Virtual power plant load prediction control method and system based on big data

Abstract

The invention discloses a virtual power plant load prediction control method and system based on big data, and relates to the technical field related to intelligent control, and the method comprises the steps: building a big data information base based on a multi-dimensional sensing data set; and introducing power plant environment parameters of the virtual power plant, performing power analysis according to the power plant environment parameters based on the big data information base, and determining a plurality of power characteristics. And performing modeling based on the plurality of power features, constructing a power plant load prediction model, and predicting the virtual power plant through the power plant load prediction model to obtain a load prediction data set. And performing power tracking on the virtual power plant according to the load prediction data set to generate a power distribution diagram. And performing data mining according to the power distribution diagram, making a power dispatching strategy, and executing the power dispatching strategy to perform adaptive control on the virtual power plant. The technical problem that energy utilization efficiency and system operation reliability are difficult to improve due to the fact that a virtual power plant load prediction control method in the prior art is low in load prediction precision is solved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent control, and particularly to a method and system for virtual power plant load prediction and control based on big data. Background Art

[0002] In traditional power systems, the power generation of power plants and the power consumption behaviors on the load side are relatively fixed and centralized. The dispatching department only needs to use a relatively simple prediction model and partial on-site monitoring data to achieve a rough supply-demand balance. However, with the rise of distributed energy systems and demand response technologies, virtual power plants, as a new energy management model, have gradually been applied. A virtual power plant integrates information from multiple dispersed power generation units, energy storage devices, and controllable loads for centralized dispatching to achieve flexible operation similar to that of conventional power plants. Since distributed energy often has volatility and randomness, and there is a strong coupling relationship between the loads in different regions and devices, traditional simple prediction and dispatching methods are difficult to meet the real-time and refined management requirements.

[0003] Therefore, in the prior art, the virtual power plant load prediction and control method has the technical problem that the load prediction accuracy is relatively low, resulting in difficulties in improving energy utilization efficiency and system operation reliability. Summary of the Invention

[0004] This application provides a method and system for virtual power plant load prediction and control based on big data, which solves the technical problem in the prior art that the virtual power plant load prediction and control method has a relatively low load prediction accuracy, resulting in difficulties in improving energy utilization efficiency and system operation reliability. By constructing a big data information library through multi-dimensional sensing data, refining power characteristics by fusing environmental parameters, and combining deep learning and reinforcement learning methods for prediction and dispatching, the ability of the virtual power plant to respond to load peak-valley fluctuations can be significantly improved, achieving the technical effects of higher energy utilization efficiency and system operation reliability.

[0005] This application provides a method for virtual power plant load prediction and control based on big data. The method includes: traversing the virtual power plant for multi-dimensional data collection to obtain a multi-dimensional sensing data set of the virtual power plant, and constructing a big data information library based on the multi-dimensional sensing data set; introducing the power plant environmental parameters of the virtual power plant, and performing power analysis based on the big data information library according to the power plant environmental parameters to determine multiple power characteristics; modeling based on the multiple power characteristics to construct a power plant load prediction model, and predicting the virtual power plant through the power plant load prediction model to obtain a load prediction data set; performing power tracking on the virtual power plant according to the load prediction data set to generate a power distribution map; performing data mining according to the power distribution map to formulate a power dispatching strategy, and executing the power dispatching strategy to perform adaptive control on the virtual power plant.

[0006] In an implementation manner, traverse the virtual power plant for multi-dimensional data collection to obtain a multi-dimensional sensing data set of the virtual power plant. The method includes: collecting data through multiple sensors arranged in the virtual power plant to obtain multiple sensing data; performing time aggregation on the multiple sensing data according to the collection time sequence to generate a sensing time sequence data set; performing spatial aggregation on the multiple sensing data according to the regional information of the virtual power plant to generate a sensing spatial data set; fusing the sensing time sequence data with the sensing spatial data to obtain a sensing spatio-temporal data set; and cleaning the sensing spatio-temporal data set to obtain the multi-dimensional sensing data set.

[0007] In an implementation manner, based on the big data information library, perform power analysis according to the power plant environment parameters to determine multiple power characteristics. The method includes: performing environmental change analysis based on the power plant environment parameters to construct an environmental change trend; mapping the environmental change trend to the big data information library for load assessment to generate a load capacity score; performing principal component analysis on the virtual power plant according to the load capacity score to determine multiple load capacity characteristics; and adding the multiple load capacity characteristics to the multiple power characteristics.

[0008] In an implementation manner, based on the multiple power characteristics, perform modeling to construct a power plant load prediction model. The method includes: performing analysis based on the power plant environment parameters combined with the environmental change trend to determine multiple environmental characteristics; retrieving the historical load data set of the virtual power plant, and partitioning the historical load data set according to the multiple environmental characteristics to obtain a training data set and a test data set; performing feature dimensionality reduction on the training data set to obtain a training dimensionality reduction data set; performing hyperparameter optimization on the training data set through random search to obtain a hyperparameter combination; and performing cross-validation on the hyperparameter combination according to the test data set, and constructing the power plant load prediction model according to the validation result.

[0009] In an implementation manner, perform prediction on the virtual power plant through the power plant load prediction model to obtain a load prediction data set. The method includes: using a long short-term memory network to perform deep learning combined with the multiple load capacity characteristics to obtain a load learning result; performing load capture on the virtual power plant based on the load learning result, performing correlation analysis according to the capture result to generate a load dependence coefficient; performing load fluctuation analysis based on the load dependence coefficient to determine power load demand information; performing load correlation analysis on the virtual power plant according to the load dependence coefficient to determine a load correlation coefficient; using the load correlation coefficient as an index to traverse and match the big data information library, and performing load balance calculation according to the matching result combined with the power load demand information to obtain the load prediction data set.

[0010] In an implementation manner, power tracking is performed on a virtual power plant according to the load prediction data set to generate a power distribution map. The method includes: retrieving a real-time power data set of the virtual power plant from the big data information library, where the real-time power data set includes a real-time load data set; adding the load prediction data set to a data control group, and performing load fluctuation comparison analysis on the real-time load data set according to the data control group to draw a load fluctuation comparison trend; performing load identification based on the load fluctuation comparison trend to determine multiple peak load data points; traversing and marking the virtual power plant according to the multiple peak load data points to generate multiple dynamic identification information; performing power tracking according to the multiple dynamic identification information to generate power dynamic information, and performing coordinate mapping according to the power dynamic information to generate the power distribution map.

[0011] In an implementation manner, data mining is performed according to the power distribution map to formulate a power dispatching strategy. The method includes: setting an expected distribution threshold according to the power dynamic information, and determining whether the power distribution map is greater than or equal to the expected distribution threshold; if the power distribution map is less than the expected distribution threshold, generating an imbalance instruction, and performing association rule mining on the power distribution map through the imbalance instruction to generate a distribution association pattern; optimizing and updating the power distribution map according to the distribution association pattern to obtain an optimized power distribution map; performing linear programming on the virtual power plant according to the optimized power distribution map to generate power constraints; and formulating the power dispatching strategy by performing reinforcement learning according to the power constraints in combination with the multiple power characteristics.

[0012] This application also provides a virtual power plant load prediction control system based on big data, including: an information library construction module for traversing the virtual power plant to collect multi-dimensional data, obtaining a multi-dimensional sensing data set of the virtual power plant, and constructing a big data information library based on the multi-dimensional sensing data set; a power feature determination module for introducing the power plant environment parameters of the virtual power plant and performing power analysis according to the power plant environment parameters based on the big data information library to determine multiple power characteristics; a load prediction module for performing modeling based on the multiple power characteristics, constructing a power plant load prediction model, and predicting the virtual power plant through the power plant load prediction model to obtain a load prediction data set; a power distribution map generation module for performing power tracking on the virtual power plant according to the load prediction data set to generate a power distribution map; and an adaptive control module for performing data mining according to the power distribution map, formulating a power dispatching strategy, and performing adaptive control on the virtual power plant by executing the power dispatching strategy.

[0013] A method and system for virtual power plant load prediction and control based on big data proposed in this application. The method includes: traversing the virtual power plant for multi-dimensional data collection to obtain a multi-dimensional sensing data set of the virtual power plant, and constructing a big data information library based on the multi-dimensional sensing data set; introducing the power plant environment parameters of the virtual power plant, and performing power analysis based on the big data information library according to the power plant environment parameters to determine multiple power characteristics; performing modeling based on the multiple power characteristics to construct a power plant load prediction model, and predicting the virtual power plant through the power plant load prediction model to obtain a load prediction data set; performing power tracking on the virtual power plant according to the load prediction data set to generate a power distribution map; performing data mining according to the power distribution map to formulate a power dispatching strategy, and executing the power dispatching strategy to perform adaptive control on the virtual power plant. This solves the technical problem in the prior art that the virtual power plant load prediction and control method has a low load prediction accuracy, resulting in difficulty in improving the energy utilization efficiency and system operation reliability. By constructing a big data information library through multi-dimensional sensing data, refining power characteristics by integrating environmental parameters, and combining deep learning and reinforcement learning methods for prediction and dispatching, the response ability of the virtual power plant to load peak-valley fluctuations can be significantly improved, achieving the technical effect of higher energy utilization efficiency and system operation reliability. Description of the Drawings

[0014] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the application. It should be understood that the operations in the front or below do not necessarily need to be executed precisely in sequence. On the contrary, according to needs, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0015] Figure 1 Schematic diagram of the flow of a method for virtual power plant load prediction and control based on big data provided by an embodiment of the present application;

[0016] Figure 2 Schematic diagram of the structure of a system for virtual power plant load prediction and control based on big data provided by an embodiment of the present application.

[0017] Description of the reference numerals: Information library construction module 11, power characteristic determination module 12, load prediction module 13, power distribution map generation module 14, adaptive control module 15. Detailed Embodiments

[0018] The above description is only an overview of the technical solution of this application. In order to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the specific embodiments of this application are specifically given below.

[0019] In order to make the purpose, technical solution and advantages of this application clearer, the following will further describe this application in detail with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of this application. All other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0020] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first / second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0021] The embodiments of this application provide a method and system for virtual power plant load prediction control based on big data, as Figure 1 shown. The method includes:

[0022] Traverse the virtual power plant to collect multi-dimensional data, obtain the multi-dimensional sensing data set of the virtual power plant, and build a big data information library based on the multi-dimensional sensing data set; introduce the power plant environment parameters of the virtual power plant, and perform power analysis based on the big data information library according to the power plant environment parameters to determine multiple power characteristics; build a power plant load prediction model based on the multiple power characteristics, and predict the virtual power plant through the power plant load prediction model to obtain a load prediction data set.

[0023] Traverse the virtual power plant for multi-dimensional data collection, obtain the multi-dimensional collected data of sensors in the virtual power plant within the historical time interval, cluster the multi-dimensional collected data to obtain a set composed of all sensing data at the same time and in the same area, and perform data cleaning to obtain the multi-dimensional sensing data set of the virtual power plant. Based on the multi-dimensional sensing data set, construct a big data information library. Among them, the sensors include: voltage, current, power generation load data. Subsequently, introduce the power plant environment parameters of the virtual power plant, and perform power analysis at the corresponding time nodes according to the power plant environment parameters based on the big data information library to determine multiple power characteristics. The environment parameters of the power plant include parameters such as temperature, humidity, wind speed, and light intensity within the historical time interval. Further, based on the multiple power characteristics, build a power plant load prediction model, and predict the virtual power plant through the power plant load prediction model to obtain a load prediction data set.

[0024] The method provided by the embodiment of the present application further includes: collecting data through multiple sensors arranged in the virtual power plant to obtain multiple sensing data; performing time aggregation on the multiple sensing data according to the collection time sequence to generate a sensing time sequence data set; performing spatial aggregation on the multiple sensing data according to the regional information of the virtual power plant to generate a sensing space data set; fusing the sensing time sequence data with the sensing space data to obtain a sensing space-time data set; performing data cleaning on the sensing space-time data set to obtain the multi-dimensional sensing data set.

[0025] Traverse the virtual power plant for multi-dimensional data collection to obtain the multi-dimensional sensing data set of the virtual power plant. The method includes: arranging sensors in each area of the virtual power plant, and collecting data through multiple sensors arranged in the virtual power plant to obtain multiple sensing data. Subsequently, perform time aggregation on the multiple sensing data according to the collection time sequence, and aggregate all sensing data within the same time into one category to generate a sensing time sequence data set. Further, perform spatial aggregation on the multiple sensing data according to the regional information of the virtual power plant, so that multiple sensor data in the same area are aggregated into one category to generate a sensing space data set. Fuse the sensing time sequence data with the sensing space data to obtain a set composed of all sensing data at the same time and in the same area, and obtain a sensing space-time data set. Further, perform data cleaning on the sensing space-time data set to remove sensing data records with duplicate timestamps, sensor data with obvious anomalies, etc., to obtain the multi-dimensional sensing data set.

[0026] The method provided by the embodiment of the present application further includes: performing environmental change analysis based on the power plant environmental parameters, constructing an environmental change trend; mapping the environmental change trend to the big data information library for load assessment, generating a load capacity score; performing principal component analysis on the virtual power plant according to the load capacity score, determining multiple load capacity characteristics; and adding the multiple load capacity characteristics to the multiple power characteristics.

[0027] Performing power analysis based on the big data information library according to the power plant environmental parameters to determine multiple power characteristics. The method includes: performing environmental change analysis based on the power plant environmental parameters, constructing an environmental change trend. When performing environmental change analysis, the environmental data is statistically aggregated at a certain time span (such as hours, days, weeks) to form a basic time series. By performing trend fitting on the environmental data according to the time series, based on the fitting result, the environmental change trend corresponding to each environmental data is obtained. Further, mapping the environmental change trend to the big data information library for load assessment, obtaining the power generation load at each corresponding time node, and obtaining the load capacity score of each time node according to the ratio of the recorded power generation load to the load that the power plant can bear. Further, performing principal component analysis on the environmental change trend of the virtual power plant according to the load capacity score, that is, obtaining principal component analysis according to the environmental parameters corresponding to the load capacity score and the environmental change trend, determining the environmental principal components affecting the load capacity score, determining multiple load capacity characteristics, and adding the multiple load capacity characteristics to the multiple power characteristics.

[0028] The method provided by the embodiment of the present application further includes: analyzing based on the power plant environmental parameters in combination with the environmental change trend to determine multiple environmental characteristics; retrieving the historical load data set of the virtual power plant, partitioning the historical load data set according to the multiple environmental characteristics to obtain a training data set and a test data set; performing feature dimensionality reduction on the training data set to obtain a training dimensionality reduction data set; performing hyperparameter optimization on the training data set through random search to obtain a hyperparameter combination; cross-validating the hyperparameter combination according to the test data set, and constructing the power plant load prediction model according to the validation result.

[0029] Performing modeling based on the multiple power characteristics to construct a power plant load prediction model. The method includes: analyzing based on the power plant environmental parameters in combination with the environmental change trend to obtain the distribution of the power plant environmental parameters in the environmental change trend, partitioning the power plant environmental parameters based on the distribution, and partitioning the power plant environmental parameters into multiple environmental characteristics, and each environmental characteristic corresponds to an environmental parameter interval. Such as high temperature and high humidity environmental characteristics, low temperature and high humidity environmental characteristics, etc.

[0030] Subsequently, a historical load dataset of the virtual power plant is extracted from the big data information database of the virtual power plant. The historical load dataset includes the corresponding load data and environmental data during the historical operation process. The historical load dataset is partitioned according to the multiple environmental characteristics to obtain datasets corresponding to different environmental characteristics, and the datasets corresponding to different environmental characteristics are partitioned into training datasets and test datasets to obtain training datasets and test datasets. When there are too many environmental characteristics and power characteristics, model training may face the situation of insufficient samples or excessive noise. Therefore, feature dimensionality reduction is required. Feature dimensionality reduction is a process of reducing the number of input variables through mathematical or statistical methods while trying to retain the most critical information of the original data. Specifically, principal component analysis can be used for further dimensionality reduction. The training dataset after dimensionality reduction is more conducive to training by machine learning or deep learning algorithms and is also beneficial to reducing the risk of overfitting.

[0031] Furthermore, hyperparameter optimization is performed on the training dataset through random search to obtain a hyperparameter combination, that is, the value range of the hyperparameters is set in advance, and then several combinations are randomly selected for testing during the training process of the training dataset. Finally, the configuration with better performance is selected to obtain the hyperparameter combination. Among them, hyperparameters refer to the parameters that need to be manually set before model training, such as learning rate, regularization coefficient, number of hidden layers of the neural network, depth of the decision tree, etc. After obtaining the hyperparameter combination, cross-validation is performed according to the test dataset. If the cross-validation result meets the expectations (such as low average error and small deviation), this hyperparameter combination can be fixed and the model is retrained with the full amount of training data, and finally a power plant load prediction model is obtained. Among them, cross-validation is a common method for testing model performance. It divides the data into several parts, and each time one part is rotated as the validation set and the rest are used as the training set for repeated training and evaluation, so as to obtain a more stable estimate of the model generalization performance.

[0032] The method provided by the embodiment of the present application further includes: performing deep learning by combining a long short-term memory network with the multiple load capacity characteristics to obtain a load learning result; capturing the load of the virtual power plant based on the load learning result, performing correlation analysis according to the capture result to generate a load dependence coefficient; performing load fluctuation analysis based on the load dependence coefficient to determine the power load demand information; performing load correlation analysis on the virtual power plant according to the load dependence coefficient to determine the load correlation coefficient; using the load correlation coefficient as an index to traverse and match the big data information database, and performing load balance calculation according to the matching result in combination with the power load demand information to obtain the load prediction dataset.

[0033] After completing the power plant load prediction model, it is necessary to predict the current and future load conditions of the virtual power plant. This prediction process not only depends on the output results of the model, but also combines a series of deep learning methods such as long short-term memory networks and the analysis of the correlation between loads. By identifying the interdependent relationships, fluctuation characteristics, and balance states of each load, a more accurate and valuable load prediction data set can be finally output. Historical load data and environmental data related to multiple load capacity characteristics (such as temperature, humidity, wind speed, illuminance, etc.) are obtained from the big data information database of the virtual power plant, and based on the power plant load prediction model, long short-term memory networks are used to perform deep learning training on the historical load data and environmental data related to multiple load capacity characteristics to obtain the load learning results. The load learning results are the load values or load time series obtained by the model.

[0034] Furthermore, based on the load learning results, load capture is performed on the virtual power plant, that is, in the latest prediction or real-time monitoring data, the correlation analysis between different regions in the virtual power plant and the overall prediction results is identified, and their correlation degree is identified to generate a load dependence coefficient. The load dependence coefficient is a quantitative index used to characterize the influence degree or coupling strength between different loads, and its value can generally be normalized between 0 and 1 (or other ranges). After collecting the load dependence coefficients of different regions, it is possible to further analyze which load entities have obvious fluctuation correlations within a certain time range, complete the load fluctuation analysis, and obtain the power load demand information, that is, within the specified prediction time period, the load demand of the overall power plant and each subsystem and the elasticity degree with respect to fluctuation events. According to the load dependence coefficient, load correlation analysis is performed on the virtual power plant to extract other load-related data, and the load correlation coefficient is determined. The higher the dependence coefficient, the higher the correlation coefficient. In the big data information database of the virtual power plant, a large amount of historical operation data, equipment information, dispatching strategies, and their effect records are usually stored. Using the "load correlation coefficient" as an index, historical scenarios or operation modes similar to the current or predicted period are searched, and the big data information database is traversed and matched. Under the conditions of the known total load demand on the demand side and dispatchable resources (power generation, energy storage, demand response, etc.), the dispatching strategies or supply-demand allocation methods of similar historical scenarios are matched to obtain the best or relatively optimized load allocation results. According to the matching results and combined with the power load demand information, load balance calculation is performed to obtain the load prediction data set.

[0035] According to the load prediction data set, power tracking is performed on the virtual power plant to generate a power distribution map; data mining is carried out according to the power distribution map to formulate a power dispatching strategy, and the power dispatching strategy is executed to perform adaptive control on the virtual power plant.

[0036] Perform power tracking on the virtual power plant according to the load prediction data set to generate a power distribution map. Further data mining is performed on the generated power distribution map to discover the load correlation relationships and potential rules between different regions or different time periods. Based on the load distribution and correlation rules, determine specific scheduling objectives (such as maintaining the load within a reasonable range, or ensuring the efficient utilization of distributed energy as much as possible), and formulate feasible power scheduling strategies through methods such as linear programming and reinforcement learning. Among them, linear programming is an optimization method that minimizes or maximizes an objective function (such as total operating cost, energy consumption level, emissions, etc.) under several linear constraints. Once the power scheduling strategy is determined, the corresponding control instructions can be sent to each subsystem such as a power generation unit, a load unit, an energy storage unit, and a demand response terminal through the centralized scheduling system of the virtual power plant. Finally, the goals of reducing operating costs, improving energy utilization efficiency, and balancing power supply and demand are achieved. It solves the technical problem in the prior art that the load prediction control method of the virtual power plant has a low load prediction accuracy, resulting in difficulty in improving the energy utilization efficiency and system operation reliability. By constructing a big data information library through multi-dimensional sensing data, refining power characteristics by fusing environmental parameters, and combining deep learning and reinforcement learning methods for prediction and scheduling, the ability of the virtual power plant to respond to load peak-valley fluctuations can be significantly improved, and the technical effects of higher energy utilization efficiency and system operation reliability can be achieved.

[0037] The method provided by the embodiment of this application further includes: retrieving the real-time power data set of the virtual power plant based on the big data information library, and the real-time power data set includes a real-time load data set; adding the load prediction data set to a data control group, and performing load fluctuation comparison analysis on the real-time load data set according to the data control group to draw a load fluctuation comparison trend; performing load identification based on the load fluctuation comparison trend to determine multiple peak load data points; traversing and marking the virtual power plant according to the multiple peak load data points to generate multiple dynamic identification information; performing power tracking according to the multiple dynamic identification information to generate power dynamic information, and performing coordinate mapping according to the power dynamic information to generate the power distribution map.

[0038] After performing load forecasting on the virtual power plant and obtaining the load forecasting data set, it is also necessary to compare, track, and visually present the forecasting results with real-time power data. The goal of this process is to identify potential peak loads, abnormal fluctuations, etc. in the current and future of the virtual power plant, and generate a "power distribution map". The power distribution map can display the power dynamic information of each area or device of the power plant in a more intuitive way, providing support for subsequent dispatching optimization and decision-making. Retrieve the real-time power data set of the virtual power plant based on the big data information library. The real-time power data set refers to the real-time load data set collected within the current time (or nearly real-time, such as in seconds or minutes). In order to judge the accuracy of the forecast, promptly identify errors and fluctuations, it is necessary to compare the load forecasting data set with the real-time power data. Add the load forecasting data set to the data control group, and conduct load fluctuation comparison analysis on the real-time load data set according to the data control group. Based on the difference between the real-time load data and the forecast data, one or more difference curves can be plotted to obtain the load fluctuation comparison trend. Use the load fluctuation comparison trend to identify loads by using the fluctuation trend to judge key load change events and determine multiple peak load data points. If during the time period from 10:00 to 11:00, the real-time load significantly exceeds the forecast value by more than 10% and maintains for a certain duration (such as 5 minutes or 10 minutes), it can be determined that this time period is a peak load area, and mark the corresponding timestamp and load value in the data. Traverse and mark the virtual power plant according to the multiple peak load data points to generate multiple dynamic identification information. Conduct power tracking based on the multiple dynamic identification information, and continue to monitor the change rules of these marked points in terms of time and space, including subsequent load decline, transfer, and peak duration. During the tracking process, collect the status updates of the above-mentioned marked points in real time, such as whether the load has declined, whether it has been smoothed by the energy storage system, and whether it has been shunted by the dispatching strategy, and then generate power dynamic information. Map the dynamic information (including regional location, load value, time information, peak identification) to the coordinate system of the visualization platform or geographic information system to form a power distribution map.

[0039] The method provided by the embodiment of the present application further includes: setting an expected distribution threshold according to the power dynamic information, and judging whether the power distribution map is greater than or equal to the expected distribution threshold; if the power distribution map is less than the expected distribution threshold, generate an imbalance instruction, and perform association rule mining on the power distribution map through the imbalance instruction to generate a distribution association pattern; optimize and update the power distribution map according to the distribution association pattern to obtain an optimized power distribution map; perform linear programming on the virtual power plant according to the optimized power distribution map to generate power constraints; and perform reinforcement learning according to the power constraints combined with the multiple power characteristics to formulate the power dispatching strategy.

[0040] Data mining is carried out according to the described power distribution map to formulate a power dispatching strategy. The method includes: after visual analysis and real-time monitoring of the power distribution map, the system will first set a desired distribution threshold for a target or ideal state based on the current power dynamic information. The desired distribution threshold is the minimum requirement or target value set for the overall level or balance degree of the power distribution map, which is used to measure whether the current system is in an ideal power consumption or generation distribution state. If the power distribution map of the virtual power plant reaches or exceeds this desired distribution threshold, at this time, the current energy supply and demand are relatively balanced, the load distribution in each region is reasonable, and the key equipment is within the safe operation range. In this case, no additional in-depth adjustment or intervention is required. However, when the system finds that the power distribution map is smaller than the set desired distribution threshold (i.e., there is an obvious load imbalance or supply-demand mismatch), an imbalance instruction is automatically generated. The imbalance instruction is used to perform association rule mining on the power distribution map. Using the historical data accumulated in the big data information database in the early stage and the current power dynamic information, potential association factors leading to imbalance or unreasonable distribution are found. For example, a sudden increase in electricity consumption in a certain region may be related to the improper staggering of the equipment shutdown time in another region, or related to the failure of the energy storage system to release electricity in time during certain periods. This mining process will output a series of distribution association patterns, such as when the load in Workshop A exceeds X kW, the energy storage module C needs to increase the discharge power by Y kW within 5 minutes and other rule relationships. After obtaining these distribution association patterns, the system will apply them to the current power distribution map for optimization and update, so as to obtain an optimized power distribution map. Based on the existing data and rules, the dispatching methods that may cause local overload or resource waste are corrected to make the overall power distribution more balanced and reasonable. According to the updated optimized power distribution map, through mathematical optimization means such as linear programming, the overall energy flow design of the virtual power plant is carried out to generate a set of power constraints. The power constraints cover multiple dimensions, for example: the maximum or minimum power limit of different load regions at specified times; the charge and discharge rate and capacity limit of the energy storage unit; the consumption requirement or grid connection capacity limit when renewable energy is accessed; the electricity consumption reduction target when the demand response mechanism is triggered. On this basis, these power constraints are considered together with various power characteristics of the virtual power plant (such as peak-valley ratio, seasonal fluctuation, energy storage response efficiency, renewable energy penetration rate, etc.), and a more intelligent and adaptive power dispatching strategy is formulated through reinforcement learning. The so-called reinforcement learning means that the system searches for the optimal or approximately optimal decision-making scheme in the process of continuous trial and error and feedback: when a certain dispatching action (such as charging in advance or peak shaving) can effectively reduce the operating cost, reduce the peak electricity consumption or improve the utilization rate of renewable energy, the system will give a corresponding positive incentive to this action and be more inclined to take similar operations in subsequent dispatching. On the contrary, if a certain dispatching leads to an increase in energy consumption cost or the state of the energy storage system cannot be maintained, a negative incentive will be generated to promote the system to correct this strategy or avoid using it.By iterating in this way, an optimal "power dispatching strategy" can ultimately be found, taking into account both economy and security. It can respond promptly to load imbalance, avoid energy waste or safety hazards, and continuously accumulate effective experience during long-term operation, thereby enhancing the overall reliability and economic benefits of the power plant.

[0041] In the above text, with reference to Figure 1 a load forecasting and control method for a virtual power plant based on big data according to an embodiment of the present invention was described in detail. Next, with reference to Figure 2 a load forecasting and control system for a virtual power plant based on big data according to an embodiment of the present invention will be described.

[0042] A load forecasting and control system for a virtual power plant based on big data according to an embodiment of the present invention solves the technical problem in the prior art that the load forecasting accuracy of the load forecasting and control method for a virtual power plant is relatively low, resulting in difficulty in improving the energy utilization efficiency and system operation reliability. By constructing a big data information library through multi-dimensional sensing data, refining power characteristics by integrating environmental parameters, and combining deep learning and reinforcement learning methods for prediction and scheduling, the ability of the virtual power plant to respond to load peak-valley fluctuations can be significantly improved, achieving the technical effect of higher energy utilization efficiency and system operation reliability. A load forecasting and control system for a virtual power plant based on big data includes: an information library construction module 11, a power characteristic determination module 12, a load forecasting module 13, a power distribution map generation module 14, and an adaptive control module 15.

[0043] The information library construction module 11 is used to traverse the virtual power plant to collect multi-dimensional data, obtain a multi-dimensional sensing data set of the virtual power plant, and construct a big data information library based on the multi-dimensional sensing data set;

[0044] The power characteristic determination module 12 is used to introduce the power plant environmental parameters of the virtual power plant, perform power analysis based on the big data information library according to the power plant environmental parameters, and determine multiple power characteristics;

[0045] The load forecasting module 13 is used to build a power plant load forecasting model based on the multiple power characteristics, and predict the virtual power plant through the power plant load forecasting model to obtain a load forecasting data set;

[0046] The power distribution map generation module 14 is used to perform power tracking on the virtual power plant according to the load forecasting data set and generate a power distribution map;

[0047] The adaptive control module 15 is used to perform data mining according to the power distribution map, formulate a power dispatching strategy, and execute the power dispatching strategy to perform adaptive control on the virtual power plant.

[0048] Next, the specific configuration of the information database construction module 11 will be described in detail. The information database construction module 11 may further include: traversing the virtual power plant to collect multi-dimensional data, obtaining a multi-dimensional sensing data set of the virtual power plant. The method includes: collecting data through multiple sensors deployed in the virtual power plant to obtain multiple sensing data; performing time aggregation on the multiple sensing data according to the collection time sequence to generate a sensing time sequence data set; performing spatial aggregation on the multiple sensing data according to the regional information of the virtual power plant to generate a sensing space data set; fusing the sensing time sequence data and the sensing space data to obtain a sensing spatio-temporal data set; cleaning the sensing spatio-temporal data set to obtain the multi-dimensional sensing data set.

[0049] Next, the specific configuration of the power feature determination module 12 will be further described in detail. The power feature determination module 12 further includes: performing power analysis based on the big data information database according to the power plant environment parameters to determine multiple power features. The method includes: performing environmental change analysis based on the power plant environment parameters to construct an environmental change trend; mapping the environmental change trend to the big data information database for load assessment to generate a load capacity score; performing principal component analysis on the virtual power plant according to the load capacity score to determine multiple load capacity features; adding the multiple load capacity features to the multiple power features.

[0050] Next, the specific configuration of the load prediction module 13 will be described in detail. The load prediction module 13 may further include: building a power plant load prediction model based on the multiple power features. The method includes: analyzing based on the power plant environment parameters combined with the environmental change trend to determine multiple environmental features; retrieving the historical load data set of the virtual power plant, dividing the historical load data set according to the multiple environmental features to obtain a training data set and a test data set; performing feature dimensionality reduction on the training data set to obtain a training dimensionality reduction data set; performing hyperparameter optimization on the training data set through random search to obtain a hyperparameter combination; performing cross-validation on the hyperparameter combination according to the test data set, and constructing the power plant load prediction model according to the verification result.

[0051] Next, the specific configuration of the load prediction module 13 will be described in detail. The load prediction module 13 further includes: predicting the virtual power plant through the power plant load prediction model to obtain a load prediction data set. The method includes: performing deep learning by using a long short-term memory network in combination with the multiple load capacity characteristics to obtain a load learning result; capturing the load of the virtual power plant based on the load learning result, performing correlation analysis according to the capture result to generate a load dependence coefficient; performing load fluctuation analysis based on the load dependence coefficient to determine the power load demand information; performing load correlation analysis on the virtual power plant according to the load dependence coefficient to determine the load correlation coefficient; using the load correlation coefficient as an index to traverse and match the big data information library, and performing load balance calculation according to the matching result in combination with the power load demand information to obtain the load prediction data set.

[0052] Next, the specific configuration of the power distribution map generation module 14 will be further described in detail. The power distribution map generation module 14 further includes: performing power tracking on the virtual power plant according to the load prediction data set to generate a power distribution map. The method includes: retrieving the real-time power data set of the virtual power plant from the big data information library, where the real-time power data set includes a real-time load data set; adding the load prediction data set to the data control group, and performing load fluctuation comparison analysis on the real-time load data set according to the data control group to draw a load fluctuation comparison trend; performing load identification based on the load fluctuation comparison trend to determine multiple peak load data points; traversing and marking the virtual power plant according to the multiple peak load data points to generate multiple dynamic identification information; performing power tracking according to the multiple dynamic identification information to generate power dynamic information, and performing coordinate mapping according to the power dynamic information to generate the power distribution map.

[0053] Next, the specific configuration of the adaptive control module 15 will be described in detail. The adaptive control module 15 further includes: performing data mining according to the power distribution map to formulate a power dispatching strategy. The method includes: setting an expected distribution threshold according to the power dynamic information, and determining whether the power distribution map is greater than or equal to the expected distribution threshold; if the power distribution map is less than the expected distribution threshold, generating an imbalance instruction, and performing association rule mining on the power distribution map through the imbalance instruction to generate a distribution association pattern; optimizing and updating the power distribution map according to the distribution association pattern to obtain an optimized power distribution map; performing linear programming on the virtual power plant according to the optimized power distribution map to generate power constraints; and formulating the power dispatching strategy by performing reinforcement learning according to the power constraints in combination with the multiple power characteristics.

[0054] The virtual power plant load prediction and control system provided by an embodiment of the present invention can execute a virtual power plant load prediction and control method provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution of the method.

[0055] Although this application makes various references to certain modules in the system according to the embodiments of this application, however, any number of different modules can be used and run on the user terminal and / or server. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0056] The above specific implementation manners do not constitute a limitation to the protection scope of this application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. A virtual power plant load prediction control method based on big data, characterized in that: The method comprises: Traversing the virtual power plant to collect multidimensional data, obtaining a multidimensional sensor data set of the virtual power plant, and building a big data information library based on the multidimensional sensor data set; Introducing power plant environmental parameters of the virtual power plant, performing power analysis according to the power plant environmental parameters based on the big data information database, and determining a plurality of power characteristics; Modeling is performed based on the multiple power characteristics to construct a power plant load prediction model, and prediction is performed on the virtual power plant through the power plant load prediction model to obtain a load prediction data set; Tracking power of the virtual power plant according to the load forecast data set to generate a power distribution map; Data mining is performed according to the power distribution map, a power dispatching strategy is formulated, and the power dispatching strategy is executed to perform adaptive control on the virtual power plant.

2. A virtual power plant load prediction control method based on big data as claimed in claim 1, characterized in that: Traversing the virtual power plant to collect multi-dimensional data, and obtaining a multi-dimensional sensor data set of the virtual power plant, the method includes: Data is collected through multiple sensors deployed in the virtual power plant to obtain multiple sensor data; Performing temporal aggregation on the plurality of sensor data according to the collection time sequence to generate a sensor time series data set; Spatially aggregating the plurality of sensor data according to regional information of the virtual power plant to generate a sensor spatial data set; Fusing the sensing time series data with the sensing space data to obtain a sensing spatiotemporal data set; The sensing spatiotemporal data set is cleaned to obtain the multidimensional sensing data set.

3. The method for load prediction and control of a virtual power plant based on big data according to claim 1, characterized in that: Based on the big data information base, power analysis is performed according to the power plant environmental parameters to determine multiple power characteristics, and the method includes: Perform environmental change analysis based on the environmental parameters of the power plant and construct environmental change trends; Mapping the environmental change trend to the big data information database for load assessment to generate a load capacity score; Performing principal component analysis on the virtual power plant according to the load capacity score to determine a plurality of load capacity characteristics; The plurality of load capability characteristics are added to the plurality of power characteristics.

4. A virtual power plant load prediction control method based on big data as claimed in claim 3, characterized in that: Modeling is performed based on the multiple power characteristics to construct a power plant load prediction model, the method comprising: Analyze the power plant environmental parameters in combination with the environmental change trend to determine multiple environmental characteristics; Retrieving a historical load data set of the virtual power plant, dividing the historical load data set according to the multiple environmental characteristics, and obtaining a training data set and a test data set; Perform feature dimensionality reduction based on the training data set to obtain a training dimensionality reduction data set; Performing hyperparameter optimization on the training data set by random search to obtain a hyperparameter combination; The hyperparameter combination is cross-validated according to the test data set, and the power plant load prediction model is constructed according to the validation results.

5. The method for load prediction and control of a virtual power plant based on big data according to claim 3, characterized in that: The load forecasting model of the power plant is used to forecast the virtual power plant to obtain a load forecasting data set, and the method includes: Using a long short-term memory network combined with the multiple load capacity characteristics to perform deep learning to obtain a load learning result; Capturing the load of the virtual power plant based on the load learning result, performing correlation analysis according to the captured result, and generating a load dependency coefficient; Perform load fluctuation analysis based on the load dependence coefficient to determine power load demand information; Performing load correlation analysis on the virtual power plant according to the load dependency coefficient to determine the load correlation coefficient; The load correlation coefficient is used as an index to traverse and match the big data information library, and a load balance calculation is performed based on the matching result combined with the power load demand information to obtain the load prediction data set.

6. The method for load prediction and control of a virtual power plant based on big data according to claim 1, characterized in that: The power of the virtual power plant is tracked according to the load forecast data set to generate a power distribution map, the method comprising: Retrieving a real-time power data set of a virtual power plant based on the big data information database, wherein the real-time power data set includes a real-time load data set; Adding the load forecast data set to a data control group, performing load fluctuation control analysis on the real-time load data set according to the data control group, and drawing a load fluctuation control trend; Perform load identification based on the load fluctuation control trend to determine multiple peak load data points; Traversing and marking the virtual power plant according to the multiple peak load data points to generate multiple dynamic identification information; Electric power tracking is performed according to the multiple dynamic identification information to generate electric power dynamic information, and coordinate mapping is performed according to the electric power dynamic information to generate the electric power distribution map.

7. A virtual power plant load prediction and control method based on big data as claimed in claim 6, characterized in that: Data mining is performed according to the power distribution map to formulate a power dispatch strategy, the method comprising: Setting an expected distribution threshold according to the power dynamic information, and determining whether the power distribution diagram is greater than or equal to the expected distribution threshold; If the power distribution graph is less than the expected distribution threshold, an imbalance instruction is generated, and association rule mining is performed on the power distribution graph through the imbalance instruction to generate a distribution association pattern; Optimizing and updating the power distribution map according to the distribution association mode to obtain a power optimization distribution map; Performing linear programming on the virtual power plant according to the power optimization distribution map to generate power constraints; Reinforcement learning is performed according to the power constraints and the multiple power characteristics to formulate the power dispatch strategy.

8. A virtual power plant load forecasting control system based on big data, characterized in that: The system comprises: An information base construction module is used to traverse the virtual power plant to collect multi-dimensional data, obtain a multi-dimensional sensor data set of the virtual power plant, and construct a big data information base based on the multi-dimensional sensor data set; A power feature determination module, used for introducing power plant environmental parameters of the virtual power plant, performing power analysis according to the power plant environmental parameters based on the big data information library, and determining a plurality of power features; A load forecasting module, used to perform modeling based on the multiple power characteristics, construct a power plant load forecasting model, forecast the virtual power plant through the power plant load forecasting model, and obtain a load forecasting data set; A power distribution map generation module, used to track the power of the virtual power plant according to the load forecast data set and generate a power distribution map; The adaptive control module is used to perform data mining according to the power distribution map, formulate a power dispatching strategy, and execute the power dispatching strategy to perform adaptive control on the virtual power plant.

Images