Load prediction method and system considering user active adjustment influence

Abstract

The invention relates to a load prediction method and system considering user active adjustment influence, and the method comprises the steps: firstly collecting and summarizing historical energy consumption data, meteorological data, holiday and festival information, a state of whether to participate in load adjustment, and a controlled quantity factor according to a selected research object, and forming an initial sequence data set; performing data cleaning on the data set to form a time sequence data pool; secondly, considering a state of whether to participate in load adjustment and a controlled quantity factor, estimating an actual load value of a user participating in a load adjustment time period (such as not participating in adjustment) by using a similar day method, and updating a data sample set to be free of manual adjustment influence; then, an optimized BP neural network algorithm is adopted, a feedback channel is established between a prediction result and input data, and the model is trained; according to the method, influence factors that the natural law is changed by manual regulation and control of the load are considered during prediction, the prediction accuracy is effectively improved, and the method is suitable for load prediction of users participating in auxiliary services of the virtual power plant.

Description

Technical Field

[0001] The present invention relates to the field of electric power information technology, and in particular to a load forecasting method and system taking into account the influence of user active regulation. Background Art

[0002] With the rapid development of information technology and the widespread deployment of smart grids, user load forecasting, as a key link in power system planning and operation, has become increasingly important. User load forecasting aims to accurately estimate power demand in the future through multi-dimensional data sources such as historical data, meteorological information, and socio-economic factors. Research in this field is not only related to the safe and stable operation of the power system, but also directly affects the supply and demand balance of the power market, the efficient allocation of energy, and the effective integration of renewable energy.

[0003] Traditional load forecasting methods include statistical models such as time series analysis and regression analysis. In recent years, with the rapid development of machine learning and deep learning technologies, forecasting models based on neural networks, support vector machines, random forests and other algorithms have gradually become mainstream. In addition, hybrid models (i.e., combining multiple models) also provide new ideas for further improving forecasting accuracy.

[0004] These existing model methods have the following technical problems when applied to load forecasting of virtual power plant users: the load of virtual power plant users has controlled components due to factors such as demand-side response, which changes the natural change process of the load. Therefore, there are different types of load history values, such as unadjusted and adjusted. Common algorithms are based on the natural change process of the load. After filtering periodic factors through algorithms such as difference, the load sequence is close to white noise. For load sequences with controlled components, common algorithms cannot be well applied, so the final prediction accuracy is limited. Summary of the invention

[0005] In order to solve the problem that the existing technology fails to consider the influencing factors of the controlled component when applied to virtual power plant user load forecasting, the present invention introduces the state and controlled quantity of whether to participate in load adjustment, uses the similar day method to estimate the actual load value when participating in the adjustment, and adopts an optimized BP neural network algorithm. By establishing a feedback channel between the prediction result and the input data, a load forecasting method and system that considers the influence of user active adjustment is provided.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme.

[0007] A load forecasting method considering the influence of user active regulation, comprising:

[0008] S1 builds a time series data pool based on the historical power load data of the target energy-consuming object, the external environment data at the corresponding time, and the load adjustment information;

[0009] S2 selects historical time series data of a certain period in the time series data pool, screens and obtains dates in the same time period that are not involved in load adjustment, and calculates the actual load value on the date that is involved in load adjustment in the historical time series data of the certain period by similar day method; inputs the actual load value into the time series data pool to update the historical time series data sequence;

[0010] S3 calculates the similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool after executing step S2 by the similar day method, and arranges the calculated similarities in descending order;

[0011] S4 selects the data of the first 10 evaluation object series with the largest similarity based on the similarity sequence after executing step S3, and constructs the BP neural network model training samples;

[0012] S5 builds a BP neural network model, uses the network model training samples as input of the BP neural network model, and trains the BP neural network model;

[0013] S6 executes steps S1 to S5 multiple times to complete the training of the BP neural network model;

[0014] S7 obtains the predicted value of power load at a certain moment in the future through actual historical data and the trained BP neural network model.

[0015] Preferably, step S2 specifically includes:

[0016] In the time series data pool, take the historical time series data (X t-n+ 1,...,X t );

[0017] For each time interval (X j ,...,X k ) Take the historical data of the previous 30 days and filter the dates that are not involved in load adjustment in the same time period;

[0018] The similar day method is used to estimate the actual load value (Y j’ ,...,Y k’ ) and update the historical time series data sequence (X t-n+1 ,...,X j’ ,...,X k’ ,...,X t )~(Y t-n+1 ,...,Y j’ ,...,Y k’ ,...,Yt ).

[0019] Preferably, step S4 specifically includes:

[0020] Select the best similar p-day data set as the BP neural network model training sample;

[0021] Step S5 specifically includes:

[0022] Take the best similarity p days of historical data at time q every day (X 11 , ...X 1q, X 21 ,...,X 2q , ..., X p1 ,...,X pq ) as the model input, and the load data of q consecutive moments in the future (Y t+1 ,...,Y t+q) As the model output, a BP neural network model is constructed, and the model is trained using the data in the time series data pool.

[0023] Preferably, the BP neural network model includes one hidden layer, which includes 20 neurons.

[0024] Preferably, the BP neural network model further comprises using a genetic algorithm to optimize the weights and thresholds of the BP neural network model, calculating the optimal weights and thresholds and performing model training;

[0025] The weights and thresholds of the BP neural network model are expressed as

[0026] S=R*S1+S1*S2+S1+S2

[0027] Calculated; where S is the encoding length, S1 is the number of hidden layer neurons, S2 is the number of output layer neurons, and R is the number of input neurons;

[0028] The process of using genetic algorithm to optimize the weights and thresholds of the BP neural network model, calculating the optimal weights and thresholds and training the model also includes:

[0029] The weights and thresholds of the BP neural network are encoded, the population size is set to 50, and the initial population is generated; the inverse of the sum of squares of the actual output and expected output of the BP neural network is used as the fitness function of the genetic algorithm, and the fitness value of each individual is determined by the fitness function;

[0030] The selection, crossover and mutation cycles are used until the set number of genetic generations is exceeded, and the individual with the largest fitness value is selected as the optimal individual of the group.

[0031] Use the optimal individual of the current group to decompose it into the connection weight and connection threshold of the BP neural network;

[0032] The BP network is trained using new weights and thresholds, and simulated and tested using untrained data.

[0033] Preferably, the input of the BP neural network model includes: 10 days of historical days optimized by the similar day method, the actual power load for the period not participating in the adjustment or the estimated power load for the period participating in the adjustment, the temperature, humidity, air pressure, wind direction at the predicted time, whether it is a holiday, and whether to participate in load regulation.

[0034] Preferably, the similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool includes consideration of the similarity of meteorological factors, date factors, and time distance factors.

[0035] In a second aspect, the present invention provides a load forecasting system that takes into account the impact of user active adjustment, including:

[0036] Data acquisition module, used to:

[0037] Based on the historical power load data of the target energy-consuming object, the external environment data at the corresponding time, and the load adjustment information, a time series data pool is constructed;

[0038] Analytical modeling modules for:

[0039] Based on selecting the historical time series data of a certain period in the time series data pool, filter out the dates in the same time period that are not involved in load adjustment, and calculate the actual load value on the date in the historical time series data of the certain period that is involved in load adjustment by similar day method; input the actual load value into the time series data pool to update the historical time series data sequence;

[0040] The similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool is calculated by the similar day method, and the calculated similarities are arranged in descending order;

[0041] Based on the similarity sequence, data of the day to be predicted with a similarity value greater than a first preset threshold are selected to construct a BP neural network model training sample;

[0042] Construct a BP neural network model, use the network model training samples as the input of the BP neural network model, and train the BP neural network model;

[0043] The above process is performed multiple times to complete the training of the BP neural network model;

[0044] The load prediction module is used to: obtain the predicted value of the power load at a certain moment in the future through actual historical data and the trained BP neural network model; and output the predicted value of the power load at a certain moment in the future.

[0045] In a third aspect, the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0047] It can be seen from the technical solutions provided by the above embodiments of the present invention that the present invention provides a load forecasting method and system considering the influence of user active adjustment. The method first collects and summarizes historical energy consumption data, meteorological data and holiday information, the state of participating in load adjustment and the controlled quantity factors according to the selected research object to form an initial sequence data set; the data set is cleaned to form a time series data pool; then the state of participating in load adjustment and the controlled quantity factors are considered, and the actual load value of the user participating in the load adjustment period (such as not participating in the adjustment) is estimated by using the similar day method, and the data sample set is updated to not contain the influence of human adjustment; then the optimized BP neural network algorithm is adopted to train the model by establishing a feedback channel between the prediction result and the input data. This method predicts the load value in the time period of the user's active adjustment in the collected data sample, and restores the actual load value that should occur if it does not participate in the adjustment, so as to provide a reliable data basis for a series of subsequent short-term load forecasting results. Taking the power load forecasting of an industrial user in a provincial capital city in 2023 as an example, after adopting the present invention, the accuracy of user load forecasting is increased from 79.4% to 83.89%. The specific situation is as follows: the user's historical data is collected for a total of 90 days, of which 23 days participate in valley filling demand response in different time periods. According to the present invention, the load occurrence value of the valley filling period is corrected to varying degrees. From the experimental verification, the present invention can effectively improve the accuracy of load prediction of energy-consuming objects in application.

[0048] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0050] Figure 1 A processing flow chart of a load forecasting method considering the influence of user active adjustment provided by the present invention;

[0051] Figure 2 An architectural diagram of a load forecasting method considering the influence of user active adjustment provided by the present invention;

[0052] Figure 3 A comparison diagram of the actual user load value (participating in valley filling) and the predicted load value that should occur without participating in the adjustment in a load forecasting method considering the influence of user active adjustment provided by the present invention;

[0053] Figure 4 A comparison chart of the prediction effects of a load prediction method considering the influence of user active adjustment provided by the present invention;

[0054] Figure 5 A logic block diagram of a load forecasting system considering the influence of user active adjustment provided by the present invention.

[0055] In the figure:

[0056] 501. Data acquisition module 502. Analysis and modeling module 503. Load forecasting module. DETAILED DESCRIPTION

[0057] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0058] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0059] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0060] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0061] See also Figure 1 and 2 The present invention provides a load forecasting method considering the influence of user active adjustment, comprising:

[0062] S1 builds a time series data pool based on the historical power load data of the target energy-consuming object, the external environment data at the corresponding time, and the load adjustment information;

[0063] S2 selects historical time series data of a certain period in the time series data pool, screens and obtains dates in the same time period that are not involved in load adjustment, and calculates the actual load value on the date that is involved in load adjustment in the historical time series data of the certain period by similar day method; inputs the actual load value into the time series data pool to update the historical time series data sequence;

[0064] S3 calculates the similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool after executing step S2 by the similar day method, and arranges the calculated similarities in descending order;

[0065] S4 selects the data of the first 10 evaluation object series with the largest similarity based on the similarity sequence after executing step S3, and constructs the BP neural network model training samples;

[0066] S5 builds a BP neural network model, uses the network model training samples as input of the BP neural network model, and trains the BP neural network model;

[0067] S6 executes steps S1 to S5 multiple times to complete the training of the BP neural network model;

[0068] S7 obtains the predicted value of power load at a certain moment in the future through actual historical data and the trained BP neural network model.

[0069] In the preferred embodiment provided by the present invention, the specific execution process of the above steps is as follows.

[0070] 1. Build a time series dataset

[0071] According to the selected energy-consuming objects, their power consumption data are collected and summarized, and the external environmental data and load adjustment information at the corresponding time are recorded; the summarized historical data are pre-processed, outliers are removed, and then normalized, and the obtained data is used as a time series data pool for backup.

[0072] The existence of significant abnormal data points has a serious impact on the statistical characteristics of the load curve and the extraction of the characteristic curve. In the data preprocessing stage, it is necessary to first remove obvious abnormal data points. Including: ① Zero value and null value recognition, for null values, no record or record is NULL as the judgment, for zero values, first judge the rationality of the zero value based on the previous and subsequent load values. If there are near-zero data before and after, and zero values ​​have appeared in history, or the user mark can appear zero value, then it is a reasonable zero value, otherwise it is an abnormal zero value. ② Continuous constant value, the judgment of continuous constant value is judged by the load difference between adjacent points. If the load difference between two adjacent points is less than the detection window width, it is judged to be continuous and constant. ③ Abnormal step value, for the detection of significant abnormal step, the user load at two adjacent moments can be subtracted to obtain the user load increment. Assuming that the load increment obeys the overall distribution, the 3σ principle is used as the judgment criterion.

[0073] Abnormal data repair fully utilizes the load characteristic curve information. After repair, the repaired data maintains continuity with the adjacent normal load data; the repaired data should be similar in shape to the load characteristic curve. For each abnormal data segment, find its corresponding characteristic curve segment, perform linear transformation on the characteristic curve segment, and keep the endpoints of the segment matching its adjacent normal data.

[0074] 2. Actively adjust behavior recognition

[0075] Some of the user's active adjustment behaviors are recorded, but some may not be recorded. It is necessary to identify the user's active adjustment behaviors. After the load is adjusted, the curve shape will have abnormal conditions in multiple periods, and its characteristic quantity will change. The characteristic embodiment method is used for identification. The load sequence of the multi-day natural change process is decomposed in the frequency domain to obtain the load sequence spectrum. The spectrum is combined according to the cycle length to obtain the time domain components of different cycle lengths. The components of different cycles are accumulated to obtain the basic characteristic curve. Compare the basic characteristic curve with the characteristic curve to be identified. If there is an obvious trend deviation, it is judged as the user's active adjustment curve.

[0076] Another way to identify user active adjustment behavior is to use pattern matching. The basic operation is: if the load curve to be identified is highly similar to the load curve that has been marked with user active adjustment behavior in the past, and is highly different from the load characteristic curve that has not been adjusted by the main body in the past, then the dual criteria can be used to determine whether user active adjustment behavior has occurred on the day to be identified.

[0077] For example, if historical data shows that a user has made active adjustments but has not been marked, the user can be actively marked and used in subsequent active adjustment behavior identification.

[0078] 3. Separation of controlled components of historical load

[0079] In load forecasting, reference samples are an important basis for forecasting, and the forecast results are obtained by calculating the reference samples. In the implementation system of the forecasting algorithm, the reference samples are the input link. If there is a problem with the input data, it will be reflected in the forecast results after the forecast calculation. When forecasting the load of virtual power plant users, the collected historical load data has two states: "participating in adjustment" and "not participating in adjustment". The load value in the "participating in adjustment" state changes its natural load law. If the collected load value in the participating adjustment state is directly input into the forecasting model, it will cause deviations in the forecast results.

[0080] From the time series data pool, take historical time series data (X t-n+1 ,...,X t ), for each time interval (X j ,...,X k ) Take the historical data of the previous 30 days, and select the dates in the same time period that are not involved in load adjustment. According to the actual load values ​​at time Xj-1 and Xk+1, use the similar day method to select the best similar day. Use the correlation between the external environment data and the actual load value of the best similar day to predict the selected "participation adjustment" time interval (X j ,...,X k ), if the actual load value (Yj’ ,...,Y k’ ), update the predicted actual load value to the historical time series number (X t-n+1 ,...,X j’ ,...,X k’ ,...,X t )~(Y t-n+1 ,...,Y j’ ,...,Y k’ ,...,Y t ).

[0081] 4. Uncontrolled component load series prediction

[0082] From the updated time series data set (Xt-n+1,...,Xj',...,Xk',...,Xt)~(Y t-n+1 ,...,Y j’ ,...,Y k’ ,...,Y t ) Using the similar day method, taking into account meteorological factors, date factors, and time distance factors, after calculating the similarity between the predicted day and the historical day, the top 10 groups of evaluation object series with the largest similarity are selected as training samples of the BP neural network. The historical data (X 11 , ...X 1q ,X 21 ,...,X 2q , ..., X p1 ,...,X pq ) is used as the model input, and the load data of the next q consecutive moments ( Yt+1 ,...,Y t+q) as the model output, and construct the BP neural network model. The number of hidden layers of the BP neural network is determined to be 1, and the number of neurons included is set to 20 according to the experimental results. The weight threshold of the BP neural network is optimized by using a genetic algorithm. The weight and threshold length are determined by the scale of the input data in the BP neural network. The length of the weight threshold is determined by the following formula: S = R*S1+S1*S2+S1+S2, where S is the encoding length, S1 is the number of hidden layer neurons, S2 is the number of output layer neurons, and R is the number of input neurons; the weights and thresholds of the BP neural network are encoded, and the population size is set to 50 to generate the initial population; the inverse of the sum of squares of the error between the actual output and the expected output of the BP neural network is used as the fitness function of the genetic algorithm, and the fitness value of each individual is determined by the fitness function. The larger the fitness value, the smaller the error; the selection, crossover, and mutation cycle operations are used until the set genetic generation is exceeded, and the individual with the largest fitness value is selected as the optimal individual of the group; the optimal individual of the current group is decomposed into the connection weights and thresholds of the BP neural network. The BP network is trained with new weights and thresholds, and untrained data is used for simulation testing. The daily feature value data of the predicted day is input into the trained BP neural network to obtain the simulated predicted value, and the power load value of the predicted day is obtained after denormalization.

[0083] 5. Correction of controlled components in predicted load

[0084] Based on the initial prediction results, whether there are factors of active adjustment in the electricity consumption behavior of users on the operating day. If so, make corrections to the base load. First, make a prediction of the correction amount, and select the similarity of the adjustment behavior. If there have been similar adjustments in historical data, the above-mentioned method of separating the controlled components can be used to calculate the controlled components and superimpose them on the initial prediction results; if there have been no similar adjustments in history, make corrections to the initial prediction results based on the period and amplitude of the active adjustment.

[0085] After the results are corrected, rationality checks are required. If the corrected load is higher than the maximum value of the user's load or lower than the minimum value of the user's load, the load in this period is limited to the maximum or minimum value. At the same time, the change rate of the user's load must be checked. If it exceeds the maximum change rate of the load in history, the load change process is limited to the load value supported by the maximum change rate. Some users have constraints on the overall power consumption. The load of such users also needs to check the overall power consumption in a certain period of time. If the power consumption exceeds or is insufficient, the correction amount is further adjusted to constrain the load on the overall power consumption.

[0086] In a second aspect, the present invention provides a load forecasting system that takes into account the impact of user active adjustment, such as Figure 5 As shown, including:

[0087] The data acquisition module 501 is used to:

[0088] Based on the historical power load data of the target energy-consuming object, the external environment data at the corresponding time, and the load adjustment information, a time series data pool is constructed;

[0089] The analysis and modeling module 502 is used to:

[0090] Based on selecting the historical time series data of a certain period in the time series data pool, filter out the dates in the same time period that are not involved in load adjustment, and calculate the actual load value on the date in the historical time series data of the certain period that is involved in load adjustment by similar day method; input the actual load value into the time series data pool to update the historical time series data sequence;

[0091] The similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool is calculated by the similar day method, and the calculated similarities are arranged in descending order;

[0092] Based on the similarity sequence, data of the day to be predicted with a similarity value greater than a first preset threshold are selected to construct a BP neural network model training sample;

[0093] Construct a BP neural network model, use the network model training samples as the input of the BP neural network model, and train the BP neural network model;

[0094] The above process is performed multiple times to complete the training of the BP neural network model;

[0095] The load prediction module 503 is used to: obtain the predicted value of the power load at a certain moment in the future through actual historical data and the trained BP neural network model; and output the predicted value of the power load at a certain moment in the future.

[0096] In a third aspect, the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0097] The present invention also provides a computer-readable storage medium on which a computer program is stored, and the program performs the steps of the above method when executed by a processor.

[0098] The present invention also provides a processor, which is used to run a program, wherein the above method is executed when the program is running.

[0099] In summary, the present invention provides a load forecasting method and system considering the influence of user active adjustment. The method first collects and summarizes historical energy consumption data, meteorological data and holiday information, whether to participate in load adjustment and controlled quantity factors according to the selected research object to form an initial sequence data set; the data set is cleaned to form a time series data pool; then the state and controlled quantity factors of whether to participate in load adjustment are considered, and the actual load value of the user participating in the load adjustment period (such as not participating in the adjustment) is estimated by using the similar day method, and the data sample set is updated to not contain the influence of human adjustment; then the optimized BP neural network algorithm is adopted to train the model by establishing a feedback channel between the prediction result and the input data. This method predicts the load value in the time period of the user's active adjustment in the collected data sample, and restores the actual load value that should occur if it does not participate in the adjustment, so as to provide a reliable data basis for a series of subsequent short-term load forecasting results. Taking the power load forecasting situation of an industrial user in a provincial capital city in 2023 as an example, after adopting the present invention, the accuracy of user load forecasting is increased from 79.4% to 83.89%. The specific situation is as follows: the user's historical data is collected for a total of 90 days, of which 23 days participate in valley filling demand response in different time periods. According to the present invention, the load occurrence value of the valley filling period is corrected to varying degrees. From the experimental verification, the present invention can effectively improve the accuracy of load prediction of energy-consuming objects in application (such as Figure 3 and 4 shown).

[0100] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0101] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0102] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0103] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A load forecasting method considering the influence of user active adjustment, characterized in that: include: S1 builds a time series data pool based on the historical power load data of the target energy-consuming object, the external environment data at the corresponding time, and the load adjustment information; S2 selects historical time series data of a certain period in the time series data pool, screens and obtains dates in the same time period that are not involved in load adjustment, and calculates the actual load value of the date in the historical time series data of the certain period that is involved in load adjustment by similar day method; inputs the actual load value into the time series data pool to update the historical time series data sequence; S3: calculating the similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool after executing step S2 by the similar day method, and arranging the calculated similarities in descending order; S4 selects the data of the first 10 evaluation object series with the largest similarity based on the similarity sequence after executing step S3, and constructs the BP neural network model training samples; S5: constructing a BP neural network model, taking the network model training sample as the input of the BP neural network model, and training the BP neural network model; S6 executes steps S1 to S5 multiple times to complete the training of the BP neural network model; S7 obtains the predicted value of the power load at a certain moment in the future through actual historical data and the trained BP neural network model.

2. The method according to claim 1, characterized in that Step S2 specifically includes: In the time series data pool, take the historical time series data (X) with a length of n t-n+ 1,...,X t ); For each time interval (X j ,...,X k ) Take the historical data of the previous 30 days and filter the dates that are not involved in load adjustment in the same time period; The similar day method is used to estimate the actual load value (Y j’ ,...,Y k’ ), and update to the historical time series data sequence (X) in the time series data pool t-n+1 ,...,X j’ ,...,X k’ ,...,X t )~(Y t-n+1 ,...,Y j’ ,...,Y k’ ,...,Y t ).

3. The method according to claim 2, characterized in that Step S4 specifically includes: Select the best similar p-day data set as the BP neural network model training sample; Step S5 specifically includes: Take the best similarity p days of historical data at time q every day (X 11 , ...X 1q, X 21 ,...,X 2q , ..., X p1 ,...,X pq ) as the model input, and the load data of q consecutive moments in the future (Y t+1 ,...,Y t+q) As the model output, a BP neural network model is constructed, and the model is trained using the data in the time series data pool.

4. The method according to any one of claims 1 to 3, characterized in that: The BP neural network model includes a hidden layer, which includes 20 neurons.

5. The method according to claim 4, characterized in that The BP neural network model also includes using a genetic algorithm to optimize the weights and thresholds of the BP neural network model, calculating the optimal weights and thresholds and performing model training; The weights and thresholds of the BP neural network model are expressed by S=R*S1+S1*S2+S1+S2 Calculated; where S is the encoding length, S1 is the number of hidden layer neurons, S2 is the number of output layer neurons, and R is the number of input neurons; The process of optimizing the weights and thresholds of the BP neural network model using a genetic algorithm, calculating the optimal weights and thresholds, and training the model also includes: The weights and thresholds of the BP neural network are encoded, the population size is set to 50, and the initial population is generated; the inverse of the sum of squares of the actual output and expected output of the BP neural network is used as the fitness function of the genetic algorithm, and the fitness value of each individual is determined by the fitness function; The selection, crossover and mutation cycles are used until the set number of genetic generations is exceeded, and the individual with the largest fitness value is selected as the optimal individual of the group. Use the optimal individual of the current group to decompose it into the connection weight and connection threshold of the BP neural network; The BP network is trained using new weights and thresholds, and simulated and tested using untrained data.

6. The method according to any one of claims 1 to 3, characterized in that: The input of the BP neural network model includes: 10 historical days optimized by the similar day method, the actual power load for the period not participating in the adjustment or the estimated power load for the period participating in the adjustment, the temperature, humidity, air pressure, wind direction at the prediction time, whether it is a holiday, and whether to participate in load regulation.

7. The method according to any one of claims 1 to 3, characterized in that: The similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool includes consideration of the similarity of meteorological factors, date factors, and time distance factors.

8. A load forecasting system considering the impact of user active adjustment, characterized in that: include: Data acquisition module, used to: Based on the historical power load data of the target energy-consuming object, the external environment data at the corresponding time, and the load adjustment information, a time series data pool is constructed; Analytical modeling modules for: Based on selecting historical time series data of a certain period in the time series data pool, screening to obtain dates in the same time period whose status is not involved in load adjustment, and calculating the actual load value on the date whose status is involved in load adjustment in the historical time series data of the certain period by similar day method; inputting the actual load value into the time series data pool to update the historical time series data sequence; The similarity between the load value of the day to be predicted and the load value of the historical day in the time series data pool is calculated by the similar day method, and the calculated similarities are arranged in descending order; Based on the similarity sequence, data of the day to be predicted with a similarity value greater than a first preset threshold are selected to construct a BP neural network model training sample; Constructing a BP neural network model, taking the network model training samples as input of the BP neural network model, and training the BP neural network model; The above process is performed multiple times to complete the training of the BP neural network model; The load prediction module is used to: obtain the predicted value of the power load at a certain moment in the future through actual historical data and the trained BP neural network model; and output the predicted value of the power load at a certain moment in the future.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images