WO2020098728A1 - 集群负荷预测方法、装置及存储介质 - Google Patents

集群负荷预测方法、装置及存储介质 Download PDF

Info

Publication number
WO2020098728A1
WO2020098728A1 PCT/CN2019/118309 CN2019118309W WO2020098728A1 WO 2020098728 A1 WO2020098728 A1 WO 2020098728A1 CN 2019118309 W CN2019118309 W CN 2019118309W WO 2020098728 A1 WO2020098728 A1 WO 2020098728A1
Authority
WO
WIPO (PCT)
Prior art keywords
cluster
load
individual
data
predicted
Prior art date
Application number
PCT/CN2019/118309
Other languages
English (en)
French (fr)
Inventor
黎昱
汪建
汤奕
崔晗
Original Assignee
中兴通讯股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Publication of WO2020098728A1 publication Critical patent/WO2020098728A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply

Definitions

  • the present disclosure relates to the field of electric power technology, and in particular, to a cluster load prediction method, device, and storage medium based on distributed power consumption data and external information.
  • Power system load forecasting is very important for power system operation scheduling and power generation planning and arrangement. It is not only an important aspect to ensure the safe and economic operation of the power system and to realize the scientific management and scheduling of the power grid, but also an integral part of the energy management system. Necessary for the commercial operation of the power grid. The traditional research is to analyze and forecast from the overall load and external data, but there is a problem that the forecast roughly ignores the details, which affects the degree of refinement of the cluster load forecast.
  • the main purpose of the present disclosure is to provide a cluster load prediction method, device and storage medium, aiming to improve the refinement of cluster load prediction and the prediction effect.
  • a cluster load prediction method includes: acquiring historical power data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period; learning and fitting the data through machine learning algorithms
  • the individual's historical electricity data and the external data of the same period form the individual's electricity load forecasting model; obtain the external data of the cluster to be predicted, based on the external data of the cluster to be predicted, and pass the individual's use
  • the electric load forecasting model predicts the overall load of the cluster to be predicted and outputs the cluster load forecast value.
  • an embodiment of the present disclosure also proposes a cluster load prediction device, including: a data acquisition module for acquiring historical power data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period; load prediction model Create a module for learning and fitting the individual's historical electricity data and external data in the same period through a machine learning algorithm to form an individual's electricity load forecasting model; a forecast module for acquiring external data of the cluster to be predicted, based on The external data of the cluster to be predicted, and through the individual electricity load prediction model, the overall load of the cluster to be predicted is predicted, and the cluster load prediction value is output.
  • An embodiment of the present disclosure also provides a cluster load prediction device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, when the computer program is executed by the processor The steps of the cluster load prediction method as described above.
  • An embodiment of the present disclosure also proposes a computer-readable storage medium having a computer program stored on it, which when executed by a processor implements the steps of the cluster load prediction method described above.
  • FIG. 1 is a schematic flowchart of a first embodiment of a cluster load prediction method of the present disclosure
  • FIG. 2 is a schematic flowchart of a second embodiment of a cluster load prediction method of the present disclosure
  • FIG. 3 is a schematic flowchart of a third embodiment of a cluster load prediction method of the present disclosure.
  • FIG. 4 is a schematic block diagram of cluster load prediction according to an embodiment of the present disclosure.
  • FIG. 5 is a schematic diagram of a cluster load prediction process involved in an embodiment of the present disclosure
  • FIG. 6 is a schematic diagram of comparison between the resident user load prediction and actual measurement of the cluster load prediction method of the embodiment of the present disclosure
  • FIG. 7 is a schematic diagram comparing the prediction accuracy in test data with other methods of the cluster load prediction method of the embodiment of the present disclosure
  • Example 8 is a schematic structural diagram of an artificial neural network involved in Example 1 of the present disclosure.
  • FIG. 9 is a schematic diagram of test results using the artificial neural network shown in FIG. 8 for testing.
  • FIG. 10 is a schematic diagram of comparing the cumulative results with the actual cluster load power consumption when the cluster user load error correction is performed in Example 1 of the present disclosure
  • FIG. 11 is a schematic diagram of data results after error correction in FIG. 10;
  • Example 12 is a schematic diagram of data regression prediction effect in the training of individual load prediction algorithm in Example 2 of the present disclosure
  • Example 13 is a schematic diagram of the test result of the cluster user load error correction in Example 2 of the present disclosure.
  • Example 14 is a schematic diagram of individual load prediction results in Example 3 of the present disclosure.
  • Example 15 is a schematic diagram of a cluster user load error correction test result in Example 3 of the present disclosure.
  • Example 16 is a schematic diagram of the test result of the cluster user load error correction in Example 4 of the present disclosure.
  • 17 is a schematic diagram of the structure of the device involved in the embodiment of the present disclosure.
  • the main solution of the embodiments of the present disclosure is: by acquiring the historical power data of each individual in the sample cluster corresponding to the cluster to be predicted and the external data of the same period; learning and fitting the historical power data of the individual through the machine learning algorithm and the same External data during the period to form an individual electricity load forecasting model; combined with individual electricity load forecasting model and error correction model, the overall load of the cluster to be predicted is forecasted, full use of the refined measurement data in the power system, and through the machine
  • the learning algorithm mines the data error between the individual and the cluster load, combines the individual power consumption data and weather and holidays and other external information, comprehensively considers the load change trend, improves the cluster load forecasting degree of refinement and prediction effect, and achieves rapid and accurate abnormal users Identify and provide data support for power operations such as power dispatching and power generation planning.
  • the overall load and external data are used to analyze and predict, but there is a problem that the forecast roughly ignores the details, which affects the degree of refinement of the cluster load forecast.
  • the present disclosure provides a solution that can improve the refinement of cluster load prediction and the prediction effect.
  • Mainly use the user electricity information collected by smart meters and external information such as weather and holidays to comprehensively consider the load change trend.
  • Commonly used algorithms include neural networks, support vector machines, extreme learning machines and decision trees, etc., to decompose the large task of cluster load prediction into load predictors for a large number of individuals Task, make full use of the refined measurement data in the power system, and then help the macro load forecast through the micro-research analysis of the individual data, take more measurement data into the load forecast, and improve the refinement of the cluster load forecast Degree and prediction effect, to achieve rapid and accurate discrimination of abnormal users.
  • the first embodiment of the present disclosure proposes a cluster load prediction method, including:
  • Step S101 Obtain historical power consumption data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period;
  • the solution of this embodiment involves predicting the load of the cluster in the power system. Based on the distributed power data and external data, combined with machine learning algorithms, the individual performs micro-load forecasting, and analyzes the relationship between the individual load forecasting result and the overall load data of the cluster. Cluster load prediction results, to achieve rapid and accurate identification of abnormal users.
  • the historical power consumption data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period are obtained.
  • the sample cluster corresponding to the cluster to be predicted may be a cluster of all individuals in a certain area, and the power consumption data of each individual in the sample cluster corresponding to the cluster to be predicted is distributed power consumption data.
  • the historical electricity consumption data of the time period is used as the basis for forecasting the electricity consumption load of the cluster to be predicted at a certain time period in the future.
  • the distributed electricity consumption data mentioned here refers to individual electricity consumption data in the whole electricity consumption, including not limited to residential users for communities, commercial areas for cities, and provincial administrative units for the whole country. Therefore, the application scope of the proposed forecasting method covers such individual and entire power consumers with subordinate relationship, and the overall (cluster) power load is predicted by the individual power load.
  • the historical total load of the cluster to which the individual belongs in the same period as each individual's historical power consumption data must be obtained as the overall load of the sample cluster corresponding to the cluster to be predicted.
  • each individual's historical power consumption data and external data of the same period are obtained from the overall load of the sample cluster.
  • the sample cluster is 873 households in a certain area from June to July in a certain area
  • the overall load of the sample cluster is the total electricity load of 873 households in a certain area from June to July in a certain area
  • the historical electricity consumption data of each individual is a certain area From June to July, the historical electricity consumption data of each user in 873 households in a certain area.
  • the clusters to be predicted are 873 households in a certain area in August in a certain area, and the overall load of the clusters to be forecasted refers to predicting 873 households in a certain area in August in a certain area based on the actual data of the sample cluster (873 areas in a certain area from June to July) Total electricity load.
  • the rules for selecting the sample cluster corresponding to the cluster to be predicted can be set according to the actual situation, for example, a cluster that is separated from the cluster to be predicted by a preset time period (such as 1 month, 10 months, etc.) can be selected as the sample cluster, such as ,
  • the cluster to be predicted is a cluster user from 2013 to 2018, and the sample cluster is a cluster user from 2003 to 2012.
  • external data may include: meteorological data, economic data, and social information, including weather data, One or more of holiday information, public traffic flow conditions and / or regional GDP are selected according to different power consumption scenarios.
  • Weather information here includes but is not limited to temperature, humidity, light and wind speed, etc.
  • Economic data includes but not limited to financial index, macro financial index, etc.
  • Social information includes but not limited to whether it is a weekend, whether it is a statutory holiday and whether there are major social activities .
  • the historical electricity data of each individual in the sample cluster corresponding to the cluster to be predicted and the external data in the same period can be in the following situations:
  • the first power consumption scenario :
  • the distributed electricity data is the electricity load of 873 households in a certain area from June to August in a certain area, and the external data is meteorological data (collected at 1-hour intervals) and holiday information.
  • the cluster load forecasting problem (prediction target) is based on the actual data of June and July (the actual data of the sample cluster corresponding to the cluster to be predicted), and the training model to the cluster load of 873 households per day in August (total load of the cluster to be predicted) Forecast daily consumption.
  • the distributed electricity consumption data is the electricity consumption of 2004-2015 in some provinces, municipalities and autonomous regions (6), and the external data is the regional GDP from 2004-2015.
  • the cluster load forecasting problem (forecasting target) is based on the distributed electricity data (actual data of the sample cluster) from 2004 to 2012 and external data from 2004 to 2015, and predicts the national annual electricity consumption from 2013 to 2015 (to be predicted) Overall cluster load).
  • the third power consumption scenario is the third power consumption scenario:
  • the distributed electricity data is the electricity load curve of each business district in a city
  • the external data is the flow of public transportation (bus, subway, rental) within the coverage of each business district (1 hour accuracy)
  • the meteorological information of the location of each business district (Temperature, rainfall, 1 hour accuracy).
  • the cluster load prediction problem (prediction target) is to train through historical data and predict the electricity load curve of the city's commercial district based on external data for the next day.
  • Step S102 learning and fitting the individual's historical electricity data and external data in the same period through a machine learning algorithm to form an individual's electricity load forecasting model
  • a machine learning algorithm is used to learn and fit the data association relationship between the individual's historical electricity data and external data in the same period to form an individual electricity load forecasting model.
  • the machine learning algorithm may use an artificial neural network or a linear regression algorithm
  • the obtained individual electricity load prediction model may be a neural network model or a linear regression model.
  • the structure of the artificial neural network can contain three layers of neurons, namely the input layer, the hidden layer and the output layer. This structure can fit various complex data relationships.
  • Step S103 Obtain the external data of the cluster to be predicted, based on the external data of the cluster to be predicted, and predict the overall load of the cluster to be predicted through the individual electricity load prediction model, and output the cluster load prediction value.
  • the external data of the cluster to be predicted is obtained, and then, based on the external data of the cluster to be predicted, and through the individual electricity load prediction model, the overall load of the cluster to be predicted is predicted, and the cluster is output Load forecast value.
  • the overall load of the cluster to be predicted may be all electricity consumption data of the cluster to be predicted within a certain period in the future relative to the sample cluster.
  • the current distributed power consumption data is the electricity load of 873 households in a certain area from June to August in a certain area, and the actual acquisition is from a certain area as a sample cluster from June to August.
  • the historical electricity load of 873 households in a certain area in July, the external data is meteorological data (collected at 1-hour intervals) and holiday information.
  • the overall load of the cluster to be predicted is the daily power consumption of the cluster load formed by 873 households every day in August.
  • the external data of the cluster to be predicted is input to the individual electricity load prediction model, the overall load of the cluster to be predicted is predicted, and the cluster load prediction value is output.
  • the training prediction model is applied to the cluster load daily power consumption of 873 households in August (that is, the overall cluster to be predicted Load) to predict, and finally output the predicted value of the daily power consumption of the cluster load formed by 873 households every day in August.
  • the electricity consumption data in June and July and external data are input into the training model to establish a functional relationship between electricity consumption and external data to generate an individual electricity load forecasting model.
  • the external data of August needs to be input into the individual electricity load forecasting model.
  • to predict tomorrow's electricity consumption data first obtain tomorrow's weather data, and input the weather data to obtain tomorrow's electricity data.
  • the individual's historical power consumption data is the electricity consumption of 2004-2012 in some provinces, municipalities and autonomous regions (6) in the country, and the external data in the same period is the regional GDP from 2004-2015 ( GDP). Based on the distributed electricity data from 2004 to 2012 and external data from 2004 to 2015, it is possible to predict the national annual electricity consumption from 2013 to 2015.
  • the present disclosure decomposes the large task of cluster load forecasting into load forecasting subtasks for a large number of individuals, making full use of the refined measurement data in the power system, and thus analyzing the individual data as a macro load Forecasting provides help, taking more measurement data into load forecasting, using refined influence factors to improve the degree of refinement of cluster load forecasting, and building individual electricity load forecasting models through machine learning algorithms, combined with individual electricity consumption Data and external information such as weather and holidays comprehensively consider the load change trend, which improves the cluster load forecasting effect.
  • the use of machine learning algorithms accelerates the load forecasting speed, which can realize the rapid update of short-term load forecasting.
  • the second embodiment of the present disclosure proposes a cluster load prediction method.
  • external data of the cluster to be predicted is obtained based on the cluster to be predicted.
  • External data, and through the individual electricity load prediction model, to predict the overall load of the cluster to be predicted, before outputting the cluster load forecast value it also includes:
  • Step S1031 creating an error correction model between the individual power load and the overall load of the cluster
  • the overall load of the cluster to be predicted is predicted by the individual electricity load prediction model, and the output cluster load prediction value includes:
  • Step S1032 Obtain the external data of the cluster to be predicted, based on the external data of the cluster to be predicted, combined with the individual electricity load prediction model and error correction model, predict the overall load of the cluster to be predicted, and output the cluster load prediction value .
  • a method of machine learning is used to fit the relationship between the individual data of multiple users and the cluster load to create an error correction model between the individual power load and the overall cluster load.
  • the following scheme may be adopted to create an error correction model between the individual power load and the overall load of the cluster:
  • the overall load of the sample cluster is the historical power consumption data of all individuals of the sample cluster corresponding to the statistical cluster to be predicted, and relative to the distributed power consumption data of individuals, the overall load of the sample cluster is centralized data, as shown in FIG. 5.
  • the error correction model may be a linear regression model or an extreme learning machine model.
  • the cluster load prediction value is output.
  • the overall load of the cluster to be predicted may be all electricity consumption data of the cluster to be predicted within a certain period in the future relative to the sample cluster.
  • the current distributed power consumption data is the electricity load of 873 households in a certain area from June to August in a certain area, and the actual acquisition is from a certain area as a sample cluster from June to August.
  • the historical electricity load of 873 households in a certain area in July the external data is meteorological data (collected at 1-hour intervals) and holiday information.
  • the overall load of the cluster to be predicted is the daily power consumption of the cluster load formed by 873 households in August. Based on the actual data (historical electricity consumption data) in June and July, the forecasting model and error correction model are trained for 873 households in August
  • the daily electricity consumption of the formed cluster load (that is, the overall cluster load to be predicted) is predicted.
  • the historical electricity data of each individual in the cluster to be predicted and the external data of the same period are obtained; the historical electricity data of the individual and the data of the external data of the same period are correlated through a machine learning algorithm Learn and fit the relationship to form an individual electricity load prediction model; create an error correction model between the individual electricity load and the overall cluster load; obtain external data of the cluster to be predicted, based on the external data of the cluster to be predicted, and Combined with individual power load forecasting model and error correction model, the overall load of the cluster to be predicted is predicted, and the cluster load forecast value is output to achieve rapid and accurate discrimination of abnormal users, and provide data support for power business such as power dispatching and power generation scheduling.
  • the present disclosure decomposes the large task of cluster load forecasting into load forecasting subtasks for a large number of individuals, making full use of the refined measurement data in the power system, and thus analyzing the individual data as a macro load Forecasting provides help, taking more measurement data into load forecasting, using refined influence factors to improve the degree of refinement of cluster load forecasting, and mining data errors between individuals and cluster loads through machine learning algorithms, combined with individual use Electric data and external information such as weather and holidays comprehensively consider the load change trend, which improves the cluster load forecasting effect.
  • the use of machine learning algorithms accelerates the speed of load forecasting, which can realize the rapid update of short-term load forecasting.
  • a third embodiment of the present disclosure proposes a cluster load prediction method. Based on the embodiment shown in FIG. 2 above, the method in step S101 above: obtains each individual individual in the sample cluster corresponding to the cluster to be predicted After the historical electricity data and external data of the same period, it also includes:
  • Step S1011 Pre-process the individual's historical electricity data and external data in the same period
  • step S102 learning and fitting the individual's historical electricity data and external data of the same period through a machine learning algorithm to form an individual's electricity load prediction model includes:
  • Step S1021 Learning and fitting the pre-processed individual's historical electricity data and external data in the same period through a machine learning algorithm to form an individual electricity load prediction model.
  • preprocessing of the individual's historical electricity data and external data is mainly to deal with abnormal situations such as data loss and data errors, to form standardized electricity data, and to provide accurate data for subsequent load forecasting, thereby increasing cluster load The degree of refinement and prediction of prediction.
  • the preprocessing method can be mean smoothing, adjusting the time scale of electricity data and external data to the same level, quantizing some external data, and normalizing external data to eliminate the effect of absolute value and many more.
  • one of the historical electricity data and external data can be pre-processed as needed.
  • the cluster load prediction value can also be output to the power business platform to provide data support for the power business platform; or, abnormal power consumption based on the cluster load prediction value Behavior detection.
  • the historical power data of the individual in the cluster to be predicted and the external data of the same period are preprocessed; through machine learning The algorithm learns and fits the data association relationship between the individual's historical electricity data and external data in the same period to form an individual electricity load prediction model; creates an error correction model between the individual electricity load and the overall cluster load; Obtain the external data of the cluster to be predicted, based on the external data of the cluster to be predicted, combined with the individual electricity load prediction model and error correction model, predict the overall load of the cluster to be predicted, output the cluster load prediction value, and realize the abnormal user
  • the rapid and accurate discrimination of data provides data support for power dispatching and generation planning and other power businesses.
  • the present disclosure decomposes the large task of cluster load forecasting into load forecasting subtasks for a large number of individuals, making full use of the refined measurement data in the power system, and thus analyzing the individual data as a macro load Forecasting provides help, taking more measurement data into load forecasting, using refined influence factors to improve the degree of refinement of cluster load forecasting, and mining data errors between individuals and cluster loads through machine learning algorithms, combined with individual use Electric data and external information such as weather and holidays comprehensively consider the load change trend, which improves the cluster load forecasting effect.
  • the use of machine learning algorithms accelerates the speed of load forecasting, which can realize the rapid update of short-term load forecasting.
  • FIG. 4 The principle block diagram of the cluster load prediction involved in this embodiment is shown in FIG. 4.
  • the cluster load prediction process involved in this embodiment is shown in FIG. 5.
  • the main process includes:
  • the machine learning algorithm is used to learn and fit the data association relationship between the individual's electricity data and external data to form an individual electricity load forecast model
  • step 3 carry out load prediction for each user separately to obtain the individual load prediction cumulative value; analyze the data association relationship between the individual load prediction cumulative value and the cluster load (cluster load real value) , Perform error fitting to form an error correction model between individual power consumption and overall load.
  • FIG. 6 is a schematic diagram of comparison between the resident user load prediction and actual measurement of the cluster load prediction method of the embodiment of the present disclosure.
  • FIG. 7 is a schematic diagram comparing the prediction accuracy in test data with other methods of the cluster load prediction method of the embodiment of the present disclosure.
  • the embodiments of the present disclosure have the following advantages:
  • the distributed electricity data is the load of 873 households in a certain area from June to August in a certain area (every 5 minutes collection interval), the external data is meteorological data (1 hour collection interval) and holiday information, and the forecast target is based on June and July Actual data, training prediction model predicts the daily power consumption of the cluster load formed by 873 households in August every day.
  • the abnormal value is the nth day;
  • L n is the power consumption of the nth day;
  • m is the average smoothed value range, and it can be set as a positive integer according to the usage requirements;
  • L i is the power consumption of the i day Power.
  • the meteorological information is processed on a daily average basis to form daily meteorological data, including temperature, humidity, rainfall, and air pressure.
  • the data of user 242 on June 4 is taken as an example:
  • an artificial neural network is used to construct an individual load prediction model.
  • the structure of the artificial neural network is shown in FIG. 8, which includes three layers of neurons, which are the input layer and the hidden layer. Layer and output layer, the structure can fit various complex data relationships.
  • the test results are shown in Figure 9, comparing the errors of training, verification and testing respectively, and the prediction errors are mainly concentrated around ⁇ 5%.
  • step two load forecasting is performed for each user, for a total of 873 households. After accumulating the prediction results and comparing with the actual cluster load power consumption, as shown in Figure 10, the 15-day load is predicted with an error of about 8.3%. In order to correct the error, the multiple linear regression model is used to fit the data to construct the error correction model.
  • the formula is as follows:
  • L total is the total power consumption of the real cluster load
  • k i is the regression coefficient
  • L i is the predicted value of the i-th user
  • b is the regression deviation coefficient
  • the load of the cluster to be predicted can be predicted according to the above load prediction model and error correction model, and the cluster load predicted value is output.
  • the embodiments of the present disclosure improve the prediction accuracy of the cluster load.
  • the distributed electricity consumption data is the electricity consumption of 2004-2015 in some provinces, municipalities and autonomous regions (6), and the external data is the regional GDP of 2004-2015.
  • the data is shown in Table 1 below.
  • the problem of load forecasting is based on the distributed electricity data from 2004 to 2012 and external data from 2004 to 2015, and forecasting the national annual electricity consumption from 2013 to 2015.
  • the data source is regular and has been processed.
  • the data is authentic and reliable, and the availability is high, so no preprocessing is required in this embodiment.
  • This embodiment belongs to a long-term load forecast, and the data sample is small. Therefore, the linear regression analysis is used to analyze the data relationship between the annual power consumption of each provincial administrative unit and the regional GDP, and the load forecast model is constructed.
  • the regression formula is as follows:
  • L is the annual power consumption
  • k is the regression coefficient
  • b is the regression deviation coefficient.
  • M is the regional gross product
  • the correlation between electricity consumption and GDP is roughly linear, and there is a constant term used to adjust the deviation.
  • the load of the cluster to be predicted can be predicted according to the above load prediction model and error correction model, and the cluster load predicted value is output.
  • the embodiments of the present disclosure improve the prediction accuracy of the cluster load.
  • Distributed electricity consumption data is the electricity load curve of each business district of a city (5-minute accuracy)
  • external data is the flow of public transportation (bus, subway, rental) within the coverage of each business district (1-hour accuracy) and each business district Location meteorological information (temperature, rainfall, 1 hour accuracy).
  • the problem of cluster load forecasting is to predict the power load curve of the city's commercial districts based on the external data of the next day through the training of historical data.
  • the time scale of electricity data and external data needs to be adjusted to the same level.
  • the external data is extended and the 1-hour precision data is averaged to one data every 5 minutes.
  • P is the traffic density of public transportation
  • N bus is the number of buses per unit time
  • N metro is the number of subways per unit time
  • N taxi is the number of taxis per unit time
  • T is the length of unit time.
  • n hidden is the number of hidden layer nodes
  • n hidden is the number of input attributes
  • n hidden is the number of output results.
  • the prediction result is shown in Figure 14. In the load curve of 24 hours a day, the predicted value and the true value have a high degree of fitting, but the error is relatively large between individual periods such as 17: 00-20: 00 The number of hidden layer nodes of the neural network model.
  • the prediction result of the load curve of a single business district in step two it is accumulated and compared with the load curve of each business district cluster, and error fitting is carried out through an extreme learning machine to construct an error correction model to obtain accurate results, as shown in Figure 15.
  • the relative error of the 24-hour load forecast is less than 5%.
  • the extreme learning machine algorithm selected in this embodiment has the characteristics of fast training speed and high prediction accuracy.
  • the load of the cluster to be predicted can be predicted according to the above load prediction model and error correction model, and the cluster load predicted value is output.
  • the distributed power data is the power data collected by a smart outlet of a power user's home.
  • the external data is temperature, holiday information, and real-time number of people in the household.
  • the prediction goal is to predict the day of the home based on the power data collected by each smart socket and external data. Total electricity consumption.
  • the forecast target is the total daily power consumption, and the original data needs to be processed to form daily typical data.
  • the frequency of data collected by smart sockets needs to be accumulated throughout the day; the temperature data changes greatly throughout the day, and the highest temperature of the day is selected as the attribute ; Holiday information is a logical quantity, which needs to be converted into the final scale factor through the mining of historical data to participate in prediction.
  • the real-time family population is changing, which can be simplified into two attributes: the number of working population and the number of permanent household population.
  • the data format is shown in Table 3 below.
  • the problem of individual load prediction in this embodiment is the estimable prediction of incomplete data, and the power consumption of multiple smart sockets in the home is predicted based on external data.
  • MATLAB is a computer programming language, named after Matrix Laboratory, originally intended to process computer data in a matrix manner, it integrates numerical calculation and visualization environment together, It is very intuitive, and provides a large number of functions, more and more toolboxes make its application range more and more extensive) to optimize parameters and build load forecasting models.
  • L total is the total power consumption of the real cluster load (ie, the total household electricity consumption)
  • n is the number of smart sockets
  • k i is the regression coefficient
  • L i is the predicted value of the ith smart socket
  • b is the regression deviation coefficient
  • the prediction result is shown in Figure 16.
  • the relative errors of training, verification and testing are shown in FIG. 16, and the error distribution of the prediction method used in this embodiment is ⁇ 10%, further illustrating the effectiveness of the cluster load prediction method of the embodiment of the present disclosure.
  • the embodiments of the present disclosure are based on the cluster load forecasting method of distributed power data and external information, combined with machine learning algorithms to perform micro load forecasting on individuals, and analyze the relationship between individual load forecasting results and the overall load data of the cluster.
  • the result of cluster load forecasting improves the degree of refinement and prediction effect of cluster load forecasting, and the use of machine learning algorithms accelerates the speed of load forecasting, which can realize the rapid update of short-term load forecasting.
  • an embodiment of the present disclosure also proposes a cluster load prediction device, including: a data acquisition module for acquiring historical power data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period; load prediction The model creation module is used to learn and fit the individual's historical electricity data and external data of the same period through a machine learning algorithm to form an individual electricity load prediction model; the prediction module is used to pass the individual electricity load The prediction model predicts the overall load of the cluster to be predicted and outputs the predicted value of the cluster load.
  • the apparatus further includes: an error correction model creation module for creating an error correction model between the individual's electricity load and the overall cluster load; the prediction module is also used for combining the individual's use The electric load forecasting model and error correction model predict the overall load of the cluster to be predicted and output the cluster load forecast value.
  • an embodiment of the present disclosure also proposes a cluster load prediction device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being used by the processor
  • the following operations are implemented during execution: acquiring the historical electricity data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period; learning and fitting the individual's historical electricity data through the machine learning algorithm to the same period External data of the individual to form an individual electricity load forecasting model; obtain the external data of the cluster to be predicted, based on the external data of the cluster to be predicted, and predict the overall load of the cluster to be predicted through the individual electrical load forecasting model , Output cluster load forecast value.
  • the cluster load prediction apparatus of this embodiment may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is used to implement connection communication between these components.
  • the user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may further include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • the memory 1005 may be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as a disk memory.
  • the memory 1005 may optionally be a storage device independent of the foregoing processor 1001.
  • FIG. 17 does not constitute a limitation on the device, and may include more or less components than those illustrated, or combine certain components, or have different component arrangements.
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a cluster load prediction program.
  • the network interface 1004 is mainly used to connect to a network server and perform data communication with the network server;
  • the user interface 1003 is mainly used to connect to a client (user side) and perform data communication with the client;
  • the processor 1001 can be used to call the cluster load prediction program stored in the memory 1005, and perform the following operations: obtain the historical power consumption data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period; Combine the individual's historical electricity data and the external data of the same period to form an individual's electricity load forecasting model; obtain the external data of the cluster to be predicted, based on the external data of the cluster to be predicted, and pass the individual
  • the model of electricity load forecasting is used to forecast the overall load of the cluster to be predicted and output the predicted value of the cluster load.
  • the processor 1001 may also be used to call the cluster load prediction program stored in the memory 1005 and perform the following operations: create an error correction model between the individual's electricity load and the overall cluster load; the obtaining is to be predicted The external data of the cluster is based on the external data of the cluster to be predicted, and through the individual electricity load prediction model, the overall load of the cluster to be predicted is predicted, and the step of outputting the predicted value of the cluster load includes:
  • the external data of the cluster to be predicted is obtained, based on the external data of the cluster to be predicted, combined with the individual electricity load forecasting model and error correction model, the overall load of the cluster to be predicted is predicted, and the cluster load forecast value is output.
  • the processor 1001 may also be used to call the cluster load prediction program stored in the memory 1005, and perform the following operations: obtain the historical total of the cluster to which the individual belongs in the same period as the historical electricity data of each individual Load as the overall load of the sample cluster corresponding to the cluster to be predicted; based on the individual's electricity load prediction model, the individual's electricity load in the sample cluster corresponding to the predicted cluster is predicted to obtain the individual electricity load forecast value ; Accumulate the individual electricity load forecast values to obtain the individual load forecast cumulative value; analyze the data association relationship between the individual load forecast cumulative value and the overall load of the sample cluster to form individual electricity load and cluster overall load Error correction model.
  • the processor 1001 may also be used to call the cluster load prediction program stored in the memory 1005, and perform the following operations: pre-process the individual's historical electricity data and external data in the same period; through the machine The learning algorithm learns to fit the pre-processed individual's historical electricity data and external data in the same period to form an individual's electricity load forecasting model.
  • the processor 1001 may also be used to call the cluster load prediction program stored in the memory 1005 and perform the following operations: output the cluster load prediction value to the power business platform to provide data for the power business platform Support; or, based on the cluster load prediction value, perform abnormal power consumption behavior detection.
  • an embodiment of the present disclosure also proposes a computer-readable storage medium that stores a computer program on the computer-readable storage medium.
  • the computer program When the computer program is executed by a processor, the following operations are implemented: acquiring a sample cluster corresponding to the cluster to be predicted Each individual ’s historical electricity data and external data in the same period; learning and fitting the individual ’s historical electricity data and the external data in the same period through a machine learning algorithm to form an individual ’s electricity load forecasting model;
  • the external data of the cluster to be predicted is based on the external data of the cluster to be predicted, and through the individual electricity load prediction model, the overall load of the cluster to be predicted is predicted, and the cluster load prediction value is output.
  • the following operations are also implemented: creating an error correction model between the individual electricity load and the overall load of the cluster; the acquiring external data of the cluster to be predicted is based on the Predict the external data of the cluster, and use the individual electricity load prediction model to predict the overall load of the cluster to be predicted, and the steps of outputting the predicted value of the cluster load include:
  • the external data of the cluster to be predicted is obtained, based on the external data of the cluster to be predicted, combined with the individual electricity load forecasting model and error correction model, the overall load of the cluster to be predicted is predicted, and the cluster load forecast value is output.
  • a cluster load prediction method, device, and storage medium learn and fit an individual ’s historical electricity data and external data in the same period through a machine learning algorithm to form an individual ’s electricity load forecast
  • the model through the individual electricity load forecasting model, the overall load of the cluster to be predicted is predicted, and the cluster load forecast value is output to achieve rapid and accurate discrimination of abnormal users, and provide data support for power operations such as power dispatching and power generation scheduling; in addition
  • the individual load forecasting model and error correction model can be combined to predict the overall load of the cluster to be predicted.
  • the present disclosure decomposes the large task of cluster load forecasting into load forecasting subtasks for a large number of individuals, making full use of the refined measurement data in the power system, and thus analyzing the individual data as a macro load Forecasting provides help, taking more measurement data into load forecasting, using refined influence factors to improve the degree of refinement of cluster load forecasting, and mining data errors between individuals and cluster loads through machine learning algorithms, combined with individual use Electric data and external information such as weather and holidays comprehensively consider the load change trend, which improves the cluster load forecasting effect.
  • the use of machine learning algorithms accelerates the speed of load forecasting, which can realize the rapid update of short-term load forecasting.
  • the present disclosure improves the degree of refinement and prediction effect of cluster load prediction.
  • a cluster load prediction method, device, and storage medium proposed by embodiments of the present disclosure, by acquiring historical power data of each individual in the sample cluster corresponding to the cluster to be predicted and external data of the same period; learning and fitting through machine learning algorithms
  • the individual's historical electricity data and the external data in the same period form the individual's electricity load prediction model; obtain the external data of the cluster to be predicted, based on the external data of the cluster to be predicted, and pass the individual electricity load prediction model, Predict the overall load of the cluster to be predicted, output the cluster load forecast value, realize fast and accurate identification of abnormal users, and provide data support for power dispatching and power generation planning and other power services; in addition, it can also combine individual power load forecasting models and errors Revise the model to predict the overall load of the cluster to be predicted.
  • the present disclosure decomposes the large task of cluster load forecasting into load forecasting subtasks for a large number of individuals, making full use of the refined measurement data in the power system, and thus analyzing the individual data as a macro load Forecasting provides help, taking more measurement data into load forecasting, using refined influence factors to improve the degree of refinement of cluster load forecasting, and mining data errors between individuals and cluster loads through machine learning algorithms, combined with individual use Electric data and external information such as weather and holidays comprehensively consider the load change trend, which improves the cluster load forecasting effect.
  • the use of machine learning algorithms accelerates the speed of load forecasting, which can realize the rapid update of short-term load forecasting.

Abstract

一种集群负荷预测方法、装置及存储介质，其方法包括：获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据（S101）；通过机器学习算法学习拟合个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型（S102）；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值（S103）；其中，还可以结合个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测。

Description

集群负荷预测方法、装置及存储介质
本公开要求享有2018年11月14日提交的名称为“集群负荷预测方法、装置及存储介质”的中国专利申请CN201811350529.0的优先权，其全部内容通过引用并入本文中。
技术领域
本公开涉及电力技术领域，尤其涉及一种基于分布式用电数据与外部信息的集群负荷预测方法、装置及存储介质。
背景技术
电力系统负荷预测对于电力系统运行调度以及发电计划安排是非常重要的内容，它既是保证电力系统安全经济运行和实现电网科学管理及调度的重要方面，又是能量管理系统的组成部分，也是今后进行电网商业化运营所必需的内容。传统研究是从总体负荷及外部数据去进行分析预测，但存在预测粗略忽略细节的问题，影响集群负荷预测的精细化程度。
发明内容
本公开的主要目的在于提供一种集群负荷预测方法、装置及存储介质，旨在提高集群负荷预测的精细化程度及预测效果。
为实现上述目的，本公开提供的一种集群负荷预测方法，包括：获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法学习拟合所述个体的历史用电数据与所述同一时期的外部数据，形成个体的用电负荷预测模型；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
此外，本公开实施例还提出一种集群负荷预测装置，包括：数据获取模块，用于获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；负荷预测模型创建模块，用于通过机器学习算法学习拟合所述个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；预测模块，用于获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
本公开实施例还提出一种集群负荷预测装置，包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述计算机程序被所述处理器执行时实现如上所述的集群负荷预测方法的步骤。
本公开实施例还提出一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如上所述的集群负荷预测方法的步骤。
附图说明
图1是本公开集群负荷预测方法第一实施例的流程示意图；
图2是本公开集群负荷预测方法第二实施例的流程示意图；
图3是本公开集群负荷预测方法第三实施例的流程示意图；
图4是本公开实施例涉及的集群负荷预测原理框图；
图5是本公开实施例涉及的集群负荷预测流程示意图；
图6是本公开实施例集群负荷预测方法的居民用户负荷预测与实际测量对比示意图；
图7是本公开实施例集群负荷预测方法的在测试数据中预测精度与其他方法的对比示意图；
图8是本公开实例1中涉及的人工神经网络结构示意图；
图9是采用图8所示的人工神经网络进行测试的测试结果示意图；
图10是本公开实例1中集群用户负荷误差修正时将预测结果累加后与真实集群负荷用电量进行对比示意图；
图11是图10中经过误差修正后的数据结果示意图；
图12是本公开实例2中个体负荷预测算法训练中数据回归预测效果示意图；
图13是本公开实例2中集群用户负荷误差修正测试结果示意图；
图14是本公开实例3中个体负荷预测结果示意图；
图15是本公开实例3中集群用户负荷误差修正测试结果示意图；
图16是本公开实例4中集群用户负荷误差修正测试结果示意图；
图17是本公开实施例方案涉及的装置结构示意图。
本公开目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
为了使本公开的技术方案更加清楚、明了，下面将结合附图作进一步详述。
具体实施方式
应当理解，此处所描述的实施例仅仅用以解释本公开，并不用于限定本公开。
本公开实施例的主要解决方案是：通过获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法学习拟合个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；结合个体的用电负荷预测模型及 误差修正模型，对待预测集群总体负荷进行预测，充分利用电力系统中的精细化量测数据，且通过机器学习算法挖掘个体与集群负荷间的数据误差，结合个体用电数据与气象和节假日等外部信息，综合考虑负荷变化趋势，提高集群负荷预测的精细化程度及预测效果，实现对异常用户的快速准确判别，为电力调度和发电计划安排等电力业务提供数据支撑。
由于一些情况中，是从总体负荷及外部数据去进行分析预测，但存在预测粗略忽略细节的问题，影响集群负荷预测的精细化程度。
本公开提供一种解决方案，可以提高集群负荷预测的精细化程度及预测效果。主要利用智能电表采集的用户用电信息以及气象和节假日等外部信息，综合考虑负荷变化趋势。涉及数据分析相关技术，包括关联性分析、数值关系拟合和误差分析等。采用各类人工智能算法进行数据分析成为高效的技术手段，常用的算法包括神经网络、支持向量机、极限学习机和决策树等，将集群负荷预测这个大任务分解成对于大量个体的负荷预测子任务，充分利用电力系统中的精细化量测数据，进而通过对个体数据的微观研究分析为宏观负荷预测提供帮助，将更多的量测数据考虑进负荷预测中，提高集群负荷的预测精细化程度及预测效果，实现对异常用户的快速准确判别。
在一个实施例中，如图1所示，本公开第一实施例提出一种集群负荷预测方法，包括：
步骤S101，获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；
本实施例方案涉及对电力系统中集群负荷进行预测，基于分布式用电数据与外部数据，结合机器学习算法对个体进行微观负荷预测，并分析个体负荷预测结果与集群的总体负荷数据关系得出集群负荷预测结果，实现对异常用户的快速准确判别。
在一个实施例中，首先，获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据。
其中，待预测集群对应的样本集群可以是某个地区所有个体的集群，待预测集群对应的样本集群中的每个个体的用电量数据为分布式用电数据，可以根据预测需要统计某一时段的历史用电数据，用来作为对未来某一时段待预测集群用电负荷预测的依据。
此处所提分布式用电数据泛指用电整体中的个体用电数据，包括不限于居民用户之于小区，商业区之于城市，省级行政单位之于全国。因此，所提预测方法的应用范围覆盖此类存在从属关系的用电个体与整体，通过个体用电负荷预测整体(集群)用电负荷。
因此，首先要获取与每个个体的历史用电数据同一时期的个体所属集群的历史总负荷，作为所述待预测集群对应的样本集群总体负荷。
然后，从样本集群总体负荷中获取每个个体的历史用电数据以及同一时期的外部数据。
比如：样本集群是某地区6月至7月某区域873户，样本集群总体负荷为某地区6月 至7月某区域873户的总用电负荷，每个个体的历史用电数据是某地区6月至7月某区域873户中各个用户的历史用电数据。
待预测集群是某地区8月某区域873户，预测待预测集群的总体负荷是指基于样本集群(某地区6月至7月某区域873户)的实际数据预测某地区8月某区域873户的总用电负荷。
其中，选择待预测集群对应的样本集群的规则，可以根据实际情况设定，比如可以选择与待预测集群相隔预设时间段(比如1个月、10个月等)的集群作为样本集群，比如，待预测集群是2013～2018年期间的集群用户，样本集群是2003～2012年期间的集群用户。
此外，考虑用户用电量与气象、经济及其他社会因素的强相关性，需要考虑利用外部数据进行负荷预测，其中，外部数据可以包括：气象数据、经济数据、社会信息，比如包括气象数据、节假日信息、公共交通流量情况和/或地区生产总值中的一种或多种，根据不同的用电场景进行选择。
此处气象信息包括但不限于温度、湿度、光照和风速等，经济数据包括但不限于金融指数、宏观财经指数等，社会信息包括但不限于是否周末、是否法定假日和是否有重大社会活动等。
为了确保用电数据和外部数据的正确映射关系，需要将两者形成一一映射的数据结构。
比如，待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据可以是以下几种情况：
第一种用电场景：
分布式用电数据为某地区6月至8月某区域873户的用电负荷，外部数据为气象数据(1小时间隔采集)和节假日信息。集群负荷预测问题(预测目标)为根据6、7月份实际数据(待预测集群对应的样本集群的实际数据)，训练预测模型对8月份每天的873户形成的集群负荷(待预测集群总体负荷)日用电量进行预测。
第二种用电场景：
分布式用电数据为全国部分省市自治区(6个)的2004～2015年度电力消费量，外部数据为2004～2015年的地区生产总值。集群负荷预测问题(预测目标)为基于2004～2012年的分布式用电数据(样本集群的实际数据)和2004～2015年的外部数据，预测2013～2015年的全国年度电力消费量(待预测集群总体负荷)。
第三种用电场景：
分布式用电数据为某市各商业区用电负荷曲线，外部数据为各商业区覆盖范围内公共交通(公交、地铁、出租)流量情况(1小时精度)，各商业区所处位置气象信息(气温、 雨量，1小时精度)。集群负荷预测问题(预测目标)为通过历史数据的训练，基于未来一天的外部数据预测全市商业区的用电负荷曲线。
步骤S102，通过机器学习算法学习拟合所述个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；
在得到待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据后，通过统计表等方式形成个体的历史用电数据与外部数据(气象数据和节假日信息等)之间的映射关系，得到两者的数据关联关系。
通过机器学习算法对所述个体的历史用电数据与同一时期的外部数据的数据关联关系进行学习拟合，形成个体的用电负荷预测模型。
其中，机器学习算法可以采用人工神经网络或线性回归算法，得到的个体的用电负荷预测模型可以为神经网络模型或线性回归模型。
人工神经网络的结构可以包含三层神经元，分别是输入层、隐含层和输出层，该结构能够拟合各种复杂的数据关系。
步骤S103，获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
在一个实施例中，首先，获取待预测集群的外部数据，然后，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
其中，该待预测的集群总体负荷可以是相对样本集群的未来某一段时间内的待预测集群所有用电数据。
以上述第一种用电场景为例，比如当前统计的分布式用电数据为某地区6月至8月某区域873户的用电负荷，实际获取的是作为样本集群的某地区6月至7月某区域873户的历史用电负荷，外部数据为气象数据(1小时间隔采集)和节假日信息。待预测的集群总体负荷为8月份每天的873户形成的集群负荷日用电量。
在获取待预测集群的外部数据后，将待预测集群的外部数据输入至个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
以上述第一种用电场景为例，根据6、7月份实际数据(历史用电数据)，训练预测模型对8月份每天的873户形成的集群负荷日用电量(即待预测的集群总体负荷)进行预测，最终输出8月份每天的873户形成的集群负荷日用电量的预测值。
也就是说，6、7月用电数据和外部数据输入训练模型，建立用电量和外部数据之间的函数关系，生成个体的用电负荷预测模型，在预测8月份用电量时，只需要将8月份的外部数据输入个体的用电负荷预测模型即可。在实际使用中，比如需要预测明天的用电数据，则首先获取明天气象数据，将气象数据输入，则得到明天的用电数据。
再比如上述第二种用电场景，个体的历史用电数据为全国部分省市自治区(6个)的2004～2012年度电力消费量，同一时期外部数据为2004～2015年的地区生产总值(GDP)。可以基于2004～2012年的分布式用电数据和2004～2015年的外部数据，预测2013～2015年的全国年度电力消费量。
将2004-2012年用电数据和GDP输入训练模型，建立用电量和GDP之间的函数关系，在预测2013年时，只需要将2013年的GDP数据输入模型即可。实际使用中，如预测2019年用电量，则需要知道2019年的GDP，这个值可以由外部提供。
本实施例通过上述方案，通过获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法学习拟合个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值，实现对异常用户的快速准确判别，为电力调度和发电计划安排等电力业务提供数据支撑；此外还可以结合个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测。相比一些情况，本公开将集群负荷预测这个大任务分解成对于大量个体的负荷预测子任务，充分利用了电力系统中的精细化量测数据，从而通过对个体数据的微观研究分析为宏观负荷预测提供帮助，将更多的量测数据考虑进负荷预测中，利用精细化影响因素提高了集群负荷预测的精细化程度，而且通过机器学习算法构建个体的用电负荷预测模型，结合个体用电数据与气象和节假日等外部信息，综合考虑负荷变化趋势，提高了集群负荷的预测效果；此外，利用机器学习算法加快了负荷预测的速度，可以实现短期负荷预测的快速更新。
如图2所示，本公开第二实施例提出一种集群负荷预测方法，基于上述图1所示的实施例，在上述步骤S103，获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值之前还包括：
步骤S1031，创建个体的用电负荷与集群总体负荷间的误差修正模型；
上述步骤S103，通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值包括：
步骤S1032，获取待预测集群的外部数据，基于所述待预测集群的外部数据，并结合所述个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
由于个体用户的负荷预测存在误差且数据预处理过程中剔除了部分用户数据，因此需要建立各用户与集群负荷的映射关系。本实施例采用机器学习的方法，将多个用户的个体数据与集群负荷的关系进行拟合，创建个体的用电负荷与集群总体负荷间的误差修正模 型。
其中，作为一种实施方式，创建个体的用电负荷与集群总体负荷间的误差修正模型可以采用如下方案：
获取与所述每个个体的历史用电数据同一时期的个体所属集群的历史总负荷，作为所述待预测集群对应的样本集群总体负荷；
基于所述个体的用电负荷预测模型对待预测集群对应的样本集群中个体的用电负荷进行预测，得到各个个体的用电负荷预测值；
累计各个个体的用电负荷预测值，得到个体的负荷预测累加值；
分析所述个体的负荷预测累加值与样本集群总体负荷的数据关联关系，进行误差拟合，形成个体的用电负荷与集群总体负荷间的误差修正模型。
其中，样本集群总体负荷为统计的待预测集群对应的样本集群的所有个体历史用电数据，相对个体的分布式用电数据，该样本集群总体负荷为集中式数据，参照图5所示。
其中，误差修正模型可以为线性回归模型或极限学习机模型。
在获得个体的用电负荷预测模型及误差修正模型后，结合所述个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
在一个实施例中，待预测的集群总体负荷可以是相对样本集群的未来某一段时间内的待预测集群所有用电数据。
以上述第一种用电场景为例，比如当前统计的分布式用电数据为某地区6月至8月某区域873户的用电负荷，实际获取的是作为样本集群的某地区6月至7月某区域873户的历史用电负荷，外部数据为气象数据(1小时间隔采集)和节假日信息。待预测的集群总体负荷为8月份每天的873户形成的集群负荷日用电量，根据6、7月份实际数据(历史用电数据)，训练预测模型及误差修正模型对8月份每天的873户形成的集群负荷日用电量(即待预测的集群总体负荷)进行预测。
本实施例通过上述方案，通过获取待预测集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法对所述个体的历史用电数据与同一时期的外部数据的数据关联关系进行学习拟合，形成个体的用电负荷预测模型；创建个体的用电负荷与集群总体负荷间的误差修正模型；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并结合个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值，实现对异常用户的快速准确判别，为电力调度和发电计划安排等电力业务提供数据支撑。相比一些情况，本公开将集群负荷预测这个大任务分解成对于大量个体的负荷预测子任务，充分利用了电力系统中的精细化量测数据，从而通过对个体数据的微观研究分析为宏观负荷预测提供帮助，将更多的量测数据考虑进负荷预测中，利用精细化影响因素提高了集群负荷预测的精细化程度，而且通过机器学习算法挖掘个体与集群负荷 间的数据误差，结合个体用电数据与气象和节假日等外部信息，综合考虑负荷变化趋势，提高了集群负荷的预测效果；此外，利用机器学习算法加快了负荷预测的速度，可以实现短期负荷预测的快速更新。
如图3所示，本公开第三实施例提出一种集群负荷预测方法，基于上述图2所示的实施例，该方法在上述步骤S101：获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据之后，还包括：
步骤S1011，对所述个体的历史用电数据以及同一时期的外部数据进行预处理；
上述步骤S102：通过机器学习算法学习拟合所述个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型包括：
步骤S1021，通过机器学习算法学习拟合预处理后的个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型。
其中，考虑到电力系统配电网中数据流量较大，存在部分用户的用电数据传输丢失或错误等情况，为了防止此类错误信息对于负荷预测的影响，故须通过数据预处理方法将原始数据进行剔除或重新加工。而言，包括校验异常值、检查数据格式和查询数据空缺等，最终确保用电数据的完整性、真实性和可用性。
因此，对所述个体的历史用电数据、外部数据进行预处理主要是对数据缺失、数据错误等异常情况进行处理，形成规范化的用电数据，为后续负荷预测提供精准数据，从而提高集群负荷预测的精细化程度和预测效果。
其中，预处理的方式可以是均值平滑处理、将用电数据和外部数据的时间尺度调整到同一水平、对一些外部数据的量化处理、对外部数据进行归一化，以消除绝对值大小的影响等等。
需要说明的是，根据需要可以选择对历史用电数据、外部数据其中之一进行预处理。
在一个实施例中，在获得待预测集群负荷预测值后，还可以将集群负荷预测值输出至电力业务平台，为所述电力业务平台提供数据支持；或者，基于集群负荷预测值进行异常用电行为检测。
本实施例通过上述方案，通过获取待预测集群中每个个体的历史用电数据以及同一时期的外部数据，对所述个体的历史用电数据以及同一时期的外部数据进行预处理；通过机器学习算法对所述个体的历史用电数据与同一时期的外部数据的数据关联关系进行学习拟合，形成个体的用电负荷预测模型；创建个体的用电负荷与集群总体负荷间的误差修正模型；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并结合个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值，实现对异常用户的快速准确判别，为电力调度和发电计划安排等电力业务提供数据支撑。相比一些情况，本公开将集群负荷预测这个大任务分解成对于大量个体的负荷预测子任 务，充分利用了电力系统中的精细化量测数据，从而通过对个体数据的微观研究分析为宏观负荷预测提供帮助，将更多的量测数据考虑进负荷预测中，利用精细化影响因素提高了集群负荷预测的精细化程度，而且通过机器学习算法挖掘个体与集群负荷间的数据误差，结合个体用电数据与气象和节假日等外部信息，综合考虑负荷变化趋势，提高了集群负荷的预测效果；此外，利用机器学习算法加快了负荷预测的速度，可以实现短期负荷预测的快速更新。
本实施例涉及的集群负荷预测原理框图如图4所示。
本实施例涉及的集群负荷预测流程如图5所示。主要流程包括：
1、采集待预测集群对应的样本集群中每个个体的历史用电量数据，并对数据缺失，数据错误等异常情况进行处理，形成规范化的用电数据，存储至负荷数据库；
2、采集待预测集群对应的样本集群中个体用电数据同一时期的气象数据、经济数据以及节假日信息等外部数据，存储至外部信息数据库，并形成用电数据与外部数据的映射关系；
3、通过机器学习算法对个体的用电数据与外部数据的数据关联关系进行学习拟合，形成个体的用电负荷预测模型；
4、根据步骤3中个体的用电负荷预测模型，对各个用户分别进行负荷预测，得到个体的负荷预测累加值；分析个体的负荷预测累加值与集群负荷(集群负荷真实值)的数据关联关系，进行误差拟合，形成个体用电量与总体负荷间的误差修正模型。
5、结合步骤3和步骤4中形成的负荷预测模型与误差修正模型，对待预测的集群负荷数据进行预测，最终输出集群用户的负荷预测值，为电力调度和发电计划安排等电力业务提供数据支撑。
参照图6，图6是本公开实施例集群负荷预测方法的居民用户负荷预测与实际测量对比示意图。
参照图7，图7是本公开实施例集群负荷预测方法的在测试数据中预测精度与其他方法的对比示意图。
与一些情况相比，本公开实施例具有如下优点：
1、通过提高个体的负荷预测精度，将更多的量测数据考虑进负荷预测中，利用精细化影响因素提高对集群负荷的预测精度；
2、通过机器学习算法挖掘个体与集群负荷间的数据误差，提高了集群负荷的预测效果；
3、利用机器学习算法加快了负荷预测的速度，可以实现短期负荷预测的快速更新。
以下通过实例对本公开实施例方案进行详细阐述：
实例1
分布式用电数据为某地区6月至8月某区域873户的负荷(每5分钟采集间隔)，外部数据为气象数据(1小时采集间隔)和节假日信息，预测目标为根据6、7月份实际数据，训练预测模型对8月份每天的873户形成的集群负荷日用电量进行预测。
一、数据预处理
1)将单个用户的用电数据进行换算，求出每个用户每天的用电量；
2)将用电量异常值进行均值平滑处理，公式为：
Figure PCTCN2019118309-appb-000001
上述公式中，异常值为第n天；L n为第n天的用电量；m为均值平滑的取值天数范围，根据使用需求设为正整数即可；L i为第i天的用电量。
3)将气象信息进行按日均值处理，形成每日气象数据，包括气温、湿度、雨量和气压。
二、个体负荷预测算法训练
根据步骤一中获取的数据，以用户242在6月4日的数据为例结构如下：
用户编号 用电量 气温 湿度 雨量 气压 是否节假日
242 4.31 24 68 0 1008.2 否
由于数据内容较多，不在此罗列。按照此格式整理的数据进行个体的负荷预测，本实施例采用人工神经网络构建个体负荷预测模型，人工神经网络结构如图8所示，其包含了三层神经元，分别是输入层、隐含层和输出层，该结构能够拟合各种复杂的数据关系，测试结果如图9所示，分别将训练、验证和测试的误差进行比较，预测误差主要集中于±5％左右。
三、集群用户负荷误差修正
根据步骤二中个体负荷预测方法，对各个用户分别进行负荷预测，共计873户。将预测结果累加后与真实集群负荷用电量进行对比，如图10所示，对15天的负荷进行了预测，误差约为8.3％。为了修正误差，采用多元线性回归模型对数据进行拟合构建误差修正模型，公式如下：
Figure PCTCN2019118309-appb-000002
式中，L total为真实集群负荷用电总量，k i为回归系数，L i为第i个用户的预测值，b为回归偏差系数。
经过误差修正后的数据结果如图11所示，相对误差降低至1.2％。
之后，即可根据上述负荷预测模型及误差修正模型对待预测集群负荷进行预测，输出集群负荷预测值。
相比一些情况，本公开实施例提高了集群负荷的预测精度。
实例2
分布式用电数据为全国部分省市自治区(6个)的2004～2015年度电力消费量，外部数据为2004～2015的地区生产总值，数据如下表1所示。负荷预测问题为基于2004～2012年的分布式用电数据和2004～2015年的外部数据，预测2013～2015年的全国年度电力消费量。
表1电力消费量与生产总值
Figure PCTCN2019118309-appb-000003
Figure PCTCN2019118309-appb-000004
一、数据预处理
本实施例数据来源正规，已经过处理，数据真实可靠，可用性高，故本实施例中无需进行预处理工作。
二、个体负荷预测算法训练
本实施例属于长时间周期的负荷预测，且数据样本较小，故通过线性回归分析各省级行政单位的年度电力消费量与地区生产总值的数据关系，构建负荷预测模型，回归公式如下：
L＝kM+b
式中，L为年度电力消费量，k为回归系数，b为回归偏差系数。数据回归预测效果如图12所示，M是地区生产总值，用电量和GDP间的相关关系大致为线性关系，且有一个常数项用于调整偏差。
三、集群用户负荷误差修正
由于本实施例中分布式数据只选取了集群数据间部分信息，因此，这是一个不完全分布式数据预测集群用电负荷的问题。单纯累加6个省级行政单位用电信息对于整体而言差异较大，因此需要进行关联性挖掘，本实施例采用极限学习机学习两者间的数据关系，构建误差修正模型，测试结果如图13所示。从2013年到2016年的三年中，负荷预测的相对误差分别为2.5％，3.5％，2.25％，效果较好。
之后，即可根据上述负荷预测模型及误差修正模型对待预测集群负荷进行预测，输出集群负荷预测值。
相比一些情况，本公开实施例提高了集群负荷的预测精度。
实例3
分布式用电数据为某市各商业区用电负荷曲线(5分钟精度)，外部数据为各商业区覆盖范围内公共交通(公交、地铁、出租)流量情况(1小时精度)以及各商业区所处位置气象信息(气温、雨量，1小时精度)。集群负荷预测问题为通过历史数据的训练，基于未来一天的外部数据预测全市商业区的用电负荷曲线。
一、数据预处理
1.根据预测问题所需的用电负荷曲线，需要将用电数据和外部数据的时间尺度调整到同一水平。对外部数据进行外延扩展，将1小时精度的数据平均到每5分钟一个数据。
2.此外对于外部数据进行归一化，以消除绝对值大小的影响。
3.对于公共交通进行量化处理，形成公共交通流量密度指标，公式如下：
Figure PCTCN2019118309-appb-000005
式中，P为公共交通流量密度，N bus为单位时间内公共汽车数量，N metro为单位时间内地铁数量，N taxi为单位时间内出租车数量，T为单位时间长度。处理完的数据结构如下表2所示。
表2实例3数据结构
Figure PCTCN2019118309-appb-000006
二、个体商业区负荷预测算法训练
本实施例中负荷预测影响因素较多，既包括公共交通数据，又包括多种气象指标，且数据量较大，因此考虑采用神经网络构建个体的用电负荷预测模型进行个体商业区负荷预测。通过经验公式选择隐层节点数：
Figure PCTCN2019118309-appb-000007
式中，n hidden为隐层节点数，n hidden为输入属性个数，n hidden为输出结果个数。预测结果如图14所示，在一天24小时的负荷曲线中，预测值与真实值的拟合程度较高，但在个别时段如17时-20时之间误差相对较大，由上述公式得到神经网络模型的隐层节点数。
三、集群用户负荷误差修正
根据步骤二中单个商业区负荷曲线预测结果，进行累加后与各商业区集群负荷曲线进行比较，通过极限学习机进行误差拟合，构建误差修正模型，得到精确结果，如图15所示。24小时的负荷预测相对误差均低于5％。本实施例选用的极限学习机算法具有训练速 度快，预测精度高等特点。
之后，即可根据上述负荷预测模型及误差修正模型对待预测集群负荷进行预测，输出集群负荷预测值。
实例4
分布式用电数据为某电力用户家庭智能插座采集的用电数据，外部数据为气温、节假日信息、家庭实时人数，预测目标为根据各智能插座采集的用电数据及外部数据预测该家庭该日的用电总量。
一、数据预处理
本实例预测目标为日用电总量，需要将原始数据进行处理形成日典型数据，如智能插座采集数据频率较高需要进行全天累加；气温数据全天变化较大，选取当日最高温度作为属性；节假日信息为逻辑量，需要通过历史数据的规律挖掘转变为最终的比例系数参与预测，家庭实时人口处于变化之中，可以简化为工作人口数量与常住家庭人口数量两个属性。数据格式如下表3所示。
表3实例4数据结构
Figure PCTCN2019118309-appb-000008
二、个体负荷预测算法训练
本实施例的个体负荷预测问题为不完全数据的估计性预测，基于外部数据对家庭多个智能插座用电量进行预测。采用决策树进行回归预测，通过MATLAB工具箱(MATLAB是一门计算机编程语言，取名来源于Matrix Laboratory，本意是专门以矩阵的方式来处理计算机数据，它把数值计算和可视化环境集成到一起，非常直观，而且提供了大量的函数，工具箱越来越多，使其应用范围越来越广泛)进行参数优化，构建负荷预测模型。
三、集群用户负荷误差修正
基于步骤二中对家庭各智能插座的用电量预测，挖掘智能插座预测负荷和集群负荷间的数据关系，采用多元线性回归法进行预测，构建误差修正模型，公式如下：
Figure PCTCN2019118309-appb-000009
式中L total为真实集群负荷用电总量(即家庭用电总量)，n为智能插座数量，k i为回归系数，L i为第i个智能插座的预测值，b为回归偏差系数，预测结果如图16所示。图 16中展示了训练、验证和测试的相对误差，本实施例所采用的预测方法的误差分布为±10％，进一步说明了本公开实施例集群负荷预测方法的有效性。
相比一些情况，本公开实施例基于分布式用电数据与外部信息的集群负荷预测方法，结合机器学习算法对个体进行微观负荷预测，并分析个体负荷预测结果与集群的总体负荷数据关系得出集群负荷预测结果，提高了集群负荷预测的精细化程度和预测效果，而且利用机器学习算法加快了负荷预测的速度，可以实现短期负荷预测的快速更新。
对应地，本公开实施例还提出一种集群负荷预测装置，包括：数据获取模块，用于获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；负荷预测模型创建模块，用于通过机器学习算法学习拟合所述个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；预测模块，用于通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
在一个实施例中，所述装置还包括：误差修正模型创建模块，用于创建个体的用电负荷与集群总体负荷间的误差修正模型；所述预测模块，还用于结合所述个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
此外，本公开实施例还提出一种集群负荷预测装置，包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述计算机程序被所述处理器执行时实现如下操作：获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法学习拟合所述个体的历史用电数据与所述同一时期的外部数据，形成个体的用电负荷预测模型；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
在一个实施例中，如图17所示，本实施例集群负荷预测装置可以包括：处理器1001，例如CPU，网络接口1004，用户接口1003，存储器1005，通信总线1002。其中，通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard)，可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如WI-FI接口)。存储器1005可以是高速RAM存储器，也可以是稳定的存储器(non-volatile memory)，例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解，图17中示出的装置结构并不构成对装置的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
如图17所示，作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及集群负荷预测程序。
在图17所示的装置中，网络接口1004主要用于连接网络服务器，与网络服务器进行数据通信；用户接口1003主要用于连接客户端(用户端)，与客户端进行数据通信；而处理器1001可以用于调用存储器1005中存储的集群负荷预测程序，并执行以下操作：获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法学习拟合所述个体的历史用电数据与所述同一时期的外部数据，形成个体的用电负荷预测模型；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
在一个实施例中，处理器1001还可以用于调用存储器1005中存储的集群负荷预测程序，并执行以下操作：创建个体的用电负荷与集群总体负荷间的误差修正模型；所述获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值的步骤包括：
获取待预测集群的外部数据，基于所述待预测集群的外部数据，并结合所述个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
在一个实施例中，处理器1001还可以用于调用存储器1005中存储的集群负荷预测程序，并执行以下操作：获取与所述每个个体的历史用电数据同一时期的个体所属集群的历史总负荷，作为所述待预测集群对应的样本集群总体负荷；基于所述个体的用电负荷预测模型对待预测集群对应的样本集群中个体的用电负荷进行预测，得到各个个体的用电负荷预测值；累计各个个体的用电负荷预测值，得到个体的负荷预测累加值；分析所述个体的负荷预测累加值与所述样本集群总体负荷的数据关联关系，形成个体的用电负荷与集群总体负荷间的误差修正模型。
在一个实施例中，处理器1001还可以用于调用存储器1005中存储的集群负荷预测程序，并执行以下操作：对所述个体的历史用电数据以及同一时期的外部数据进行预处理；通过机器学习算法学习拟合预处理后的个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型。
在一个实施例中，处理器1001还可以用于调用存储器1005中存储的集群负荷预测程序，并执行以下操作：将所述集群负荷预测值输出至电力业务平台，为所述电力业务平台提供数据支持；或者，基于所述集群负荷预测值进行异常用电行为检测。
此外，本公开实施例还提出一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如下操作：获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法学习拟合所述个体的历史用电数据与所述同一时期的外部数据，形成个体的用电负荷预测模型； 获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
在一个实施例中，所述计算机程序被处理器执行时还实现如下操作：创建个体的用电负荷与集群总体负荷间的误差修正模型；所述获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值的步骤包括：
获取待预测集群的外部数据，基于所述待预测集群的外部数据，并结合所述个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
本实施例实现集群负荷预测的原理，请参照上述各实施例，在此不再赘述。
相比一些情况，本公开实施例提出的一种集群负荷预测方法、装置及存储介质，通过机器学习算法学习拟合个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；通过个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值，实现对异常用户的快速准确判别，为电力调度和发电计划安排等电力业务提供数据支撑；此外还可以结合个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测。相比一些情况，本公开将集群负荷预测这个大任务分解成对于大量个体的负荷预测子任务，充分利用了电力系统中的精细化量测数据，从而通过对个体数据的微观研究分析为宏观负荷预测提供帮助，将更多的量测数据考虑进负荷预测中，利用精细化影响因素提高了集群负荷预测的精细化程度，而且通过机器学习算法挖掘个体与集群负荷间的数据误差，结合个体用电数据与气象和节假日等外部信息，综合考虑负荷变化趋势，提高了集群负荷的预测效果；此外，利用机器学习算法加快了负荷预测的速度，可以实现短期负荷预测的快速更新。
本公开提高了集群负荷预测的精细化程度和预测效果。
本公开实施例提出的一种集群负荷预测方法、装置及存储介质，通过获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；通过机器学习算法学习拟合个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值，实现对异常用户的快速准确判别，为电力调度和发电计划安排等电力业务提供数据支撑；此外还可以结合个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测。相比一些情况，本公开将集群负荷预测这个大任务分解成对于大量个体的负荷预测子任务，充分利用了电力系统中的精细化量测数据，从而通过对个体数据的微观研究分析为宏观负荷预测提供帮助，将更多的量测数据考虑进负荷预测中，利用精细化影响因素提高了集群负荷预测的精细化程度，而且通过机器学习算法挖掘个体与集群负荷间的数据误差，结合个体用电数据与气 象和节假日等外部信息，综合考虑负荷变化趋势，提高了集群负荷的预测效果；此外，利用机器学习算法加快了负荷预测的速度，可以实现短期负荷预测的快速更新。
以上所述仅为本公开的优选实施例，并非因此限制本公开的专利范围，凡是利用本公开说明书及附图内容所作的等效结构或流程变换，或直接或间接运用在其它相关的技术领域，均同理包括在本公开的专利保护范围内。

Claims (10)

  1. 一种集群负荷预测方法，其中，包括：
    获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；
    通过机器学习算法学习拟合所述个体的历史用电数据与所述同一时期的外部数据，形成个体的用电负荷预测模型；
    获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
  2. 根据权利要求1所述的方法，其中，所述获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值的步骤之前还包括：
    创建个体的用电负荷与集群总体负荷间的误差修正模型；
    所述获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值的步骤包括：
    获取待预测集群的外部数据，基于所述待预测集群的外部数据，并结合所述个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
  3. 根据权利要求2所述的方法，其中，所述创建个体的用电负荷与集群总体负荷间的误差修正模型的步骤包括：
    获取与所述每个个体的历史用电数据同一时期的个体所属集群的历史总负荷，作为所述待预测集群对应的样本集群总体负荷；
    基于所述个体的用电负荷预测模型对待预测集群对应的样本集群中个体的用电负荷进行预测，得到各个个体的用电负荷预测值；
    累计各个个体的用电负荷预测值，得到个体的负荷预测累加值；
    分析所述个体的负荷预测累加值与所述样本集群总体负荷的数据关联关系，形成个体的用电负荷与集群总体负荷间的误差修正模型。
  4. 根据权利要求1-3中任一项所述的方法，其中，所述获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据的步骤之后，还包括：
    对所述个体的历史用电数据以及同一时期的外部数据进行预处理；
    所述通过机器学习算法学习拟合所述个体的历史用电数据与同一时期的外部数据，形 成个体的用电负荷预测模型的步骤包括：
    通过机器学习算法学习拟合预处理后的个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型。
  5. 根据权利要求4所述的方法，其中，所述外部数据包括：气象数据、经济数据、社会信息中的一种或多种；所述个体的用电负荷预测模型为神经网络模型或线性回归模型；所述误差修正模型为线性回归模型或极限学习机模型。
  6. 根据权利要求1-3中任一项所述的方法，其中，所述方法还包括：
    将所述集群负荷预测值输出至电力业务平台，为所述电力业务平台提供数据支持；或者，基于所述集群负荷预测值进行异常用电行为检测。
  7. 一种集群负荷预测装置，其中，包括：
    数据获取模块，用于获取待预测集群对应的样本集群中每个个体的历史用电数据以及同一时期的外部数据；
    负荷预测模型创建模块，用于通过机器学习算法学习拟合所述个体的历史用电数据与同一时期的外部数据，形成个体的用电负荷预测模型；
    预测模块，用于获取待预测集群的外部数据，基于所述待预测集群的外部数据，并通过所述个体的用电负荷预测模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
  8. 根据权利要求7所述的集群负荷预测装置，其中，所述装置还包括：
    误差修正模型创建模块，用于创建个体的用电负荷与集群总体负荷间的误差修正模型；
    所述预测模块，还用于获取待预测集群的外部数据，基于所述待预测集群的外部数据，并结合所述个体的用电负荷预测模型及误差修正模型，对待预测集群总体负荷进行预测，输出集群负荷预测值。
  9. 一种集群负荷预测装置，其中，包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述计算机程序被所述处理器执行时实现如权利要求1-6中任一项所述的集群负荷预测方法的步骤。
  10. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1-6中任一项所述的集群负荷预测方 法的步骤。
PCT/CN2019/118309 2018-11-14 2019-11-14 集群负荷预测方法、装置及存储介质 WO2020098728A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811350529.0 2018-11-14
CN201811350529.0A CN111191811A (zh) 2018-11-14 2018-11-14 集群负荷预测方法、装置及存储介质

Publications (1)

Publication Number Publication Date
WO2020098728A1 true WO2020098728A1 (zh) 2020-05-22

Family

ID=70707026

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/118309 WO2020098728A1 (zh) 2018-11-14 2019-11-14 集群负荷预测方法、装置及存储介质

Country Status (2)

Country Link
CN (1) CN111191811A (zh)
WO (1) WO2020098728A1 (zh)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111564848B (zh) * 2020-06-09 2022-05-03 剑科云智(深圳)科技有限公司 一种微型电网的智能电力调度方法和用电负荷预测装置
CN112116153B (zh) * 2020-09-18 2022-10-04 上海电力大学 一种耦合Copula与堆叠式LSTM网络的园区多元负荷联合预测方法
CN112232886B (zh) * 2020-10-30 2021-06-04 南方电网能源发展研究院有限责任公司 一种电价概率预测方法、系统、计算机设备和存储介质
CN112633738A (zh) * 2020-12-30 2021-04-09 安徽广志电气有限公司 一种配电控制柜终端负荷分配方法
CN113131476B (zh) * 2021-04-28 2022-06-14 南方电网科学研究院有限责任公司 一种用电负荷预测方法
CN113222290B (zh) * 2021-06-03 2022-11-15 玲珑集团有限公司 一种空压机与密炼机集群的用电负荷协同优化方法
CN113360270B (zh) * 2021-06-30 2024-02-27 杭州数梦工场科技有限公司 一种数据清洗任务处理方法及装置
CN116956198B (zh) * 2023-09-18 2023-12-26 湖南迈格瑞特物联科技有限公司 基于物联网的智慧用电数据分析方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102402726A (zh) * 2011-11-04 2012-04-04 中国电力科学研究院 一种基于区域负荷分析的大规模配电网电力电量预测方法
CN102982387A (zh) * 2012-10-18 2013-03-20 安徽工程大学 一种短期电力负荷预测方法
US20170317495A1 (en) * 2014-10-26 2017-11-02 Green Power Labs Inc. Forecasting net load in a distributed utility grid
CN107622329A (zh) * 2017-09-22 2018-01-23 深圳市景程信息科技有限公司 基于多时间尺度长短时记忆神经网络的电力负荷预测方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104376371B (zh) * 2014-10-31 2017-10-31 国家电网公司 一种基于拓扑的配网分层负荷预测方法
CN104881706B (zh) * 2014-12-31 2018-05-25 天津弘源慧能科技有限公司 一种基于大数据技术的电力系统短期负荷预测方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102402726A (zh) * 2011-11-04 2012-04-04 中国电力科学研究院 一种基于区域负荷分析的大规模配电网电力电量预测方法
CN102982387A (zh) * 2012-10-18 2013-03-20 安徽工程大学 一种短期电力负荷预测方法
US20170317495A1 (en) * 2014-10-26 2017-11-02 Green Power Labs Inc. Forecasting net load in a distributed utility grid
CN107622329A (zh) * 2017-09-22 2018-01-23 深圳市景程信息科技有限公司 基于多时间尺度长短时记忆神经网络的电力负荷预测方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
MA. LI ET AL: "Development of Short-term Load Forecasting System Based on Provincial Power Grid", NORTH CHINA ELECTRIC POWER, vol. 10, 31 October 2003 (2003-10-31) *

Also Published As

Publication number Publication date
CN111191811A (zh) 2020-05-22

Similar Documents

Publication Publication Date Title
WO2020098728A1 (zh) 集群负荷预测方法、装置及存储介质
JP6742894B2 (ja) データ予測システムおよびデータ予測方法
CN110610280A (zh) 一种电力负荷短期预测方法、模型、装置及系统
US20190228329A1 (en) Prediction system and prediction method
JP2020501491A (ja) 動的エネルギーストレージシステム制御のためのシステムおよび方法
Li et al. Federated learning-based short-term building energy consumption prediction method for solving the data silos problem
Khan et al. Genetic algorithm based optimized feature engineering and hybrid machine learning for effective energy consumption prediction
Xu et al. Prediction of thermal energy inside smart homes using IoT and classifier ensemble techniques
Liu et al. Edge sensing data-imaging conversion scheme of load forecasting in smart grid
CN113033110B (zh) 一种基于交通流模型的重点区域人员应急疏散系统及方法
Jin et al. A review and reflection on open datasets of city-level building energy use and their applications
WO2017212880A1 (ja) データ予測システムおよびデータ予測方法
Lan et al. Understanding the relationship between urban morphology and solar potential in mixed-use neighborhoods using machine learning algorithms
CN115907822A (zh) 一种考虑区域及经济影响的负荷特征指标关联性挖掘方法
CN112862172A (zh) 国网95598停电投诉预测方法、装置、计算机设备和存储介质
CN111144628A (zh) 一种分布式供能的冷热电负荷预测模型系统及方法
Grolinger et al. Energy slices: benchmarking with time slicing
Tena-García et al. Implementing data reduction strategies for the optimal design of renewable energy systems
Pham et al. Research on applying machine learning models to predict the electricity generation capacity of rooftop solar energy systems on buildings
Gao et al. A two-stage decision framework for GIS-based site selection of wind-photovoltaic-hybrid energy storage project using LSGDM method
CN111160663B (zh) 一种预测方法及装置
Subramanian et al. Optimal power and reserve capacity procurement policies with deferrable loads
Ghasemi et al. Big Data Analytics in Smart Energy Systems and Networks: A Review
Yue et al. Deep Learning–Based Monitoring Sustainable Decision Support System for Energy Building to Smart Cities with Remote Sensing Techniques
Zou et al. Predicting the electric power consumption of office buildings based on dynamic and static hybrid data analysis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19885095

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13/09/2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19885095

Country of ref document: EP

Kind code of ref document: A1