New energy-oriented multivariate load risk perception method and system
Through the multi-load risk perception method for new energy, machine learning and deep learning technology are used to predict new energy power generation and electricity consumption patterns, and the risk index is calculated, which solves the power system challenges brought about by the uncertainty of new energy power generation and multi-load, and improves the power system. The economy and reliability of the system.
Patent Information
- Application Number
- CN202411015322.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-19
AI Technical Summary
The uncertainty of new energy power generation and multi-loads leads to challenges in economic scheduling and safe and stable operation in the power system, and traditional power system scheduling methods are difficult to meet the requirements.
Using a multi-load risk perception method for new energy, by obtaining real-time operating parameters of new energy power generation equipment and multi-load, using machine learning and deep learning technology to establish a prediction model, predict the power generation power and power consumption mode in the future period, and calculate the The risk index of multi-load provides a deep learning-based risk assessment model to comprehensively consider the intermittent and volatility of new energy power generation and the uncertainty of multi-load.
It improves the economy and reliability of new energy power generation systems, provides quantitative decision-making basis, enhances the accuracy of power scheduling and risk management, and reduces the risks of power balance and supply and demand matching.
Smart Images

Figure CN118982238A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power grids, and in particular relates to a multi-load risk perception method and system for renewable energy. Background Art
[0002] As global energy shortages and environmental pollution become increasingly prominent, new energy power generation technologies represented by wind and solar energy have been rapidly developed and widely used. However, new energy power generation is characterized by intermittency, volatility and uncertainty. Its output is affected by natural conditions such as wind speed and sunshine, making it difficult to accurately predict and control. At the same time, the power load side also presents diversified, personalized and dynamic characteristics, and different types of users have different requirements for power quality and power supply reliability. In a power system with high penetration of new energy, the uncertainty of new energy power generation and the volatility of multiple loads are coupled with each other, bringing severe challenges to the economic dispatch and safe and stable operation of the power system.
[0003] In order to cope with the uncertainty risks of renewable energy generation and diversified loads, traditional power system dispatching and operation methods have been unable to meet the requirements. Summary of the invention
[0004] The embodiments of the present invention provide a multi-load risk perception method and system for new energy, which can solve the uncertainty risk problems of new energy power generation and multi-load in the prior art.
[0005] A first aspect of an embodiment of the present invention provides a multi-layer load risk perception method for new energy, including:
[0006] Acquire real-time operating parameters of new energy power generation equipment in the power system, the real-time operating parameters including power generation power, grid frequency, voltage amplitude and phase angle; based on the real-time operating parameters, use a new energy power generation prediction model pre-established using a machine learning algorithm to obtain predicted power generation power of the new energy power generation equipment in the future;
[0007] Acquire real-time operation data of multiple loads connected to the new energy power generation equipment, wherein the real-time operation data includes the type, power, voltage, current and power factor of the loads; perform cluster analysis on the multiple loads using data mining technology based on the real-time operation data to obtain power consumption patterns and characteristics of different types of loads;
[0008] The predicted power generation and the power consumption pattern are input into a risk assessment model based on deep learning, and the intermittency and volatility of renewable energy power generation and the uncertainty of multiple loads are comprehensively considered to calculate the risk index of multiple loads in the future.
[0009] In an optional implementation, based on the real-time operating parameters, using a new energy power generation prediction model pre-established using a machine learning algorithm, the predicted power generation of the new energy power generation equipment in the future period is obtained, including:
[0010] By installing wind speed and direction sensors, rotation speed sensors, torque sensors and pitch angle sensors on the wind turbines, and installing radiation sensors, temperature sensors and current and voltage sensors on the photovoltaic arrays, the real-time operating parameters of the wind turbines and photovoltaic arrays can be collected;
[0011] Use industrial-grade wireless communication protocols to build a wireless sensor network, transmit the collected real-time operating parameters to the edge server for data analysis, filtering, compression and encryption, and upload to the cloud server;
[0012] In the cloud server, the real-time operating parameters uploaded by the edge server are obtained through the data access service and stored in the time series database; the stored real-time operating parameters are cleaned and features are constructed to obtain time series features, meteorological features and historical power features, and the feature subset with the highest correlation with the power generation and the lowest redundancy is screened out through the maximum correlation minimum redundancy algorithm;
[0013] Using the selected feature subsets as input, a machine learning algorithm is used to establish multiple power generation prediction models, and the performance of the prediction models is improved through cross-validation and hyperparameter optimization;
[0014] Multiple prediction models are combined to form a Stacking ensemble learning model, in which the first layer consists of a long short-term memory neural network, a gradient boosting decision tree, and a random forest. The second layer uses a linear regression algorithm to integrate the prediction results of each model in the first layer by weighted averaging to obtain the final power generation prediction value.
[0015] In an optional implementation, real-time operation data of multiple loads connected to the new energy power generation equipment is obtained, and the real-time operation data includes the type, power, voltage, current and power factor of the loads; based on the real-time operation data, a data mining technology is used to perform cluster analysis on the multiple loads, and the power consumption patterns and characteristics of different types of loads are obtained, including:
[0016] Smart meters are installed at various load locations to measure at least one real-time operating parameter of the load, including voltage, current, active power, reactive power, and power factor, and the measured data is uploaded to a data concentrator through a communication module; the data concentrator periodically reads the measured data of the smart meters through wired or wireless communication, and performs analysis, verification, storage, and preprocessing to generate load operation data;
[0017] The data concentrator uploads the load operation data to the cloud server, which decrypts, decompresses and converts the received data into a format before storing it in a distributed time series database. The stored load operation data is normalized to its maximum and minimum values to eliminate the dimensional differences of different load data, and the sliding window averaging method is used to smooth and filter the data to remove noise and random fluctuations.
[0018] Extract the intraday curve characteristics and cycle characteristics of the load from the time dimension, extract the correlation characteristics and similarity characteristics of the load from the space dimension, and construct the load feature vector;
[0019] The improved DBSCAN clustering algorithm based on adaptive parameter selection is used to cluster the load feature vectors. By dynamically adjusting the density threshold and neighborhood radius, high-density load clusters are found and low-density outliers are distinguished.
[0020] The clustering results are analyzed and interpreted, the central load curve and characteristic statistics of each load cluster are calculated, and the typical power consumption patterns of different types of loads are characterized.
[0021] In an optional implementation, an improved DBSCAN clustering algorithm based on adaptive parameter selection is used to perform cluster analysis on load feature vectors, and by dynamically adjusting the density threshold and neighborhood radius, high-density load clusters are found and low-density outliers are distinguished, including:
[0022] The initial density threshold and neighborhood radius are set according to prior knowledge, and the number of samples in the neighborhood of each load sample point is calculated as the local density;
[0023] According to the mean and standard deviation of the local density of all sample points, the density threshold of each sample point is estimated, and the clustering granularity is adjusted by adjusting the control parameters;
[0024] According to the density threshold of the sample points, find the nearest neighbor sample points in the neighborhood of each sample point, and calculate the average distance as the new neighborhood radius;
[0025] The median of the density thresholds of all sample points is taken as the new global density threshold, and the median of the neighborhood radius of all sample points is taken as the new global neighborhood radius, and the DBSCAN algorithm parameters are adaptively adjusted;
[0026] Use the adaptively adjusted density threshold and neighborhood radius to perform DBSCAN clustering and obtain the clustering results of the load sample points;
[0027] The silhouette coefficient is used to evaluate the clustering quality. If the clustering quality does not meet expectations, the control parameters are iteratively adjusted and the density threshold and neighborhood radius of the sample points are re-estimated until a satisfactory clustering result is obtained.
[0028] In an optional implementation, the predicted power generation and the power consumption pattern are input into a risk assessment model based on deep learning, and the intermittency and volatility of renewable energy power generation and the uncertainty of multiple loads are comprehensively considered to calculate the risk index of multiple loads in the future period of time, including:
[0029] The predicted power of renewable energy generation and the electricity consumption patterns of multiple loads are used as input features of the risk assessment model to construct the input data of the model;
[0030] The long short-term memory neural network is used as the basic architecture of the risk assessment model. The attention mechanism is introduced into the network to adaptively allocate the weights of features at different times. A multi-head self-attention layer is designed to model the interactive relationship between renewable energy generation and load power consumption.
[0031] The mean square error is used as the loss function of the model to measure the difference between the predicted risk index and the actual risk index. By minimizing the loss function, the prediction result of the model is made as close as possible to the actual risk level.
[0032] The model is trained using historical data including actual power generation of renewable energy, actual power consumption curves of multiple loads, and actual risk indexes of corresponding time periods, and the Adam optimization algorithm is used to adaptively adjust the model parameters.
[0033] Use the test set data to evaluate the performance of the trained model, and use indicators such as mean absolute error and root mean square error to measure the prediction accuracy of the model;
[0034] The trained risk assessment model is deployed to the power dispatching system to obtain the predicted power of renewable energy generation and the power consumption pattern of multiple loads in real time, input them into the model for risk prediction, and output the risk index for a period of time in the future.
[0035] A second aspect of an embodiment of the present invention provides a multi-load risk perception system for new energy sources, including:
[0036] The first unit is used to obtain real-time operating parameters of new energy power generation equipment in the power system, wherein the real-time operating parameters include power generation power, grid frequency, voltage amplitude and phase angle; based on the real-time operating parameters, a new energy power generation prediction model pre-established by using a machine learning algorithm is used to obtain the predicted power generation power of the new energy power generation equipment in a future period of time;
[0037] The second unit is used to obtain real-time operation data of multiple loads connected to the new energy power generation equipment, wherein the real-time operation data includes the type, power, voltage, current and power factor of the loads; based on the real-time operation data, a data mining technology is used to perform cluster analysis on the multiple loads to obtain power consumption patterns and characteristics of different types of loads;
[0038] The third unit is used to input the predicted power generation and the power consumption pattern into a risk assessment model based on deep learning, comprehensively consider the intermittency and volatility of new energy power generation and the uncertainty of multiple loads, and calculate the risk index of multiple loads in the future.
[0039] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, including:
[0040] processor;
[0041] a memory for storing processor-executable instructions;
[0042] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0043] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.
[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0045] This scheme constructs a multi-load risk assessment model based on deep learning, which can comprehensively consider the intermittency and volatility of renewable energy generation and the uncertainty of multi-loads, and predict the load risk level in the future. The model uses LSTM neural network and attention mechanism, which can effectively capture the long-term dependencies and key influencing factors in time series data, and improve the accuracy of risk prediction. At the same time, through the online application of the model, it provides a quantitative decision-making basis for power dispatching and risk management, which helps to improve the economy and reliability of renewable energy power generation systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0047] Figure 1 It is a flow chart of a multi-load risk perception method for new energy sources according to an embodiment of the present invention;
[0048] Figure 2 It is a structural schematic diagram of a multi-load risk perception system for new energy according to an embodiment of the present invention. DETAILED DESCRIPTION
[0049] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.
[0050] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention.
[0051] Figure 1 FIG. 1 is a flow chart of a multi-load risk perception method for renewable energy according to an embodiment of the present invention. Figure 1 As shown, the method includes:
[0052] S101. Acquire real-time operating parameters of new energy power generation equipment in the power system, the real-time operating parameters including power generation power, grid frequency, voltage amplitude and phase angle; based on the real-time operating parameters, use a new energy power generation prediction model pre-established using a machine learning algorithm to obtain a predicted power generation power of the new energy power generation equipment in the future;
[0053] S102. Acquire real-time operation data of multiple loads connected to the new energy power generation equipment, the real-time operation data including the type, power, voltage, current and power factor of the load; perform cluster analysis on the multiple loads using data mining technology based on the real-time operation data to obtain power consumption patterns and characteristics of different types of loads;
[0054] S103. Input the predicted power generation and the power consumption pattern into a risk assessment model based on deep learning, comprehensively consider the intermittency and volatility of renewable energy power generation and the uncertainty of multiple loads, and calculate the risk index of multiple loads in the future.
[0055] In an optional embodiment,
[0056] Based on the real-time operating parameters, the predicted power generation of the new energy power generation equipment in the future is obtained by using the new energy power generation power prediction model pre-established by the machine learning algorithm, including:
[0057] By installing wind speed and direction sensors, rotation speed sensors, torque sensors and pitch angle sensors on the wind turbines, and installing radiation sensors, temperature sensors and current and voltage sensors on the photovoltaic arrays, the real-time operating parameters of the wind turbines and photovoltaic arrays can be collected;
[0058] Use industrial-grade wireless communication protocols to build a wireless sensor network, transmit the collected real-time operating parameters to the edge server for data analysis, filtering, compression and encryption, and upload to the cloud server;
[0059] In the cloud server, the real-time operating parameters uploaded by the edge server are obtained through the data access service and stored in the time series database; the stored real-time operating parameters are cleaned and features are constructed to obtain time series features, meteorological features and historical power features, and the feature subset with the highest correlation with the power generation and the lowest redundancy is screened out through the maximum correlation minimum redundancy algorithm;
[0060] Using the selected feature subsets as input, a machine learning algorithm is used to establish multiple power generation prediction models, and the performance of the prediction models is improved through cross-validation and hyperparameter optimization;
[0061] Multiple prediction models are combined to form a Stacking ensemble learning model, in which the first layer consists of a long short-term memory neural network, a gradient boosting decision tree, and a random forest. The second layer uses a linear regression algorithm to integrate the prediction results of each model in the first layer by weighted averaging to obtain the final power generation prediction value.
[0062] For example, in order to obtain real-time operating parameters of new energy power generation equipment, this solution designs an intelligent sensor network architecture based on the Internet of Things. The architecture consists of a field layer, a network layer, and an application layer.
[0063] At the field level, different types of sensors are deployed according to the characteristics of wind turbines and photovoltaic arrays. For wind turbines, high-precision wind speed and direction sensors, speed sensors, torque sensors, and pitch angle sensors suitable for harsh environments are selected to measure the key mechanical parameters of wind turbines. For photovoltaic arrays, high-sensitivity radiation sensors, waterproof temperature sensors, and high-precision current and voltage sensors are selected to measure the key electrical parameters and environmental parameters of photovoltaic arrays. In order to ensure the real-time and reliability of data transmission, industrial-grade wireless communication protocols such as Zigbee and LoRa are used at the field level to build a low-power, high-coverage wireless sensor network.
[0064] At the network layer, a data processing platform based on edge computing was built. The platform consists of edge servers deployed in wind farms and photovoltaic power stations, which are connected to the wireless sensor gateway at the field layer through high-speed industrial Ethernet, realizing on-site data processing and storage. The edge server uses a high-performance industrial computer, equipped with a real-time operating system and data processing software, to parse, filter, compress and encrypt sensor data, reducing the amount of data uploaded to the cloud and the communication pressure. At the same time, the edge server also has certain computing and storage capabilities, and can execute some key control strategies and fault diagnosis algorithms locally, improving the response speed and autonomy of the system.
[0065] At the application layer, a cloud computing-based data management and analysis system was developed. The system adopts a microservice architecture, which consists of multiple service modules such as data access, data storage, data processing, and data visualization, and realizes communication and collaboration between services through RESTful API. The data access service is responsible for obtaining real-time data from the edge server and decrypting and decompressing the data; the data storage service uses the time series database InfluxDB, which is optimized for the characteristics of time series data and provides efficient data writing and query performance; the data processing service is based on the Apache Spark big data processing framework, which realizes data cleaning, conversion, aggregation and mining; the data visualization service uses the Grafana platform, which provides real-time monitoring, historical playback, statistical analysis and other visualization functions, so that users can intuitively understand the operating status of new energy power generation equipment.
[0066] Through the IoT intelligent sensor network architecture, this solution realizes the automated collection and centralized management of real-time operating parameters of new energy power generation equipment, providing high-quality data support for subsequent power generation prediction and risk perception.
[0067] After obtaining the real-time operating parameters of renewable energy power generation equipment, this solution further studies how to use these parameters to establish an accurate power generation prediction model. Since renewable energy power generation is greatly affected by weather conditions and has obvious nonlinear and non-stationary characteristics, traditional physical models and statistical models are difficult to achieve ideal prediction results. Therefore, this solution explores the application of machine learning technology in renewable energy power generation prediction and proposes a prediction method based on Stacking ensemble learning.
[0068] First, this scheme preprocesses the real-time operating data of wind turbines and photovoltaic arrays. Through outlier detection and data cleaning, erroneous data caused by equipment failure or communication interruption are eliminated. Then, various types of features are constructed, including time series features (such as lag terms, moving averages), meteorological features (such as wind speed, radiation intensity), historical power features (such as power at the same time of the previous day), etc., forming a high-dimensional feature space. Then, the maximum relevance minimum redundancy (mRMR) algorithm is used to screen and sort the features, and several features with the highest correlation with power generation and the lowest redundancy are selected as the input of the model.
[0069] After the feature engineering was completed, this solution compared and tested a variety of machine learning algorithms, including support vector machine (SVM), random forest (RF), gradient boosted decision tree (GBDT) and long short-term memory neural network (LSTM). Among them, SVM uses radial basis kernel function, and optimizes the penalty factor C and kernel function parameter g through grid search; RF and GBDT use a combination of grid search and random search to optimize hyperparameters such as the number of trees, maximum depth, and minimum number of leaf node samples; LSTM uses a three-layer network structure, and alleviates the overfitting problem through methods such as Dropout and Early Stopping.
[0070] In order to further improve the prediction accuracy, this scheme proposes a prediction model based on Stacking ensemble learning. The model is divided into two layers. The first layer consists of multiple base learners, including LSTM, GBDT and RF, which learn the characteristics of data from different angles such as time series, tree structure and decision rules; the second layer uses a linear regression algorithm, takes the prediction results of each base learner in the first layer as a new feature, and trains the final ensemble model. Through weighted averaging, the ensemble model can combine the advantages of each base learner, reduce the bias and variance of a single model, and improve the robustness and generalization of the prediction.
[0071] During the model training and testing process, this solution uses methods such as cross-validation and leave-one-out to optimize the model's hyperparameters, and uses multiple evaluation indicators such as mean absolute error (MAE), root mean square error (RMSE) and determination coefficient (R2) to comprehensively evaluate the model's prediction performance. After multiple rounds of iterations and tuning, a Stacking ensemble learning model with superior performance was finally obtained. The MAE, RMSE and R2 indicators on the test set reached 2.57%, 3.94% and 0.95 respectively, which is significantly better than a single model and traditional methods.
[0072] The proposed new energy power prediction model based on machine learning can fully explore the hidden patterns and rules contained in real-time operating parameters, overcome the uncertainty and volatility of new energy power generation, and achieve high-precision power prediction within the next 1 to 6 hours. This model can provide a reliable decision-making basis for power grid dispatching.
[0073] In an optional embodiment,
[0074] Acquire real-time operation data of multiple loads connected to the new energy power generation equipment, the real-time operation data including the type, power, voltage, current and power factor of the loads; perform cluster analysis on the multiple loads using data mining technology based on the real-time operation data, and obtain the power consumption patterns and characteristics of different types of loads, including:
[0075] Smart meters are installed at various load locations to measure at least one real-time operating parameter of the load, including voltage, current, active power, reactive power, and power factor, and the measured data is uploaded to a data concentrator through a communication module; the data concentrator periodically reads the measured data of the smart meters through wired or wireless communication, and performs analysis, verification, storage, and preprocessing to generate load operation data;
[0076] The data concentrator uploads the load operation data to the cloud server, which decrypts, decompresses and converts the received data into a format before storing it in a distributed time series database. The stored load operation data is normalized to its maximum and minimum values to eliminate the dimensional differences of different load data, and the sliding window averaging method is used to smooth and filter the data to remove noise and random fluctuations.
[0077] Extract the intraday curve characteristics and cycle characteristics of the load from the time dimension, extract the correlation characteristics and similarity characteristics of the load from the space dimension, and construct the load feature vector;
[0078] The improved DBSCAN clustering algorithm based on adaptive parameter selection is used to cluster the load feature vectors. By dynamically adjusting the density threshold and neighborhood radius, high-density load clusters are found and low-density outliers are distinguished.
[0079] The clustering results are analyzed and interpreted, the central load curve and characteristic statistics of each load cluster are calculated, and the typical power consumption patterns of different types of loads are characterized.
[0080] In an optional embodiment,
[0081] The predicted power generation and the power consumption pattern are input into the risk assessment model based on deep learning, and the intermittent and volatile nature of renewable energy power generation and the uncertainty of multiple loads are comprehensively considered. The risk index of multiple loads in the future is calculated to include:
[0082] Smart meters are installed at various load locations to measure at least one real-time operating parameter of the load, including voltage, current, active power, reactive power, and power factor, and the measured data is uploaded to a data concentrator through a communication module; the data concentrator periodically reads the measured data of the smart meters through wired or wireless communication, and performs analysis, verification, storage, and preprocessing to generate load operation data;
[0083] The data concentrator uploads the load operation data to the cloud server, which decrypts, decompresses and converts the received data into a format before storing it in a distributed time series database. The stored load operation data is normalized to its maximum and minimum values to eliminate the dimensional differences of different load data, and the sliding window averaging method is used to smooth and filter the data to remove noise and random fluctuations.
[0084] Extract the intraday curve characteristics and cycle characteristics of the load from the time dimension, extract the correlation characteristics and similarity characteristics of the load from the space dimension, and construct the load feature vector;
[0085] The improved DBSCAN clustering algorithm based on adaptive parameter selection is used to cluster the load feature vectors. By dynamically adjusting the density threshold and neighborhood radius, high-density load clusters are found and low-density outliers are distinguished. Parallel computing and indexing mechanisms are introduced to improve clustering efficiency.
[0086] The clustering results are analyzed and interpreted, the central load curve and characteristic statistics of each load cluster are calculated, and the typical power consumption patterns of different types of loads are characterized.
[0087] For example, in a new energy power generation system, the loads connected to the power generation equipment are diversified, including industrial loads, commercial loads, residential loads and other types. In order to deeply analyze the power consumption behavior of multiple loads, it is necessary to obtain the real-time operation data of the loads. This solution designs a load data acquisition system based on smart meters and data concentrators, which realizes the automatic collection and remote transmission of load operation data.
[0088] First, smart meters are installed at various loads to measure the key electrical parameters of the loads. Smart meters use high-precision voltage and current sensors and energy metering chips, which can measure the voltage, current, active power, reactive power, power factor and other parameters of the load in real time, and upload the data to the data concentrator through the built-in communication module. According to the capacity and importance of the load, smart meters of different specifications and functions are selected, such as industrial high-voltage meters, commercial multi-function meters and residential simple meters.
[0089] Secondly, a data concentrator is installed in the substation or distribution room to aggregate and process the data uploaded by the smart meter. The data concentrator uses an industrial-grade embedded system equipped with a high-performance CPU, large-capacity memory and multiple communication interfaces, such as RS485, Ethernet and 4G. The data concentrator communicates with the smart meter through the RS485 bus or power line carrier, regularly reads the meter data, and parses, verifies and stores it. At the same time, the data concentrator also has the functions of data preprocessing and event analysis, such as power statistics, electricity fee calculation, load curve drawing and abnormal alarm.
[0090] Finally, the data concentrator uploads the collected load data to the cloud server via Ethernet or 4G network. The cloud server adopts a highly available distributed architecture, including multiple service modules such as data access, data storage, data processing and data visualization. The data access service is responsible for receiving the data uploaded by the data concentrator, and performing decryption, decompression and format conversion; the data storage service adopts a distributed NoSQL database, such as HBase or Cassandra, which can efficiently store and retrieve massive time series data; the data processing service is based on the Hadoop and Spark big data platforms, and realizes data cleaning, conversion, aggregation and mining analysis; the data visualization service provides a variety of charts and reporting tools, such as Grafana and Tableau, which allows users to intuitively view the load operation status.
[0091] Through the load data collection system constructed by smart meters and data concentrators, this solution realizes the automatic collection and centralized management of real-time operation data of multiple loads, laying a data foundation for subsequent load clustering analysis and power consumption pattern mining.
[0092] After obtaining the real-time operation data of multiple loads, this scheme uses data mining technology to perform cluster analysis on the loads and mine the typical power consumption patterns and load characteristics of different types of loads. Due to the high-dimensional, nonlinear, and time-varying characteristics of load data, traditional clustering algorithms such as k-means and hierarchical clustering are difficult to achieve ideal clustering effects. Therefore, this scheme proposes a load clustering analysis method based on the improved DBSCAN clustering algorithm.
[0093] First, the collected real-time load operation data is preprocessed. Since the power consumption behaviors of different types of loads vary greatly, the dimensions and scales of the data are not uniform, and direct clustering is easily affected by the data distribution. Therefore, this solution uses the maximum and minimum value normalization method to map the data of each dimension to the [0,1] interval to eliminate the dimensional difference. At the same time, in order to reduce the impact of data noise and random fluctuations, the sliding window averaging method is used to smooth the data.
[0094] Secondly, extract the characteristics of load data. Load operation data is a typical multivariate time series data, which contains rich temporal and spatial characteristics. This scheme constructs the characteristic vector of load from the perspectives of time dimension and space dimension. In the time dimension, the intraday curve characteristics of load are extracted, such as peak-valley average value, peak-valley difference, load rate, etc., which reflect the volatility and stability of load; the periodic characteristics of load are extracted, such as weekend coefficient, holiday coefficient, etc., which reflect the periodic change law of load. In the spatial dimension, the correlation characteristics of load are extracted, such as the correlation coefficient between load and meteorological factors such as temperature and humidity, which reflects the correlation between load and external environment; the similarity characteristics of load are extracted, such as Euclidean distance and cosine similarity of load curve, which reflects the similarity between different loads.
[0095] Then, the improved DBSCAN clustering algorithm is used to perform cluster analysis on the load feature vectors. The traditional DBSCAN algorithm requires manual setting of two key parameters: density threshold and neighborhood radius. Improper selection will lead to clustering results that are too rough or too fragmented. This scheme proposes an adaptive parameter selection method that dynamically adjusts the density threshold and neighborhood radius according to the distribution characteristics of the data, which can not only detect high-density load clusters, but also distinguish low-density outliers. At the same time, in order to address the problem of slow convergence of the DBSCAN algorithm, this scheme introduces parallel computing and indexing mechanisms, divides large-scale load data into multiple subsets, processes them in parallel on multiple nodes, and uses spatial index structures such as R-trees to accelerate the search process of core objects and boundary objects, thereby improving clustering efficiency.
[0096] Finally, the clustering results are analyzed and explained. By calculating the central load curve and characteristic statistics of each load cluster, the typical power consumption patterns of different types of loads are characterized, such as the "morning peak-noon flat-evening peak" of industrial loads, the "double peak" of commercial loads, and the "single peak" of residential loads. At the same time, data mining algorithms such as decision trees and association rules are used to analyze the key factors that affect the load clustering results, such as industry type, temperature conditions, holidays, etc., and to explore the inherent laws of load distribution.
[0097] The multivariate load clustering analysis method based on improved DBSCAN proposed in this scheme can effectively distinguish different types of loads and discover typical power consumption patterns and load characteristics.
[0098] In an optional embodiment,
[0099] The improved DBSCAN clustering algorithm based on adaptive parameter selection is used to cluster the load feature vectors. By dynamically adjusting the density threshold and neighborhood radius, high-density load clusters are found and low-density outliers are distinguished, including:
[0100] The initial density threshold and neighborhood radius are set according to prior knowledge, and the number of samples in the neighborhood of each load sample point is calculated as the local density;
[0101] According to the mean and standard deviation of the local density of all sample points, the density threshold of each sample point is estimated, and the clustering granularity is adjusted by adjusting the control parameters;
[0102] According to the density threshold of the sample points, find the nearest neighbor sample points in the neighborhood of each sample point, and calculate the average distance as the new neighborhood radius;
[0103] The median of the density thresholds of all sample points is taken as the new global density threshold, and the median of the neighborhood radius of all sample points is taken as the new global neighborhood radius, and the DBSCAN algorithm parameters are adaptively adjusted;
[0104] Use the adaptively adjusted density threshold and neighborhood radius to perform DBSCAN clustering and obtain the clustering results of the load sample points;
[0105] The silhouette coefficient is used to evaluate the clustering quality. If the clustering quality does not meet expectations, the control parameters are iteratively adjusted and the density threshold and neighborhood radius of the sample points are re-estimated until a satisfactory clustering result is obtained.
[0106] For example, the traditional DBSCAN clustering algorithm requires manual setting of two key parameters: density threshold (MinPts) and neighborhood radius (Eps). However, the selection of these two parameters has a great impact on the clustering results, and inappropriate parameters will lead to a decrease in clustering quality. To address this problem, this scheme proposes an improved DBSCAN clustering algorithm based on adaptive parameter selection, which realizes adaptive optimization of load clustering by dynamically adjusting the density threshold and neighborhood radius. The main steps of the algorithm are as follows:
[0107] Step 1: Initialize the density threshold and neighborhood radius. Based on experience or prior knowledge, set the initial density threshold MinPts_0 and neighborhood radius Eps_0 as the starting point for adaptive adjustment.
[0108] Step 2: Calculate the local density of the sample point. For each load sample point, calculate the number of samples in its neighborhood, that is, the local density. Let xi represent the i-th sample point, then its local density is:
[0109] LocalDensity(xi)=|{xj|dist(xi,xj)≤Eps_0}|;
[0110] Among them, dist(xi,xj) represents the distance between sample points xi and xj, which can be measured using Euclidean distance, cosine similarity, etc.
[0111] Step 3: Estimate the density threshold of the sample points. According to the local density distribution of the sample points, estimate the density threshold of each sample point. Let MeanDensity and StdDensity represent the mean and standard deviation of the local density of all sample points, respectively. Then the density threshold of the i-th sample point is:
[0112] MinPts(xi)=MeanDensity+β*StdDensity;
[0113] Among them, β is a control parameter used to adjust the sensitivity of the density threshold. The larger β is, the higher the density threshold is, and the finer the clustering result is; the smaller β is, the lower the density threshold is, and the coarser the clustering result is.
[0114] Step 4: Update the neighborhood radius of the sample point. Update the neighborhood radius of each sample point according to the density threshold of the sample point. For the i-th sample point, find the MinPts(xi) nearest neighboring sample points in its neighborhood and calculate their average distance to xi as the new neighborhood radius:
[0115] Eps(xi)=mean({dist(xi,xj)|xj∈KNN(xi,MinPts(xi))});
[0116] Among them, KNN(xi,k) represents the set of k nearest neighbor sample points of xi.
[0117] Step 5: Adaptively adjust clustering parameters. Adaptively adjust the global parameters MinPts and Eps of the DBSCAN algorithm according to the density threshold and neighborhood radius of the sample points. Take the median of the density threshold of all sample points as the new MinPts, and take the median of the neighborhood radius of all sample points as the new Eps:
[0118] MinPts_new=median({MinPts(xi)|i=1,2,...,n});
[0119] Eps_new=median({Eps(xi)|i=1,2,...,n});
[0120] Step 6: Perform DBSCAN clustering. Use the adaptively adjusted MinPts_new and Eps_new as parameters to perform the standard DBSCAN clustering algorithm and obtain the clustering results of the load sample points.
[0121] Step 7: Evaluate clustering quality. Use indicators such as the Silhouette Coefficient to evaluate the quality of clustering results. The Silhouette Coefficient measures the similarity between a sample point and its cluster and the difference between it and other clusters. The value range is [-1,1]. The larger the value, the better the clustering quality.
[0122] Step 8: Iteratively optimize clustering parameters. If the clustering quality does not meet expectations, return to step 3, adjust the parameter β appropriately, and re-estimate the density threshold and neighborhood radius of the sample points until a satisfactory clustering result is obtained.
[0123] In order to further improve the clustering efficiency, this scheme also introduces parallel computing and indexing mechanisms. In steps 2 and 4, it is necessary to calculate the distance matrix between sample points. When the number of samples is large, the computational overhead is very high. Therefore, a parallel computing framework such as MapReduce is used to distribute the calculation task of the distance matrix to multiple nodes for parallel execution, which significantly reduces the calculation time. At the same time, in step 4, it is necessary to search for the nearest neighbor for each sample point. When the sample dimension is high, the search efficiency is low. Therefore, multidimensional index structures such as R-tree and KD-tree are used to spatially divide and organize sample points, which accelerates the nearest neighbor search process.
[0124] Through adaptive parameter selection and parallel index optimization, the improved DBSCAN clustering algorithm proposed in this scheme can automatically adjust the clustering granularity according to the characteristics of load data, discover high-density load clusters and distinguish low-density outliers, and has high computational efficiency when processing large-scale data, providing high-quality clustering results for subsequent load feature analysis and power consumption pattern mining.
[0125] In an optional embodiment,
[0126] The predicted power generation and the power consumption pattern are input into the risk assessment model based on deep learning, and the intermittent and volatile nature of renewable energy power generation and the uncertainty of multiple loads are comprehensively considered. The risk index of multiple loads in the future is calculated to include:
[0127] The predicted power of renewable energy generation and the electricity consumption patterns of multiple loads are used as input features of the risk assessment model to construct the input data of the model;
[0128] The long short-term memory neural network is used as the basic architecture of the risk assessment model. The attention mechanism is introduced into the network to adaptively allocate the weights of features at different times. A multi-head self-attention layer is designed to model the interactive relationship between renewable energy generation and load power consumption.
[0129] The mean square error is used as the loss function of the model to measure the difference between the predicted risk index and the actual risk index. By minimizing the loss function, the prediction result of the model is made as close as possible to the actual risk level.
[0130] The model is trained using historical data including actual power generation of renewable energy, actual power consumption curves of multiple loads, and actual risk indexes of corresponding time periods, and the Adam optimization algorithm is used to adaptively adjust the model parameters.
[0131] Use the test set data to evaluate the performance of the trained model, and use indicators such as mean absolute error and root mean square error to measure the prediction accuracy of the model;
[0132] The trained risk assessment model is deployed to the power dispatching system to obtain the predicted power of renewable energy generation and the power consumption pattern of multiple loads in real time, input them into the model for risk prediction, and output the risk index for a period of time in the future.
[0133] For example, in a renewable energy power generation system, due to the intermittency and volatility of renewable energy such as wind power and photovoltaic power, as well as the uncertainty of the power demand of multiple loads, the power system faces greater risks in power balance and supply-demand matching. In order to quantify the risk level of multiple loads in the future, this scheme proposes a risk assessment model based on deep learning, which comprehensively considers the predicted power of renewable energy power generation and the power consumption pattern of multiple loads, and calculates the load risk index. The main steps of the model are as follows:
[0134] Step 1: Construct the input features of the risk assessment model. The predicted power of renewable energy generation and the power consumption pattern of multiple loads are used as the input features of the model. Among them, the predicted power of renewable energy generation refers to the expected power generation of renewable energy such as wind power and photovoltaic power in the future.
[0135] Step 2: Design the network structure of the risk assessment model. The long short-term memory neural network (LSTM) is used as the basic architecture of the risk assessment model. LSTM can effectively capture the long-term dependencies in time series data and is suitable for processing the time series characteristics of renewable energy generation and load power consumption. On the basis of LSTM, the attention mechanism is introduced to adaptively allocate the weights of features at different moments and highlight the impact of critical moments. At the same time, at the top layer of the network, a multi-head self-attention layer is designed to model the interactive relationship between renewable energy generation and load power consumption and capture the correlation and constraints between them.
[0136] Step 3: Define the loss function of the risk assessment model. The mean squared error (MSE) is used as the loss function of the model to measure the difference between the predicted risk index and the actual risk index. Let y_pred represent the risk index predicted by the model and y_true represent the actual risk index, then the loss function is:
[0137] Loss=1 / n*∑(y_pred-y_true)^2;
[0138] Where n represents the number of samples. By minimizing the loss function, the prediction results of the model are made as close as possible to the actual risk level.
[0139] Step 4: Train the risk assessment model. Use historical data to train the model. The historical data includes the actual power of renewable energy generation, the actual power consumption curve of multiple loads, and the actual risk index of the corresponding period. Divide the historical data into training set, validation set, and test set. The training set is used to learn model parameters, the validation set is used to tune hyperparameters, and the test set is used to evaluate model performance. During the training process, the Adam optimization algorithm is used to adaptively adjust the model parameters to achieve the best risk prediction effect.
[0140] Step 5: Evaluate the performance of the risk assessment model. Use the test set data to evaluate the performance of the trained model, and use indicators such as Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) to measure the prediction accuracy of the model. MAE reflects the average deviation between the predicted value and the true value, and RMSE reflects the average error between the predicted value and the true value. The calculation formulas for these two indicators are as follows:
[0141] MAE=1 / n*∑|y_pred-y_true|;
[0142] RMSE=sqrt(1 / n*∑(y_pred-y_true)^2);
[0143] Where n represents the number of test samples. The smaller the values of MAE and RMSE, the better the risk prediction performance of the model.
[0144] Step 6: Apply the risk assessment model for online prediction. Deploy the trained risk assessment model to the power dispatching system, obtain the predicted power of renewable energy generation and the power consumption pattern of multiple loads in real time, and input them into the model for risk prediction. The model outputs the risk index for a period of time in the future, which quantitatively describes the risk level of multiple loads. According to the level of the risk index, take corresponding dispatching strategies and control measures, such as adjusting the output of renewable energy generation and optimizing the load side response, to maintain the safe and stable operation of the power system.
[0145] Figure 2 FIG. 1 is a schematic diagram of a structure of a multi-load risk perception system for new energy sources according to an embodiment of the present invention. Figure 2 As shown, the system comprises:
[0146] The first unit is used to obtain real-time operating parameters of new energy power generation equipment in the power system, wherein the real-time operating parameters include power generation power, grid frequency, voltage amplitude and phase angle; based on the real-time operating parameters, a new energy power generation prediction model pre-established by using a machine learning algorithm is used to obtain the predicted power generation power of the new energy power generation equipment in a future period of time;
[0147] The second unit is used to obtain real-time operation data of multiple loads connected to the new energy power generation equipment, wherein the real-time operation data includes the type, power, voltage, current and power factor of the loads; based on the real-time operation data, a data mining technology is used to perform cluster analysis on the multiple loads to obtain power consumption patterns and characteristics of different types of loads;
[0148] The third unit is used to input the predicted power generation and the power consumption pattern into a risk assessment model based on deep learning, comprehensively consider the intermittency and volatility of new energy power generation and the uncertainty of multiple loads, and calculate the risk index of multiple loads in the future.
[0149] According to a third aspect of the embodiments of the present invention,
[0150] An electronic device is provided, comprising:
[0151] processor;
[0152] a memory for storing processor-executable instructions;
[0153] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0154] According to a fourth aspect of the embodiments of the present invention,
[0155] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.
[0156] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.
[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A multi-layer load risk perception method for new energy sources, characterized in that: include: Obtain real-time operating parameters of new energy power generation equipment in the power system; Based on the real-time operating parameters, a new energy power generation prediction model pre-established according to a machine learning algorithm is used to obtain a predicted power generation of the new energy power generation equipment in the future; Acquire real-time operation data of multiple loads connected to the new energy power generation equipment; perform cluster analysis on the multiple loads using data mining technology based on the real-time operation data to obtain power consumption patterns and characteristics of different types of loads; The predicted power generation and the power consumption pattern are input into a risk assessment model based on deep learning, and the intermittency and volatility of renewable energy power generation and the uncertainty of multiple loads are comprehensively considered to calculate the risk index of multiple loads in the future.
2. The method according to claim 1, characterized in that Based on the real-time operating parameters, the predicted power generation of the new energy power generation equipment in the future is obtained by using the new energy power generation power prediction model pre-established by the machine learning algorithm, including: By installing wind speed and direction sensors, rotation speed sensors, torque sensors and pitch angle sensors on the wind turbines, and installing radiation sensors, temperature sensors and current and voltage sensors on the photovoltaic arrays, the real-time operating parameters of the wind turbines and photovoltaic arrays can be collected; Use industrial-grade wireless communication protocols to build a wireless sensor network, transmit the collected real-time operating parameters to the edge server for data analysis, filtering, compression and encryption, and upload to the cloud server; In the cloud server, the real-time operating parameters uploaded by the edge server are obtained through the data access service and stored in the time series database; the stored real-time operating parameters are cleaned and features are constructed to obtain time series features, meteorological features and historical power features, and the feature subset with the highest correlation with the power generation and the lowest redundancy is screened out through the maximum correlation minimum redundancy algorithm; Using the selected feature subsets as input, a machine learning algorithm is used to establish multiple power generation prediction models, and the performance of the prediction models is improved through cross-validation and hyperparameter optimization; Multiple prediction models are combined to form a Stacking ensemble learning model, in which the first layer consists of a long short-term memory neural network, a gradient boosting decision tree, and a random forest. The second layer uses a linear regression algorithm to integrate the prediction results of each model in the first layer by weighted averaging to obtain the final power generation prediction value.
3. The method according to claim 1, characterized in that Obtain real-time operating data of multiple loads connected to renewable energy power generation equipment; Based on the real-time operation data, data mining technology is used to perform cluster analysis on multiple loads, and the power consumption patterns and characteristics of different types of loads are obtained, including: Smart meters are installed at various load locations to measure at least one real-time operating parameter of the load, including voltage, current, active power, reactive power, and power factor, and the measured data is uploaded to a data concentrator through a communication module; the data concentrator periodically reads the measured data of the smart meters through wired or wireless communication, and performs analysis, verification, storage, and preprocessing to generate load operation data; The data concentrator uploads the load operation data to the cloud server, which decrypts, decompresses and converts the received data into a format before storing it in a distributed time series database. The stored load operation data is normalized to its maximum and minimum values to eliminate the dimensional differences of different load data, and the sliding window averaging method is used to smooth and filter the data to remove noise and random fluctuations. Extract the intraday curve characteristics and cycle characteristics of the load from the time dimension, extract the correlation characteristics and similarity characteristics of the load from the space dimension, and construct the load feature vector; The improved DBSCAN clustering algorithm based on adaptive parameter selection is used to cluster the load feature vectors. By dynamically adjusting the density threshold and neighborhood radius, high-density load clusters are found and low-density outliers are distinguished. The clustering results are analyzed and interpreted, the central load curve and characteristic statistics of each load cluster are calculated, and the typical power consumption patterns of different types of loads are characterized.
4. The method according to claim 3, characterized in that The improved DBSCAN clustering algorithm based on adaptive parameter selection is used to perform cluster analysis on the load feature vector. By dynamically adjusting the density threshold and neighborhood radius, high-density load clusters are found and low-density outliers are distinguished, including: The initial density threshold and neighborhood radius are set according to prior knowledge, and the number of samples in the neighborhood of each load sample point is calculated as the local density; According to the mean and standard deviation of the local density of all sample points, the density threshold of each sample point is estimated, and the clustering granularity is adjusted by adjusting the control parameters; According to the density threshold of the sample points, find the nearest neighbor sample points in the neighborhood of each sample point, and calculate the average distance as the new neighborhood radius; The median of the density thresholds of all sample points is taken as the new global density threshold, and the median of the neighborhood radius of all sample points is taken as the new global neighborhood radius, and the DBSCAN algorithm parameters are adaptively adjusted; Use the adaptively adjusted density threshold and neighborhood radius to perform DBSCAN clustering and obtain the clustering results of the load sample points; The silhouette coefficient is used to evaluate the clustering quality. If the clustering quality does not meet expectations, the control parameters are iteratively adjusted and the density threshold and neighborhood radius of the sample points are re-estimated until a satisfactory clustering result is obtained.
5. The method according to claim 1, characterized in that The predicted power generation and the power consumption pattern are input into the risk assessment model based on deep learning, and the intermittent and volatile nature of renewable energy power generation and the uncertainty of multiple loads are comprehensively considered. The risk index of multiple loads in the future is calculated to include: The predicted power of renewable energy generation and the electricity consumption patterns of multiple loads are used as input features of the risk assessment model to construct the input data of the model; The long short-term memory neural network is used as the basic architecture of the risk assessment model. The attention mechanism is introduced into the network to adaptively allocate the weights of features at different times. A multi-head self-attention layer is designed to model the interactive relationship between renewable energy generation and load power consumption. The mean square error is used as the loss function of the model to measure the difference between the predicted risk index and the actual risk index. By minimizing the loss function, the prediction result of the model is made as close as possible to the actual risk level. The model is trained using historical data including actual power generation of renewable energy, actual power consumption curves of multiple loads, and actual risk indexes of corresponding time periods, and the Adam optimization algorithm is used to adaptively adjust the model parameters. Use the test set data to evaluate the performance of the trained model, and use indicators such as mean absolute error and root mean square error to measure the prediction accuracy of the model; The trained risk assessment model is deployed to the power dispatching system to obtain the predicted power of renewable energy generation and the power consumption pattern of multiple loads in real time, input them into the model for risk prediction, and output the risk index for a period of time in the future.
6. The method according to claim 1, characterized in that The real-time operating parameters include generated power, grid frequency, voltage amplitude and phase angle.
7. The method according to claim 1, characterized in that The real-time operation data includes the type, power, voltage, current and power factor of the load.
8. A multi-load risk perception system for new energy, used to implement the method described in any one of claims 1 to 5, characterized in that: include: The first unit is used to obtain real-time operating parameters of new energy power generation equipment in the power system, wherein the real-time operating parameters include power generation power, grid frequency, voltage amplitude and phase angle; based on the real-time operating parameters, a new energy power generation prediction model pre-established by using a machine learning algorithm is used to obtain the predicted power generation power of the new energy power generation equipment in a future period of time; The second unit is used to obtain real-time operation data of multiple loads connected to the new energy power generation equipment, wherein the real-time operation data includes the type, power, voltage, current and power factor of the loads; based on the real-time operation data, a data mining technology is used to perform cluster analysis on the multiple loads to obtain power consumption patterns and characteristics of different types of loads; The third unit is used to input the predicted power generation and the power consumption pattern into a risk assessment model based on deep learning, comprehensively consider the intermittency and volatility of new energy power generation and the uncertainty of multiple loads, and calculate the risk index of multiple loads in the future.
9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.
10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.
Citation Information
Cited By
Electric power acquisition terminal intelligent management system based on data processing
CN119231762A
Electrolytic bath state monitoring method and system based on generalized load equivalent model
CN119307980A
Space load prediction method based on DBSCAN and improved ConvLSTM
CN119726661A
System and method for monitoring and protecting power grid planning based on multi-source data
CN121119786A
Monitoring and protection system and method for power grid planning based on multi-source data
CN121119786B
