature dimension reduction and reconstruction are performed according to the rollback instruction and the rollback node.

Images