A clean energy power generation data cleaning and power prediction analysis method

Abstract

The application discloses a kind of clean energy power generation data cleaning, power prediction analysis method and system, belong to power prediction technical field, the specific steps of this prediction analysis method are as follows: (1) clean energy power generation data is collected and is stored in distributed manner;(2) the stored clean energy power generation data is extracted and is preprocessed;(3) neural network model is constructed to carry out power prediction;(4) the predicted result is visualized and is shown with exception alarm;(5) the predicted result is encrypted protection by security protection system;The application can simplify data analysis difficulty, focus on risk preference and strategy formulation, enhance power grid safety and stability, at the same time, more reasonable power supply plan can be made, the use of electric energy is controlled, energy utilization efficiency is improved, and the electricity price is scientifically predicted, to help users better manage electricity cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power generation prediction, in particular to a clean energy power generation data cleaning and power prediction analysis method. BACKGROUND

[0002] With the continuous expansion of the grid-connected installed capacity scale of clean energy power stations, the grid balance puts forward high requirements for the accuracy of clean energy power generation power prediction. The examination of clean energy power station power generation power prediction by energy supervision agencies in various regions is also becoming more and more strict, and the accuracy of power generation power prediction will directly affect the operation and profitability of the power station, and high accuracy of power prediction needs to be achieved through high-quality services. Therefore, when selecting suppliers, downstream power station customers will increasingly emphasize the accuracy of power prediction and rapid feedback during the service period. Therefore, developing wind power prediction can improve the enterprise's on-grid power, effectively reduce wind and light curtailment, improve grid dispatching efficiency, and improve grid operation safety;

[0003] Although artificial intelligence technology has made significant progress, there are still certain technical bottlenecks and limitations in the field of clean energy power generation power prediction. First, the timeliness and accuracy of meteorological data is still a challenge. Extreme weather frequently occurs, seriously affecting the efficiency and stability of clean energy power generation. In terms of accumulation of historical data and acquisition of real-time meteorological data, existing algorithms and technologies cannot fully meet the needs of clean energy power generation power prediction. Secondly, the meteorological conditions and clean energy power generation scenarios in different regions differ greatly, and algorithms need to be developed and optimized accordingly. Finally, the volatility of the energy market also poses certain challenges to clean energy power generation power prediction. In the face of fierce market competition, investors and operators pay more attention to power generation efficiency and stability.

[0004] Clean energy power generation power prediction has high requirements for data integrity and accuracy. In actual production processes, data loss and abnormal situations occur from time to time. Existing artificial intelligence technologies still have certain limitations in handling missing and abnormal data. For example, traditional data interpolation methods may cause data distortion, and abnormal value detection and processing algorithms may not completely eliminate the influence of abnormal data on prediction results. Therefore, to improve the accuracy and stability of clean energy power generation power prediction, the processing capability of missing and abnormal data needs to be strengthened. SUMMARY

[0005] The present application relates to the technical field of power generation prediction, in particular to a clean energy power generation data cleaning and power prediction analysis method.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] A clean energy power generation data cleaning and power prediction analysis method, the specific steps of which are as follows:

[0008] (1) Collect clean energy power generation data and perform distributed storage;

[0009] (2) Extract the stored clean energy power generation data and perform preprocessing;

[0010] (3) Construct a neural network model for power prediction;

[0011] (4) Visualize and alarm the prediction results;

[0012] (5) Encrypt the prediction results through a security protection system.

[0013] As a further scheme of the present application, the distributed storage of step (1) has the following specific steps:

[0014] Step one: divide each group of data according to a preset time interval to form multiple groups of data blocks, then generate an identifier for each group of data blocks through a hash algorithm, and collect node information;

[0015] Step two: select appropriate nodes to store each group of data blocks according to data block division rules and node load conditions, and through a load balancing algorithm; after the data block storage is completed, configure and copy a specified number of data blocks to multiple groups of nodes according to system requirements and available resources;

[0016] Step three: when the data stored by the node changes, update the data from one node to other nodes through a data synchronization algorithm, then automatically detect the node running condition, and perform data migration or repair on the faulty node.

[0017] As a further scheme of the present application, the clean energy power generation data preprocessing of step (2) has the following specific steps:

[0018] Step 1: smooth each group of data through a moving average or exponential smoothing method to reduce random fluctuations, then convert the data to the frequency domain through Fourier transform, filter out high-frequency noise, and then detect whether there are duplicate data records; if there are duplicate data, delete them;

[0019] Step 2: standardize each clean energy power generation data to unify the format, and integrate and summarize them into a sample set, then calculate the standard deviation of the sample set to screen out abnormal data in the sample set;

[0020] Step 3: Calculate the distance between the remaining data points and the known data points by KNN algorithm, find the K groups of known data points closest to the data points to be processed, and classify or regression predict the remaining data points according to the labels or values of the found K groups of known data points;

[0021] Step 4: Detect the missing values existing in each group of data, mark the location of each missing value in the corresponding data, statistically analyze and visualize the missing values existing in each group of data to obtain the distribution and influence range of the missing values, and calculate the average or median of the corresponding K groups of data points found by KNN algorithm to replace the abnormal values or missing values.

[0022] As a further scheme of the present application, the specific calculation formula of the Fourier transform in step 1 is as follows:

[0023] (1)

[0024] In the formula, represents the complex representation of the frequency domain; represents the original signal; represents the frequency;

[0025] The specific calculation formula of the standardization processing in step 2 is as follows:

[0026] (2)

[0027] (3)

[0028] (4)

[0029] In the formula, represents the sample mean of the first group of samples; represents the range of the first group of samples; represents the standard deviation of the first group of samples;

[0030] The specific calculation formula of the KNN algorithm in step 3 is as follows:

[0031] (5)

[0032] In the formula, and are the feature vectors of two groups of samples, is the number of features.

[0033] As a further scheme of the present application, the specific steps of the power prediction in step (3) are as follows:

[0034] Step I: Mine the processed clean energy power generation data timestamp features, and perform difference sequence processing to obtain the feature mean and standard deviation of the power plant, and then perform sequence length standardization processing on the obtained feature data;

[0035] Step II: Divide the collected data into training set, validation set and test set according to 14:3:3, set the number of iterations, batch processing parameters and node dropout rate, then train the neural network model through the training set, input the power time series into the model, and update the parameters of the model through back propagation, and monitor the performance of the model using the validation set;

[0036] Step III: Use the test set to evaluate the performance of the trained neural network model, calculate the loss value of the model through the root mean square error, and minimize the loss value through the Adam optimizer. If the loss value does not meet the preset threshold, retrain the neural network model, otherwise, deploy the model to the cloud platform;

[0037] Step IV: First, use one-dimensional convolutional neural network CNN to extract the spatial features of photovoltaic power data, then use multi-layer BiLSTM shared network to extract the features of time series data, and pass the extracted time correlation to the subsequent layer, then the subsequent fully connected layer assigns different weights to the BiLSTM hidden layer through the Attention layer and outputs the processing result, and decodes the output processing result through the LSTM decoder to obtain the prediction result.

[0038] A clean energy power generation data cleaning and power prediction analysis system, comprising a data acquisition module, a cleaning analysis module, a power prediction module, an operation management module, a data display module, an abnormal warning module, a system management module and a user terminal;

[0039] The data acquisition module is used for real-time acquisition and manual upload of power plant data, and stores the data in the database;

[0040] The power prediction module is used for power prediction according to real-time environmental factors using the trained prediction model;

[0041] The operation management module is used for managing clean energy power generation power information;

[0042] The data display module is used to provide a user interface to display real-time data, historical data and prediction results of the power plant, helping users understand the operation and trend of the clean energy power plant;

[0043] The abnormal warning module is used to monitor the operation state of the power plant, detect abnormal conditions and send alarm notifications to relevant personnel;

[0044] The system management module is used for providing user management, permission management, data backup and recovery functions of the system.

[0045] The user terminal is used for receiving data sent by the data display module and feeding back to the user for viewing.

[0046] Compared with the prior art, the present application has the following advantages:

[0047] The clean energy power generation data cleaning and power prediction analysis method pre-processes the clean energy power generation data through the KNN algorithm, mines the time stamp features of the processed data, and obtains the feature mean and standard deviation of the power plant through difference sequence processing. Then, the sequence length of each group of feature data is standardized, each group of collected data is randomly divided into a training set, a validation set and a test set, and each parameter is set. Then, the neural network model is trained through the training set, and the power time sequence is input into the model. The parameters of the model are updated through back propagation, and the performance of the model is monitored using the validation set. The performance of the trained neural network model is evaluated using the test set. The loss value of the model is calculated through the root mean square error, and the loss value is minimized through the Adam optimizer. The trained model is deployed to the cloud platform, and the latest collected data is input into the model. The spatial features of the photovoltaic power data are extracted using one-dimensional convolutional neural network CNN, and the time sequence data features are extracted using multi-layer BiLSTM shared network. The extracted time correlation is transmitted to the subsequent layer, and the subsequent fully connected layer allocates different weights to the BiLSTM hidden layer through the Attention layer and outputs the processing result. The output processing result is decoded by the LSTM decoder to obtain the prediction result. The method can simplify data analysis difficulty, focus on risk preference and strategy formulation, enhance power grid safety and stability, and make more reasonable power supply plans to control the use of electric energy and improve energy utilization efficiency. The method can also scientifically predict electricity prices to help users better manage electricity costs. BRIEF DESCRIPTION OF DRAWINGS

[0048] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation of the present application.

[0049] Figure 1 A flowchart of a clean energy power generation data cleaning and power prediction analysis method according to the present application is shown in the figure.

[0050] Figure 2 A system block diagram of a clean energy power generation data cleaning and power prediction analysis system according to the present application is shown in the figure. DETAILED DESCRIPTION

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0052] Reference Figure 1 This embodiment discloses a method for cleaning clean energy power generation data and predicting power output. The specific steps of this prediction and analysis method are as follows:

[0053] Collect clean energy power generation data and store it in a distributed manner.

[0054] Specifically, the data is divided into multiple data blocks according to a preset time interval. Then, an identifier for each data block is generated using a hash algorithm. Information about each node is collected. Based on the data block partitioning rules and node load, a load balancing algorithm is used to select suitable nodes to store each data block. After the data blocks are stored, a specified number of data blocks are copied to multiple nodes according to system requirements and available resources. When the data stored on a node changes, the data update is propagated from one node to other nodes using a data synchronization algorithm. Then, the node operation status is automatically detected, and data migration or repair is performed on faulty nodes.

[0055] Extract and preprocess the stored clean energy power generation data.

[0056] Specifically, the data is smoothed using moving averages or exponential smoothing to reduce random fluctuations. Then, Fourier transform is used to convert the data to the frequency domain to filter out high-frequency noise. Duplicate data records are checked and deleted if found. The clean energy power generation data are standardized to a unified format and integrated into a sample set. The standard deviation of the sample set is calculated to filter out outliers. The KNN algorithm is used to calculate the distance between the remaining data points and known data points, finding the K nearest known data points to the data points to be processed. Based on the labels or values ​​of the K known data points, the remaining data points are classified or regressed. Missing values ​​are detected in each data set, and their locations are marked. Statistical and visual analysis is performed on the missing values ​​in each data set to obtain their distribution and impact. The mean or median of the corresponding K data points found by the KNN algorithm is calculated to replace outliers or missing values.

[0057] It should be further explained that the specific calculation formula for the Fourier transform is as follows:

[0058] (1)

[0059] In the formula, Complex number representation of the frequency domain; representing the original signal; representing the frequency;

[0060] The standardization process described in step 2 has the following specific calculation formula:

[0061] (2)

[0062] (3)

[0063] (4)

[0064] In the formula, representing the sample mean of the first group of samples; representing the range of the first group of samples; representing the standard deviation of the first group of samples;

[0065] The KNN algorithm described in step 3 has the following specific calculation formula:

[0066] (5)

[0067] In the formula, and are the feature vectors of the two groups of samples, is the number of features.

[0068] A neural network model is constructed for power prediction.

[0069] Specifically, the processed clean energy power generation data timestamp features are mined, and a difference sequence processing is performed to obtain the feature mean and standard deviation of the power plant. Then, the obtained feature data is subjected to sequence length standardization processing. The collected data is randomly divided into a training set, a validation set, and a test set according to a ratio of 14:3:3. The number of iterations, batch processing parameters, and node dropout rate are set. Then, the neural network model is trained by the training set, and the power time series is input into the model. The parameters of the model are updated through back propagation. The performance of the model is monitored by using the validation set. The performance of the trained neural network model is evaluated by using the test set. The loss value of the model is calculated by using the root mean square error. The loss value is minimized by using the Adam optimizer. If the loss value does not meet the preset threshold, the neural network model is retrained. Otherwise, the model is deployed to the cloud platform. The spatial features of the photovoltaic power data are extracted by using a one-dimensional convolutional neural network (CNN). The features of the time series data are extracted by using a multi-layer BiLSTM shared network. The extracted time correlation is transmitted to the subsequent layer. The subsequent fully connected layer allocates different weights to the BiLSTM hidden layer by using an Attention layer and outputs the processing result. The processing result is decoded by using an LSTM decoder to obtain the prediction result.

[0070] The prediction result is visualized and displayed and an abnormality alarm is given.

[0071] The prediction result is encrypted and protected by a security protection system.

[0072] With reference to Figure 2 The embodiment discloses a clean energy power generation data cleaning and power prediction analysis system, which comprises a data acquisition module, a cleaning analysis module, a power prediction module, an operation management module, a data display module, an abnormality warning module, a system management module, and a user terminal.

[0073] The data acquisition module is used for real-time acquisition and manual upload of power plant data, and stores the data in a database. The power prediction module is used for power prediction by using a trained prediction model according to real-time environmental factors. The operation management module is used for management of clean energy power generation power information. The data display module is used for providing a user interface to display real-time data, historical data, and prediction results of a power plant, helping users to understand the operation and trend of a clean energy power plant. The abnormality warning module is used for monitoring the operation state of a power plant, detecting abnormal conditions, and sending an alarm notification to relevant personnel. The system management module is used for providing user management, permission management, data backup, and recovery functions of the system. The user terminal is used for receiving data sent by the data display module and feeding back to users for viewing.

Claims

1. A clean energy power generation data cleaning, power prediction analysis method, characterized in that, The specific steps of the prediction analysis method are as follows: (1) Collect clean energy power generation data and store it in a distributed manner; (2) Extract and preprocess the stored clean energy power generation data; (3) Build a neural network model for power prediction; (4) Visualize the prediction results and issue abnormal alarms; (5) Encrypt the prediction results through a security protection system; The specific steps of the clean energy power generation data preprocessing in step (2) are as follows: Step 1: Smooth each group of data using moving average or exponential smoothing method to reduce random fluctuations, then convert the data to frequency domain through Fourier transform, filter out high-frequency noise, and detect whether there are duplicate data records. If there are duplicate data, delete them; Step 2: Standardize each clean energy power generation data to unify the format, and integrate it into a sample set. Then calculate the standard deviation of the sample set to filter out abnormal data in the sample set; Step 3: Calculate the distance between the remaining data points and the known data points using KNN algorithm, find the K groups of known data points closest to the data points to be processed, and classify or regress the remaining data points according to the labels or values of the found K groups of known data points; Step 4: Detect missing values in each group of data and mark their positions in the corresponding data. Statistically and visually analyze the missing values to obtain their distribution and impact range, and calculate the average or median of the corresponding K groups of data points found by KNN algorithm to replace the abnormal values or missing values; The specific steps of the power prediction in step (3) are as follows: Step I: Mine the timestamp features of the processed clean energy power generation data, and perform difference sequence processing to obtain the feature mean and standard deviation of the power plant. Then perform sequence length normalization processing on each group of feature data obtained; Step II: Divide each group of collected data into training set, validation set and test set according to 14:3:3, and set the number of iterations, batch processing parameters and node dropout rate. Then train the neural network model through the training set, input the power time series into the model, and update the model parameters through back propagation. At the same time, use the validation set to monitor the performance of the model; Step III: Use the test set to evaluate the performance of the trained neural network model, calculate the loss value of the model through root mean square error, and minimize the loss value through Adam optimizer. If the loss value does not meet the preset threshold, retrain the neural network model. Otherwise, deploy the model to the cloud platform; Step IV: First, use one-dimensional convolutional neural network CNN to extract the spatial features of photovoltaic power data, then use multi-layer BiLSTM shared network to extract the features of time series data, and pass the extracted time correlation to the subsequent layer. Then the subsequent fully connected layer allocates different weights to the BiLSTM hidden layer through the Attention layer and outputs the processing result. Decode the output processing result through the LSTM decoder to obtain the prediction result.

2. The method of claim 1, wherein, The specific steps of the distributed storage in step (1) are as follows: Step one: according to the preset time interval to each group of data segmentation, to form a plurality of groups of data block, and then through the hash algorithm to generate each group of data block identification, collection each group of node information; Step two: according to the data block division rule and node load condition, and through the load balancing algorithm to select the appropriate node to store each group of data block, data block storage is completed, according to the system requirements and available resources to configure copy a specified number of data block to a plurality of groups of nodes; Step three: when the node stored data changes, through the data synchronization algorithm to update the data from a node to other nodes, and then automatically detect node running condition, and to the fault node data migration or repair.

3. The method of claim 1, wherein, The Fourier transform formula of step 1 is as follows: （1） wherein represents a complex representation of the frequency domain; represents the original signal; represents the frequency; The standardization processing formula of step 2 is as follows: （2） （3） （4） wherein representing the first sample mean of the group of samples; representing the first range of the group of samples; representing the first standard deviation of the group of samples; The KNN algorithm formula of step 3 is as follows: （5） wherein and are eigenvectors of the two sets of samples, is the number of features.

Images