Small sample demand response behavior day-ahead prediction method and system under multiple influence factors

Abstract

The invention discloses a small sample demand response behavior day-ahead prediction method and system under multiple influence factors, and the method comprises the steps: obtaining the electricity price, multiple influence factors, user baseline load and demand response historical data of a target region in a prediction time period; establishing a demand response behavior prediction model based on a data-driven neural network; the trained demand response behavior prediction model takes the electricity price, various influence factors, demand response historical data and user baseline load in the prediction time period as input, and outputs a demand response behavior prediction value in the prediction time period; establishing a demand response behavior day-ahead prediction model based on a Gaussian kernel density estimation method; and the trained demand response behavior day-ahead prediction model takes the demand response behavior prediction value in the prediction time period as input, outputs probability distribution of the demand response behavior in the prediction time period, determines demand response behavior day-ahead prediction results of each prediction interval, and improves the load demand response behavior day-ahead prediction accuracy under a small sample.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system data processing, and specifically relates to a method, device, storage medium and equipment for day-ahead prediction of small sample demand response behavior considering multiple influencing factors. Background Art

[0002] In the prior art, deep learning methods, such as convolutional neural networks (CNN) and long short-term memory networks (LSTM), are widely used for time series data analysis and prediction in the day-ahead forecasting of demand response behavior. Load forecasting, electricity price pricing, etc. have an impact on day-ahead forecasting. In addition, the forecasting method is key, including Gaussian kernel density estimation, which is used to estimate demand response at different confidence levels in order to more fully understand electricity price changes.

[0003] However, existing technologies require a large amount of sample data and fail to fully explore the complex nonlinear relationship between load and demand response incentive electricity prices, resulting in the model being unable to capture key features and trends. Existing technologies lack sufficient consideration of spatiotemporal correlations, and changes in load and demand response incentive electricity prices are often affected by a variety of factors, including seasonality, weather conditions, etc. Existing technologies are sensitive to outliers and noise in load and demand response incentive electricity prices, which affects the robustness and generalization ability of the model. Existing technologies fail to effectively handle the uncertainty of prediction results, which is a common and important factor in the power market. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a method and system for day-ahead prediction of small sample demand response behavior under multiple influencing factors, so as to improve the accuracy and practicality of day-ahead prediction of load demand response behavior under small sample conditions.

[0005] The present invention adopts the following technical solution.

[0006] The present invention proposes a method for predicting the demand response behavior of a small sample under multiple influencing factors, taking the whole day of the prediction day as the prediction period and taking 15 minutes or 1 hour as a prediction interval, including:

[0007] Obtain the electricity price, various influencing factors and user baseline load of the target area during the forecast period, and obtain the historical demand response data of the target area;

[0008] Based on data-driven neural networks, a demand response behavior prediction model is established; the trained demand response behavior prediction model takes the electricity price, multiple influencing factors, historical demand response data and user baseline load within the prediction period as input, and outputs the predicted value of demand response behavior within the prediction period;

[0009] Based on the Gaussian kernel density estimation method, a day-ahead prediction model for demand response behavior is established; the trained day-ahead prediction model for demand response behavior takes the predicted value of demand response behavior within the prediction period as input, and outputs the probability distribution of demand response behavior within the prediction period; the day-ahead prediction results of demand response behavior corresponding to each prediction interval are determined from the probability distribution of demand response behavior within the prediction period.

[0010] Preferably, the electricity price of the target area during the forecast period includes: demand response incentive electricity price and time-of-use electricity price;

[0011] The various factors affecting the target area during the forecast period include: meteorological data;

[0012] According to the historical data of user baseline load, based on the probability model, the user baseline load probability distribution is determined, and the user baseline load prediction values ​​at different confidence levels in the prediction period are obtained from the user baseline load probability distribution as the user baseline load in the prediction period.

[0013] Preferably, multiple different user baseline load historical data within 24 hours a day are used as cluster centers, and multiple influencing factors and demand response historical data within the prediction period are clustered to obtain multiple sample sets, which constitute the training data set of the demand response behavior prediction model.

[0014] Preferably, each sample set includes: the correlation between the user's baseline load history data and various influencing factors within the prediction period, the correlation between the user's baseline load history data and the demand response history data, the correlation between various influencing factors within the prediction period and the demand response electricity price within the prediction period, the correlation between various influencing factors within the prediction period and the time-of-use electricity price within the prediction period, the correlation between the demand response history data and the demand response electricity price within the prediction period, and the correlation between the demand response history data and the time-of-use electricity price within the prediction period.

[0015] Preferably, the user baseline load history data of the target area and the demand response incentive electricity price are used to perform the first training on the demand response behavior day-ahead prediction model, and the demand response history data of the target area is used as a label. When the absolute value of the standard error between the output value of the demand response behavior day-ahead prediction model and the label is not less than the confidence level, the first training is terminated;

[0016] The confidence level refers to different confidence levels within the prediction period corresponding to the user baseline load prediction value obtained from the user baseline load probability distribution.

[0017] Preferably, the demand response behavior day-ahead prediction model trained for the first time is trained for the second time according to the user baseline load forecast values ​​at different confidence levels in the target area within the forecast period, the demand response incentive electricity price within the forecast period and various influencing factors within the forecast period. When the absolute value of the standard error between the output value of the demand response behavior day-ahead prediction model and the demand response behavior prediction value within the forecast period is not less than the confidence level, the second training is terminated to obtain the trained demand response behavior day-ahead prediction model.

[0018] The present invention also proposes a day-ahead prediction system for small sample demand response behavior under multiple influencing factors, taking the whole day of the prediction day as the prediction period and taking 15 minutes or 1 hour as a prediction interval, including:

[0019] The acquisition module is used to obtain the electricity price, various influencing factors and user baseline load of the target area during the forecast period, and to obtain the historical data of demand response in the target area;

[0020] The demand response behavior prediction module is used to establish a demand response behavior prediction model based on a data-driven neural network; the trained demand response behavior prediction model takes the electricity price, multiple influencing factors, demand response historical data and user baseline load within the prediction period as input, and outputs the demand response behavior prediction value within the prediction period;

[0021] The demand response behavior day-ahead prediction module is used to establish a demand response behavior day-ahead prediction model based on the Gaussian kernel density estimation method; the trained demand response behavior day-ahead prediction model takes the demand response behavior prediction value within the prediction period as input, and outputs the probability distribution of the demand response behavior within the prediction period; the demand response behavior day-ahead prediction result corresponding to each prediction interval is determined from the probability distribution of the demand response behavior within the prediction period.

[0022] Preferably, the acquisition module includes a user baseline load acquisition unit;

[0023] The user baseline load acquisition unit is used to determine the user baseline load probability distribution based on the user baseline load historical data and the probability model, and obtain the user baseline load prediction values ​​at different confidence levels within the prediction period from the user baseline load probability distribution as the user baseline load within the prediction period.

[0024] Preferably, the demand response behavior prediction module includes a training data set acquisition unit; the training data set acquisition unit is used to cluster multiple influencing factors and demand response historical data within the prediction period with multiple different user baseline load historical data within 24 hours a day as the cluster center to obtain multiple sample sets, and the multiple sample sets constitute the training data set of the demand response behavior prediction model;

[0025] Each sample set includes: the correlation between user baseline load historical data and various influencing factors within the forecast period, the correlation between user baseline load historical data and demand response historical data, the correlation between various influencing factors within the forecast period and the demand response electricity price within the forecast period, the correlation between various influencing factors within the forecast period and the time-of-use electricity price within the forecast period, the correlation between demand response historical data and the demand response electricity price within the forecast period, and the correlation between demand response historical data and time-of-use electricity price within the forecast period.

[0026] Preferably, the demand response behavior day-ahead prediction module includes a model training unit; the model training unit performs a first training on the demand response behavior day-ahead prediction model using the user baseline load history data of the target area and the demand response incentive electricity price, using the demand response history data of the target area as a label, and ends the first training when the absolute value of the standard error between the output value of the demand response behavior day-ahead prediction model and the label is not less than the confidence level;

[0027] Among them, the confidence level is the different confidence levels in the prediction period corresponding to the user baseline load prediction value obtained from the user baseline load probability distribution;

[0028] According to the user baseline load forecast values ​​at different confidence levels in the target area during the forecast period, the demand response incentive electricity price during the forecast period and various influencing factors during the forecast period, the demand response behavior day-ahead forecast model trained for the first time is trained for the second time. When the absolute value of the standard error between the output value of the demand response behavior day-ahead forecast model and the demand response behavior forecast value during the forecast period is not less than the confidence level, the second training is terminated to obtain the trained demand response behavior day-ahead forecast model.

[0029] The beneficial effects of the present invention are that, compared with the prior art, it at least includes performing multiple load correlation analyses based on historical load data and demand response data and clustering to form sample data sets to constitute a training data set for a demand response behavior prediction model, and introducing multiple correlations to enrich the training data set under small sample data; combining deep learning technology and Gaussian kernel density estimation to predict small sample short-term load demand response behavior, which helps to improve the operation and management of the power market, promote more sustainable and intelligent energy utilization, and has a positive impact on both society and the economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of a method for predicting a small sample demand response behavior under multiple influencing factors proposed by the present invention;

[0031] Figure 2 is a diagram of the demand response behavior day-ahead prediction result at a confidence level of 0.95 in an embodiment of the present invention;

[0032] Figure 3 It is a comparison chart of the demand response behavior prediction value and the demand response behavior actual value obtained in the embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0034] The present invention proposes a method for predicting the demand response behavior of a small sample under multiple influencing factors. Figure 1 As shown, including:

[0035] Step 1: Obtain the electricity price, various influencing factors and user baseline load of the target area during the forecast period, and obtain the historical demand response data of the target area.

[0036] In the embodiment, the whole 24 hours of the day to be predicted is used as the prediction period, and 15 minutes or 1 hour is used as a prediction interval.

[0037] The electricity prices in the target area during the forecast period include, but are not limited to: demand response incentive electricity prices and time-of-use electricity prices;

[0038] The various factors affecting the target area during the forecast period include, but are not limited to: meteorological data;

[0039] According to the historical data of user baseline load, based on the probability model, the user baseline load probability distribution is determined, and the user baseline load prediction values ​​at different confidence levels in the prediction period are obtained from the user baseline load probability distribution as the user baseline load in the prediction period.

[0040] The user baseline load forecast values ​​at different confidence levels determined based on the user baseline load probability distribution actually take into account the uncertainty impact of demand response in the entire forecasting process through probability means and confidence levels.

[0041] Step 2: Establish a demand response behavior prediction model based on a data-driven neural network; the trained demand response behavior prediction model takes the electricity price within the prediction period, multiple influencing factors, historical demand response data, and user baseline load as input, and outputs the demand response behavior prediction value within the prediction period.

[0042] Since demand response is a reaction to demand response incentive electricity price and time-of-use electricity price under multiple influencing factors, when demand response incentive electricity price and time-of-use electricity price are fixed, there is a correlation between demand response historical data and historical data of multiple influencing factors, and this correlation cannot be completely used as the correlation between demand response prediction data and real-time data of multiple influencing factors; and since the degree and direction of influence of real-time data of multiple influencing factors on demand response prediction data are different, the correlation between demand response prediction data and real-time data of multiple influencing factors is more complex and difficult to fit; therefore, a demand response behavior prediction model is established using a data-driven neural network, and according to the electricity price, multiple influencing factors, demand response historical data and user baseline load forecast value within the forecast period, the demand response behavior prediction model is used to obtain the demand response behavior forecast value and demand response mean under multiple influencing factors within the forecast period, which is used to characterize the relationship between the real-time demand response data and the electricity price within the forecast period.

[0043] Within 24 hours of the day, the user baseline load history data includes multiple different values. Therefore, taking multiple different user baseline load history data within 24 hours of the day as the clustering center, the various influencing factors and demand response history data within the prediction period are clustered to obtain multiple sample sets, and the multiple sample sets constitute the training data set of the demand response behavior prediction model; in each sample set, the correlation between the user baseline load history data and the various influencing factors within the prediction period, and the correlation between the user baseline load history data and the demand response history data are established respectively, and each sample set also establishes the correlation between the various influencing factors within the prediction period and the demand response incentive electricity price within the prediction period, the correlation between the various influencing factors within the prediction period and the time-of-use electricity price within the prediction period, the correlation between the demand response history data and the demand response incentive electricity price within the prediction period, and the correlation between the demand response history data and the time-of-use electricity price within the prediction period through the user baseline load history data; since the demand response behavior prediction model is a data-driven neural network, the clustering processing obtains When the sample set is used as training data, not only the electricity price and multiple influencing factors in the prediction period are provided to the data driver, and the historical data of demand response and user baseline load in the target area are obtained as basic data, but also various correlations are provided to the data driver as basic data. The data driver integrates these basic data to extract sufficient information to adjust the structure and parameters of the demand response behavior prediction model, thereby realizing the effective training of the demand response behavior prediction model. The trained demand response behavior prediction model outputs the demand response behavior prediction value and the demand response mean value in the prediction period; since the training set used by the demand response behavior prediction model contains the correlation between the demand response historical data and the demand response incentive electricity price in the prediction period and the correlation with the time-of-use electricity price in the prediction period, after multiple machine learning, the output demand response behavior prediction value and demand response mean value can also characterize the correlation between the demand response real-time data and the demand response incentive electricity price in the prediction period and the correlation with the time-of-use electricity price in the prediction period.

[0044] In addition, since the user baseline load forecast values ​​correspond to different confidence levels, the confidence level that characterizes the uncertainty of demand response further enriches the sample set and constitutes the training data set for the demand response behavior prediction model.

[0045] Furthermore, the demand response behavior prediction model is trained based on the loss function method.

[0046] Furthermore, the demand response behavior prediction model adopts data-driven convolutional neural network and long short-term memory network.

[0047] Furthermore, the data in each sample set is normalized, and the normalized sample set constitutes a training data set for the demand response behavior prediction model.

[0048] Step 3: Based on the Gaussian kernel density estimation method, a demand response behavior day-ahead prediction model is established; the trained demand response behavior day-ahead prediction model takes the predicted value of demand response behavior within the prediction period as input, and outputs the probability distribution of demand response behavior within the prediction period; and the demand response behavior day-ahead prediction result corresponding to each prediction interval is determined from the probability distribution of demand response behavior within the prediction period.

[0049] Specifically, the demand response behavior day-ahead prediction model is trained for the first time using the user baseline load historical data and the demand response incentive electricity price in the target area, with the demand response historical data in the target area as the label. When the absolute value of the standard error between the output value of the demand response behavior day-ahead prediction model and the label is not less than the confidence level, the first training is terminated, where the confidence level is the different confidence levels within the prediction period corresponding to the user baseline load prediction value obtained from the user baseline load probability distribution.

[0050] The purpose of training the demand response behavior day-ahead prediction model is to optimize the structure and parameters of the probability distribution of the demand response behavior day-ahead prediction model, so that the output value of the demand response behavior day-ahead prediction model can correctly respond to the demand response incentive electricity price based on the user's baseline load history data. Moreover, since different confidence levels are introduced to characterize the uncertainty of demand response, the demand response behavior trends that meet different confidence levels output by the trained model correspond to demand response scenarios under various uncertainties. Moreover, in the power grid system for distributed power generation access, the use of a comprehensive confidence level to characterize the uncertainty of demand response and the uncertainty of new energy output can more accurately obtain the demand response behavior trend under the new power system, more comprehensively identify the potential changes and uncertainties in power demand, and provide better services for system regulation and decision-making.

[0051] Specifically, according to the user baseline load forecast value at different confidence levels in the target area during the forecast period, the demand response incentive electricity price during the forecast period, and various influencing factors during the forecast period, the demand response behavior day-ahead forecast model trained for the first time is trained for the second time. When the absolute value of the standard error between the output value of the demand response behavior day-ahead forecast model and the demand response behavior forecast value during the forecast period is not less than the confidence level, the second training is terminated to obtain the trained demand response behavior day-ahead forecast model. The second training is the fine-tuning training of the demand response behavior day-ahead forecast model. The error is reduced through such fine-tuning. At the same time, since the meteorological data during the forecast period includes not only temperature and humidity, but also seasons and sudden climate changes, adjusting the specific content of the meteorological data or introducing extreme climate disaster data during the model fine-tuning process can also realize model training in special scenarios, improve the generalization ability of the model, and thus expand the applicable fields and scenarios of the model.

[0052] In the embodiment, at a confidence level of 0.95, the demand response behavior day-ahead prediction results corresponding to each prediction interval determined from the probability distribution of the demand response behavior within the prediction period are as follows: Figure 2 As shown; Figure 2 In the figure, the vertical axis is the load power (Power load), the horizontal axis is the sampling point, the dotted line in the figure is the predicted value of the demand response behavior (predictive value) obtained by the method proposed in the present invention, the solid circle is the real value of the demand response behavior (real value), and the gray area represents the interval corresponding to the predicted value of the demand response behavior when the confidence level (Quantile) is 0.95. The comparison between the predicted value of the demand response behavior obtained by the method proposed in the present invention and the real value of the demand response behavior is shown in Figure 1. Figure 3 As shown; Figure 3 The vertical axis is load (load), the horizontal axis is time (Time), the dotted line in the figure is the predicted value (Predictive Load) of the demand response behavior obtained by the method proposed in the present invention, and the solid line is the actual value (Actual Load) of the demand response behavior.

[0053] The method proposed in the present invention not only comprehensively considers the influence of temperature, season, baseline load, electricity price and other factors on demand response behavior, but also fully considers the uncertainty of demand response and the uncertainty of new power system. Therefore, the predicted value of demand response behavior can simulate the real value of demand response behavior more accurately, making the model performance more reliable and the output result of the model more reliable, which is helpful for the optimization and planning of power system. Unlike traditional methods, the present invention not only provides a single predicted value, but also generates prediction results at different confidence levels. This means that users can obtain predictions at different confidence levels to help them better manage risks and make decisions. Therefore, the method proposed in the present invention can be applied to a variety of fields, including power market, energy management, smart grid, etc., and the confidence level can be customized according to the needs of different fields, so as to realize an adaptive adjustment and optimization according to various scenarios in different application fields, so that it has a wide range of application potential. Through accurate power load forecasting, power system managers can better plan power supply, reduce unnecessary costs and improve energy efficiency. This helps to reduce the burden on the power system and provide more reliable power services to end users. Gaussian kernel density estimation prediction enables users to make targeted decisions under different market fluctuations. Whether it is high load period or low load period, users can better plan and optimize energy use, thereby reducing the pressure on the power system. In summary, the beneficial effects of the present invention include high accuracy, multi-confidence prediction, data quality assurance and wide applicability, which can improve the management and efficiency of the power system and provide more reliable power supply for users and power systems.

[0054] The present invention also proposes a day-ahead prediction system for small sample demand response behavior under multiple influencing factors, taking the whole day of the prediction day as the prediction period and taking 15 minutes or 1 hour as a prediction interval, including:

[0055] The acquisition module is used to obtain the electricity price, various influencing factors and user baseline load of the target area during the forecast period, and to obtain the historical data of demand response in the target area;

[0056] The demand response behavior prediction module is used to establish a demand response behavior prediction model based on a data-driven neural network; the trained demand response behavior prediction model takes the electricity price, multiple influencing factors, demand response historical data and user baseline load within the prediction period as input, and outputs the demand response behavior prediction value within the prediction period;

[0057] The demand response behavior day-ahead prediction module is used to establish a demand response behavior day-ahead prediction model based on the Gaussian kernel density estimation method; the trained demand response behavior day-ahead prediction model takes the demand response behavior prediction value within the prediction period as input, and outputs the probability distribution of the demand response behavior within the prediction period; the demand response behavior day-ahead prediction result corresponding to each prediction interval is determined from the probability distribution of the demand response behavior within the prediction period.

[0058] Preferably, the acquisition module includes a user baseline load acquisition unit;

[0059] The user baseline load acquisition unit is used to determine the user baseline load probability distribution based on the user baseline load historical data and the probability model, and obtain the user baseline load prediction values ​​at different confidence levels within the prediction period from the user baseline load probability distribution as the user baseline load within the prediction period.

[0060] Preferably, the demand response behavior prediction module includes a training data set acquisition unit; the training data set acquisition unit is used to cluster multiple influencing factors and demand response historical data within the prediction period with multiple different user baseline load historical data within 24 hours a day as the cluster center to obtain multiple sample sets, and the multiple sample sets constitute the training data set of the demand response behavior prediction model;

[0061] Each sample set includes: the correlation between user baseline load historical data and various influencing factors within the forecast period, the correlation between user baseline load historical data and demand response historical data, the correlation between various influencing factors within the forecast period and the demand response electricity price within the forecast period, the correlation between various influencing factors within the forecast period and the time-of-use electricity price within the forecast period, the correlation between demand response historical data and the demand response electricity price within the forecast period, and the correlation between demand response historical data and time-of-use electricity price within the forecast period.

[0062] Preferably, the demand response behavior day-ahead prediction module includes a model training unit; the model training unit performs a first training on the demand response behavior day-ahead prediction model using the user baseline load history data of the target area and the demand response incentive electricity price, using the demand response history data of the target area as a label, and ends the first training when the absolute value of the standard error between the output value of the demand response behavior day-ahead prediction model and the label is not less than the confidence level;

[0063] Among them, the confidence level is the different confidence levels in the prediction period corresponding to the user baseline load prediction value obtained from the user baseline load probability distribution;

[0064] According to the user baseline load forecast values ​​at different confidence levels in the target area during the forecast period, the demand response incentive electricity price during the forecast period and various influencing factors during the forecast period, the demand response behavior day-ahead forecast model trained for the first time is trained for the second time. When the absolute value of the standard error between the output value of the demand response behavior day-ahead forecast model and the demand response behavior forecast value during the forecast period is not less than the confidence level, the second training is terminated to obtain the trained demand response behavior day-ahead forecast model.

[0065] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0066] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0067] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0068] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting small sample demand response behavior under multiple influencing factors, taking the whole day of the prediction day as the prediction period, and taking 15 minutes or 1 hour as a prediction interval, characterized in that: include: Obtain the electricity price, various influencing factors and user baseline load of the target area during the forecast period, and obtain the historical demand response data of the target area; Based on data-driven neural networks, a demand response behavior prediction model is established; the trained demand response behavior prediction model takes the electricity price, multiple influencing factors, historical demand response data and user baseline load within the prediction period as input, and outputs the predicted value of demand response behavior within the prediction period; Based on the Gaussian kernel density estimation method, a day-ahead prediction model for demand response behavior is established; the trained day-ahead prediction model for demand response behavior takes the predicted value of demand response behavior within the prediction period as input, and outputs the probability distribution of demand response behavior within the prediction period; the day-ahead prediction results of demand response behavior corresponding to each prediction interval are determined from the probability distribution of demand response behavior within the prediction period.

2. The method for predicting the small sample demand response behavior under multiple influencing factors according to claim 1 is characterized in that: The electricity prices in the target area during the forecast period include: demand response incentive electricity prices and time-of-use electricity prices; The various factors affecting the target area during the forecast period include: meteorological data; According to the historical data of user baseline load, based on the probability model, the user baseline load probability distribution is determined, and the user baseline load prediction values ​​at different confidence levels in the prediction period are obtained from the user baseline load probability distribution as the user baseline load in the prediction period.

3. The method for predicting small sample demand response behavior under multiple influencing factors according to claim 2 is characterized in that: Taking multiple different user baseline load historical data within 24 hours a day as the clustering center, various influencing factors and demand response historical data within the prediction period are clustered to obtain multiple sample sets, which constitute the training data set of the demand response behavior prediction model.

4. The method for predicting the small sample demand response behavior under multiple influencing factors according to claim 3 is characterized in that: Each sample set includes: the correlation between user baseline load historical data and various influencing factors within the forecast period, the correlation between user baseline load historical data and demand response historical data, the correlation between various influencing factors within the forecast period and the demand response electricity price within the forecast period, the correlation between various influencing factors within the forecast period and the time-of-use electricity price within the forecast period, the correlation between demand response historical data and the demand response electricity price within the forecast period, and the correlation between demand response historical data and time-of-use electricity price within the forecast period.

5. The method for predicting the small sample demand response behavior under multiple influencing factors according to claim 2 is characterized in that: The demand response behavior day-ahead prediction model is trained for the first time using the historical user baseline load data and the demand response incentive electricity price in the target area, with the historical demand response data in the target area as the label. When the absolute value of the standard error between the output value of the demand response behavior day-ahead prediction model and the label is not less than the confidence level, the first training is terminated. The confidence level refers to different confidence levels within the prediction period corresponding to the user baseline load prediction value obtained from the user baseline load probability distribution.

6. The method for predicting small sample demand response behavior under multiple influencing factors according to claim 5 is characterized in that: According to the user baseline load forecast values ​​at different confidence levels in the target area during the forecast period, the demand response incentive electricity price during the forecast period and various influencing factors during the forecast period, the demand response behavior day-ahead forecast model trained for the first time is trained for the second time. When the absolute value of the standard error between the output value of the demand response behavior day-ahead forecast model and the demand response behavior forecast value during the forecast period is not less than the confidence level, the second training is terminated to obtain the trained demand response behavior day-ahead forecast model.

7. A day-ahead prediction system for small sample demand response behavior under multiple influencing factors, with the whole day of the prediction day as the prediction period and 15 minutes or 1 hour as a prediction interval, characterized in that: include: The acquisition module is used to obtain the electricity price, various influencing factors and user baseline load of the target area during the forecast period, and obtain the historical data of demand response in the target area; The demand response behavior prediction module is used to establish a demand response behavior prediction model based on a data-driven neural network; the trained demand response behavior prediction model takes the electricity price, multiple influencing factors, demand response historical data and user baseline load within the prediction period as input, and outputs the demand response behavior prediction value within the prediction period; The demand response behavior day-ahead prediction module is used to establish a demand response behavior day-ahead prediction model based on the Gaussian kernel density estimation method; the trained demand response behavior day-ahead prediction model takes the demand response behavior prediction value within the prediction period as input, and outputs the probability distribution of the demand response behavior within the prediction period; the demand response behavior day-ahead prediction result corresponding to each prediction interval is determined from the probability distribution of the demand response behavior within the prediction period.

8. The day-ahead prediction system for small sample demand response behavior under multiple influencing factors according to claim 7 is characterized in that: The acquisition module includes a user baseline load acquisition unit; The user baseline load acquisition unit is used to determine the user baseline load probability distribution based on the user baseline load historical data and the probability model, and obtain the user baseline load prediction values ​​at different confidence levels within the prediction period from the user baseline load probability distribution as the user baseline load within the prediction period.

9. The system for predicting small sample demand response behavior based on multiple influencing factors according to claim 7, characterized in that: The demand response behavior prediction module includes a training data set acquisition unit; the training data set acquisition unit is used to cluster multiple influencing factors and demand response historical data within the prediction period with multiple different user baseline load historical data within 24 hours a day as the cluster center to obtain multiple sample sets, and the multiple sample sets constitute the training data set of the demand response behavior prediction model; Each sample set includes: the correlation between user baseline load historical data and various influencing factors within the forecast period, the correlation between user baseline load historical data and demand response historical data, the correlation between various influencing factors within the forecast period and the demand response electricity price within the forecast period, the correlation between various influencing factors within the forecast period and the time-of-use electricity price within the forecast period, the correlation between demand response historical data and the demand response electricity price within the forecast period, and the correlation between demand response historical data and time-of-use electricity price within the forecast period.

10. The method for predicting the small sample demand response behavior under multiple influencing factors according to claim 7, characterized in that: The demand response behavior day-ahead prediction module includes a model training unit; The model training unit uses the historical user baseline load data of the target area and the demand response incentive electricity price to perform the first training on the demand response behavior day-ahead prediction model, and uses the historical demand response data of the target area as a label. When the absolute value of the standard error between the output value of the demand response behavior day-ahead prediction model and the label is not less than the confidence level, the first training is terminated; Among them, the confidence level is the different confidence levels in the prediction period corresponding to the user baseline load prediction value obtained from the user baseline load probability distribution; According to the user baseline load forecast values ​​at different confidence levels in the target area during the forecast period, the demand response incentive electricity price during the forecast period and various influencing factors during the forecast period, the demand response behavior day-ahead forecast model trained for the first time is trained for the second time. When the absolute value of the standard error between the output value of the demand response behavior day-ahead forecast model and the demand response behavior forecast value during the forecast period is not less than the confidence level, the second training is terminated to obtain the trained demand response behavior day-ahead forecast model.

Images