Medium-term load prediction method, device and system based on electric quantity feedback correction

Abstract

The invention discloses a medium-term load prediction method, device and system based on electric quantity feedback correction, and the method comprises the steps: inputting selected first historical daily electricity consumption data into a daily electricity consumption prediction model, and obtaining a daily electricity consumption prediction value in a to-be-predicted period; based on the selected second historical daily electricity consumption data, the historical daily meteorological data and the historical holiday data corresponding to the selected second historical daily electricity consumption data, finding out historical similar days corresponding to the daily in the to-be-predicted period; inputting historical day meteorological data and historical holiday data corresponding to each historical similar day into a load-electricity consumption ratio prediction model to obtain a moment load-daily electricity consumption ratio corresponding to each day in the to-be-predicted period; and calculating a daily moment load predicted value in the to-be-predicted period based on the daily electricity consumption predicted value and the moment load-daily electricity consumption ratio. According to the method, daily electricity consumption medium-term prediction and load-daily electricity consumption decomposition based on similar days are carried out in sequence, so that medium-term prediction of the load is realized, and the accumulative amplification effect of errors in long-sequence prediction is inhibited.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power system load forecasting, and particularly relates to a medium-term load forecasting method, device and system based on power feedback correction. BACKGROUND

[0002] Power system load forecasting is an important part of power system generation planning and the basis of power system economic operation. In the current situation of rapid development and tight supply of electricity, it is extremely important to reasonably plan and operate the power system. Medium-term load forecasting provides data support for system operation and maintenance, power purchase plan management, and can assist power grid companies in optimizing investment strategies, improving asset utilization, and ensuring the safe and stable operation of the power system under supply and demand fluctuations. Therefore, the progress of medium-term load forecasting technology is of great significance to the fine management and market operation of the power system.

[0003] Medium-term load forecasting usually requires a long prediction step. For example, for daily 24 time point load forecasting in the next month, 720 prediction points need to be output continuously. Traditional artificial intelligence models are prone to error accumulation and amplification in such long sequence prediction, resulting in a serious deviation of the prediction results from the actual situation. SUMMARY

[0004] To solve the above problems, the present application provides a medium-term load forecasting method, device and system based on power feedback correction, which sequentially performs medium-term load forecasting, load-day electricity consumption decomposition based on similar days, and realizes medium-term load forecasting, which can inhibit the cumulative amplification effect of errors in long sequence prediction.

[0005] In order to achieve the above technical purposes and achieve the above technical effects, the present application realizes the following technical solutions:

[0006] In a first aspect, the present application provides a medium-term load forecasting method based on power feedback correction, comprising:

[0007] Input the selected first historical daily electricity consumption data into the preselected daily electricity consumption prediction model to obtain the daily electricity consumption prediction value of each day in the to-be-predicted period;

[0008] Based on the selected second historical daily electricity consumption data, and the corresponding historical daily weather data and historical holiday data, find out the historical similar day corresponding to each day in the to-be-predicted period;

[0009] Input the historical daily weather data and historical holiday data corresponding to each historical similar day into the pre-trained load-electricity consumption ratio prediction model to obtain the time load-day electricity consumption ratio corresponding to each day in the to-be-predicted period;

[0010] The daily time load prediction value of each day in the to-be-predicted period is calculated based on the daily electricity consumption prediction value of each day in the to-be-predicted period and the time load-daily electricity consumption ratio corresponding to each day in the to-be-predicted period.

[0011] With reference to the first aspect, optionally, the to-be-predicted period is one month, and the first historical daily electricity consumption data comprises all historical daily electricity consumption data of the two months before the to-be-predicted period.

[0012] With reference to the first aspect, optionally, the second historical daily electricity consumption data comprises all historical daily electricity consumption data of the month corresponding to the to-be-predicted period in the previous year and the month before the to-be-predicted period.

[0013] With reference to the first aspect, optionally, the historical similar day is obtained by the following method:

[0014] The correlation coefficient between the historical daily electricity consumption data and the historical daily meteorological data and the historical holiday data is calculated based on the historical daily meteorological data and the historical holiday data corresponding to the second historical daily electricity consumption data.

[0015] The historical multi-dimensional feature vector is constructed based on the historical daily meteorological data and the historical holiday data corresponding to the second historical daily electricity consumption data.

[0016] The historical multi-dimensional feature vector is standardized, and the standardized historical multi-dimensional feature vector is weighted according to the correlation coefficient between the historical daily electricity consumption data and the historical daily meteorological data and the historical holiday data, to obtain a historical daily normalized weighted feature vector.

[0017] The daily multi-dimensional feature vector corresponding to each day in the to-be-predicted period is constructed based on the predicted daily meteorological data and the holiday data of each day in the to-be-predicted period.

[0018] The Euclidean distance between the daily multi-dimensional feature vector corresponding to each day in the to-be-predicted period and the historical daily normalized weighted feature vector is calculated respectively, and the N historical days with the smallest Euclidean distance are selected to constitute the historical similar day of the to-be-predicted day.

[0019] With reference to the first aspect, optionally, the correlation coefficient between the historical daily electricity consumption data and the historical daily meteorological data is a Pearson correlation coefficient, and is calculated by the following calculation formula:

[0020]

[0021] The correlation coefficient between the historical daily electricity consumption data and the historical holiday data is a Spearman correlation coefficient, and is calculated by the following formula:

[0022] ,

[0023] wherein, represents a Pearson correlation coefficient, is the daily weather data of the day, is the daily electricity consumption data of the day, is the total number of days, is the daily weather data mean value, is the daily electricity consumption data mean value; represents a Spearman correlation coefficient, represents the rank difference of the day in the historical daily electricity consumption data sequence and the historical holiday data sequence.

[0024] In combination with the first aspect, optionally, the standardization processing of the historical multi-dimensional feature vector is specifically:

[0025] The historical multi-dimensional feature vector is normalized.

[0026] In combination with the first aspect, optionally, the daily electricity consumption prediction model is an LSTM neural network; and the load-electricity consumption ratio prediction model is an SVR model, and in the training process, the input data of the model includes historical daily weather data and historical holiday data corresponding to the historical similar day, and the label data is the time point load-daily electricity consumption ratio corresponding to the historical similar day.

[0027] In combination with the first aspect, optionally, the medium-term load prediction method further includes pre-processing the first historical daily electricity consumption data and the second historical daily electricity consumption data, and the pre-processing includes:

[0028] The first historical daily electricity consumption data and the second historical daily electricity consumption data are subjected to abnormal value detection by using the quartile method;

[0029] The abnormal value is taken as a missing value, and the missing value is filled by using a linear interpolation algorithm.

[0030] In the second aspect, the application provides a medium-term load prediction device based on electricity consumption feedback correction, which comprises:

[0031] A daily electricity consumption prediction module is configured to input the selected first historical daily electricity consumption data into a preselected daily electricity consumption prediction model to obtain daily daily electricity consumption prediction values in a to-be-predicted period.

[0032] A historical similar day determination module is configured to find out historical similar days corresponding to each day in the to-be-predicted period based on the selected second historical daily electricity consumption data and historical daily weather data and historical holiday data corresponding thereto.

[0033] The time moment load-daily power consumption ratio calculation module is configured to input historical daily meteorological data and historical holiday data corresponding to each historical similar day into a pre-trained load-power consumption ratio prediction model to obtain time moment load-daily power consumption ratios corresponding to each day in the to-be-predicted period.

[0034] The time moment load prediction value calculation module is configured to calculate time moment load prediction values of each day in the to-be-predicted period based on the daily power consumption prediction values of each day in the to-be-predicted period and the time moment load-daily power consumption ratios corresponding to each day in the to-be-predicted period.

[0035] In a third aspect, the present application provides a medium-term load prediction system based on power feedback correction, comprising a storage medium and a processor.

[0036] The storage medium is configured to store instructions.

[0037] The processor is configured to operate according to the instructions to perform the method according to any one of the first aspect.

[0038] Compared with the prior art, the present application has the following beneficial effects:

[0039] The present application adopts a two-stage coupling architecture, first performs medium-term prediction of daily power consumption of each day in the to-be-predicted period, and then performs load-power decomposition based on similar days (i.e., calculates time moment load-daily power consumption ratios corresponding to each day in the to-be-predicted period) to finally realize medium-term prediction of load. The architecture dynamically feedbacks and corrects the load decomposition process through the prediction values of daily power consumption of each day in the to-be-predicted period, significantly suppresses the cumulative amplification effect of errors in long sequence prediction, and thus provides a reliable data basis for medium-term dispatching plan and power transaction decision of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor, wherein:

[0041] Figure 1 The flowchart of the medium-term load prediction based on power feedback correction of an embodiment of the present application;

[0042] Figure 2 The flowchart of predicting daily power consumption of the next 30 days through historical 90-day daily power consumption of an embodiment of the present application;

[0043] Figure 3 The correlation coefficient calculation result diagram of an embodiment of the present application;

[0044] Figure 4 This is a comparison chart of the predicted and actual values ​​of the load-daily electricity consumption ratio at any given moment, according to an embodiment of the present invention.

[0045] Figure 5 This is a schematic diagram of the daily time-based load forecast values ​​within the forecast period according to an embodiment of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0047] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0048] Example 1

[0049] This invention provides a medium-term load forecasting method based on power feedback correction, comprising the following steps:

[0050] (1) Input the selected first historical daily electricity consumption data into the pre-selected daily electricity consumption prediction model to obtain the daily electricity consumption prediction value for each day in the prediction period;

[0051] (2) Based on the selected second historical day electricity consumption data, as well as the corresponding historical day meteorological data and historical holiday data, find the historical similar days corresponding to each day in the period to be predicted;

[0052] (3) Input the historical meteorological data and historical holiday data corresponding to each historical similar day into the pre-trained load-electricity consumption ratio prediction model to obtain the time-load-daily electricity consumption ratio corresponding to each day in the prediction period; the time-load-daily electricity consumption ratio refers to the ratio of time-load to daily electricity consumption; in the specific implementation process, 24 time-load-daily electricity consumption ratios can be set according to 24 hours;

[0053] (4) Based on the predicted daily electricity consumption for each day within the forecast period and the corresponding time-load-daily electricity consumption ratio for each day within the forecast period, calculate the predicted time-load value for each day within the forecast period. In specific implementation, the method for obtaining the predicted time-load value includes: multiplying the predicted daily electricity consumption for each day within the forecast period by the time-load-daily electricity consumption ratio for each time period to obtain several predicted time-load values.

[0054] The above scheme employs a two-stage coupled architecture. First, it performs a medium-term forecast of daily electricity consumption within the forecast period. Then, it performs load-electricity decomposition based on similar days (i.e., calculates the load-electricity ratio at the corresponding time point within the forecast period), ultimately achieving medium-term load forecasting. This architecture uses the daily electricity consumption forecast values ​​within the forecast period to dynamically correct the load decomposition process, significantly suppressing the cumulative amplification effect of errors in long-sequence forecasts, thereby providing a reliable data foundation for medium-term power system dispatch planning and power trading decisions.

[0055] In one specific embodiment of the present invention, the prediction period is one month, and the first historical daily electricity consumption data includes all historical daily electricity consumption data for the two months preceding the prediction period.

[0056] In the above scheme, by setting the first historical daily electricity consumption data to all historical daily electricity consumption data of the two months prior to the forecast period (i.e., 60 days of historical daily electricity consumption data), it can effectively include multiple short-term cycles (such as weekly cycles) and longer-term trend changes (such as monthly upward or downward trends), providing richer contextual information for the daily electricity consumption forecasting model, enabling it to better understand and model the inherent laws of electricity load.

[0057] In one specific embodiment of the present invention, the second historical daily electricity consumption data includes all historical daily electricity consumption data for the month preceding the forecast period and the corresponding month of the previous year.

[0058] In the above scheme, by setting the second historical daily electricity consumption data to include all historical daily electricity consumption data of the month before the forecast period and the month corresponding to the forecast period in the previous year, and by utilizing both "annual periodicity" and "monthly continuity" information, the similar days found not only conform to long-term patterns but also closely resemble recent conditions.

[0059] In one specific embodiment of the present invention, the historical similarity dates are obtained through the following method:

[0060] Based on the historical daily meteorological data and historical holiday data corresponding to the second historical day's electricity consumption data, the correlation coefficient between the historical daily electricity consumption data and the historical daily meteorological data and historical holiday data is calculated.

[0061] Based on the historical meteorological data and historical holiday data corresponding to the electricity consumption data of the second historical day, a historical multidimensional feature vector is constructed. In the specific implementation process, the historical multidimensional feature vector can be obtained by arranging the historical meteorological data and historical holiday data corresponding to each historical day in chronological order.

[0062] The historical multidimensional feature vector is standardized, and the standardized historical multidimensional feature vector is weighted according to the correlation coefficient between historical daily electricity consumption data, historical daily meteorological data, and historical holiday data to obtain the historical daily normalized weighted feature vector.

[0063] Based on the forecast day meteorological data and holiday data of each day within the forecast period, a daily multidimensional feature vector is constructed for each day within the forecast period. In specific implementation, the daily multidimensional feature vector can be obtained by arranging the forecast day meteorological data and holiday data of each day within the forecast period according to the order of the days.

[0064] The Euclidean distance between the daily multidimensional feature vector and the normalized weighted feature vector of historical days within the prediction period is calculated for each day. The N historical days with the smallest Euclidean distance are then selected to constitute the historical similar days for the day to be predicted. In practice, the value of N can be set according to actual needs, for example, N=5.

[0065] Based on the above scheme, the load-electricity consumption ratio prediction model can focus on learning the data patterns most relevant to the prediction target, eliminating irrelevant noise interference, thereby obtaining a more accurate and professional prediction model.

[0066] In one specific embodiment of the present invention, the correlation coefficient between the historical daily electricity consumption data and the historical daily meteorological data (including continuous variables such as daily average temperature and precipitation) is the Pearson correlation coefficient, which is calculated using the following formula:

[0067]

[0068] The correlation coefficient between the historical daily electricity consumption data and the historical number of holidays (including non-continuous variables such as week number, holiday type, and month) is the Spearman correlation coefficient, which is calculated using the following formula:

[0069] ,

[0070] In the formula, This represents the Pearson correlation coefficient. For the first Daily weather data for the day, For the first Daily electricity consumption data Total number of days This is the average of daily meteorological data. This represents the average daily electricity consumption. This represents the Spearman correlation coefficient. Indicates the first The rank difference between the historical daily electricity consumption data series and the historical holiday data series.

[0071] In one specific embodiment of the present invention, the standardization process for the historical multidimensional feature vector specifically involves:

[0072] The historical multidimensional feature vectors are normalized.

[0073] In the above scheme, normalization can eliminate dimensional differences and improve the model's robustness to small data fluctuations and outliers. In specific implementation, the historical multidimensional feature vector can be set as a five-dimensional feature vector composed of daily average temperature, daily average rainfall, monthly data, holiday data, and week number; the standardization of the historical multidimensional feature vector represents complementary normalization of the five-dimensional feature vector; the historical daily normalized weighted feature vector is calculated using the following formula:

[0074] ,

[0075] In the formula, For historical daily normalized weighted feature vectors, The correlation coefficients are for daily average temperature, daily average rainfall, monthly data, holiday data, and week number, respectively. These are the external feature vectors for the historical 60 days and the predicted 30 days, respectively.

[0076] In one specific embodiment of the present invention, the daily electricity consumption prediction model is an LSTM neural network; the load-electricity consumption ratio prediction model is an SVR model. During the training process, the input data of the model includes historical meteorological data and historical holiday data corresponding to historical similar days, and the label data is the load-daily electricity consumption ratio at the time corresponding to historical similar days.

[0077] In its implementation, the LSTM neural network comprises an input layer, a hidden layer, and an output layer. The hidden layer consists of several LSTM units, each containing a memory unit and three gating structures: a forget gate, an input gate, and an output gate. These gating structures collectively filter and transmit historical information. The forget gate controls the degree to which the memory unit's state information from the previous time step is retained. Specifically, firstly, the forget gate takes the hidden state from the previous time step and the input from the current time step as inputs, and maps the output to the [0,1] interval using the sigmoid activation function to determine the forgetting ratio of the previous time step's memory state. Secondly, the unit state is updated using the combined action of the forget gate and the input gate. Finally, the output gate controls how much of the current unit state can be output. The output gate first takes the output of the previous time step's daily electricity consumption and the input from the current time step as inputs, and controls the output range to [0,1] using the sigmoid function; then, it controls the current time step's unit state output to [−1,1] using a function; finally, it obtains the output for the current time step.

[0078] In one specific embodiment of the present invention, the medium-term load forecasting method further includes preprocessing the first historical daily electricity consumption data and the second historical daily electricity consumption data, the preprocessing including:

[0079] The quartile method was used to detect outliers in the electricity consumption data of the first and second historical days.

[0080] Outliers are treated as missing values, and a linear interpolation algorithm is used to fill in the missing values.

[0081] The outlier detection method uses the quartile method, and the calculation formula is as follows:

[0082] ,

[0083] The calculation formula used for missing value imputation is as follows:

[0084] ,

[0085] In the formula, This indicates the point in time at which to determine if daily electricity consumption is abnormal. , These represent the upper quartile and the lower quartile, respectively. Indicates the interquartile range. , Indicates if The value that needs to be replaced in case of an anomaly. This represents the time point when X^'s daily electricity consumption was normal. for The corresponding daily electricity consumption for The next day's electricity consumption will be at a time when it is at a normal level. for The corresponding daily electricity consumption.

[0086] The medium-term load forecasting method based on power feedback correction in this embodiment of the invention will be described in detail below with reference to a specific implementation method.

[0087] The data used in this embodiment are electricity load and consumption data from a district / county in Nanjing City, State Grid, CMIP6 weather data, and government-published holiday types and dates. The dataset covers the period from June 2023 to July 2024, with load data at a time granularity of 1 hour and electricity consumption data at a time granularity of 1 day. External influencing factors include meteorological factors such as daily average temperature, daily average wind speed, and daily total precipitation, as well as holiday types and week numbers, with a time granularity of 1 day.

[0088] Step 1: After acquiring the data, preprocess the data, use the quartile method to detect outliers, and then use linear interpolation to imput outliers and missing values.

[0089] Step 2: Construct a Long Short-Term Memory (LSTM) artificial neural network. A model sample set is built based on the daily electricity consumption sequence from June 1, 2023 to May 31, 2024, with the test set from June 1, 2024 to June 30, 2024. For the model sample set, the daily electricity consumption over 60 consecutive days is used as the model training input, and the daily electricity consumption over the following 30 days is used as the model training output. This allows the model to predict the electricity consumption for the next 30 days based on the historical daily electricity consumption over the past 90 days. Figure 2 As shown.

[0090] Step 3: To quantify the impact of external factors on electricity consumption changes, the correlation between daily electricity consumption and these external factors is calculated. For continuous external factors (daily average temperature, precipitation), Pearson correlation coefficient is used; for discrete external factors (weekly ordinal number, holiday code, monthly ordinal number), Spearman correlation coefficient is used. For example... Figure 3 As shown.

[0091] Step 4: Construct corresponding multi-dimensional feature vectors based on meteorological data and holiday data for historical and predicted dates, respectively. Normalize the multi-dimensional feature vectors, and then weight the normalized multi-dimensional feature vectors with the correlation coefficients (Pearson correlation coefficient or Spearman correlation coefficient) calculated in Step 3. Calculate the Euclidean distance between the weighted normalized feature vectors of the predicted date and historical dates. Take the data from the same month last year and the previous month for the predicted date, and select the 5 historical dates with the closest distance to the predicted date's feature vector. Combine these to form a historical similar day dataset of 10 samples for the predicted date. Taking June 1, 2024 as an example, its multi-dimensional feature vector and normalized vector are [29,19,0,6,1] and [0.3,0.35,0,0.01,0.04], respectively. The constructed historical similar day dataset is shown in Table 1 below. The same process is followed to complete the construction of historical similar day datasets for the remaining dates in June 2024.

[0092] Table 1. Historical Dataset of Similar Dates on June 1, 2024

[0093]

[0094] Step 5: For a given forecast date, based on the load data and daily electricity consumption data of historical similar days, calculate the load-to-daily electricity consumption ratio for that forecast date. Then, train the constructed Support Vector Machine Regression (SVR) model using the multi-dimensional feature vectors corresponding to the historical similar days. Finally, by inputting the multi-dimensional feature vectors (composed of daily meteorological data and historical holiday data) of the historical similar days corresponding to the forecast date, the load-to-daily electricity consumption ratio for the forecast time is calculated, as shown below. Figure 4 As shown. Combined with the predicted daily electricity consumption from step 2, the load for the predicted day is obtained, as shown. Figure 5 As shown.

[0095] Example 2

[0096] Based on the same inventive concept as Embodiment 1, this embodiment of the invention provides a medium-term load forecasting device based on power feedback correction, comprising:

[0097] The daily electricity consumption forecasting module is used to input the selected first historical daily electricity consumption data into the pre-selected daily electricity consumption forecasting model to obtain the daily electricity consumption forecast value within the forecast period.

[0098] The historical similar day determination module is used to find historical similar days corresponding to each day in the forecast period based on the selected second historical day's electricity consumption data, as well as its corresponding historical day's meteorological data and historical holiday data.

[0099] The time-load-daily electricity consumption ratio calculation module is used to input historical daily meteorological data and historical holiday data corresponding to each historical similar day into the pre-trained load-electricity consumption ratio prediction model to obtain the time-load-daily electricity consumption ratio corresponding to each day in the prediction period.

[0100] The time-based load forecast calculation module is used to calculate the time-based load forecast value for each day in the forecast period based on the daily electricity consumption forecast value for each day in the forecast period and the time-based load-daily electricity consumption ratio corresponding to each day in the forecast period.

[0101] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0102] This invention employs a two-stage coupled architecture. First, it performs a medium-term forecast of daily electricity consumption within the forecast period. Then, it performs load-electricity decomposition based on similar days (i.e., calculates the load-to-electricity ratio at the corresponding time point within the forecast period), ultimately achieving medium-term load forecasting. This architecture uses the predicted daily electricity consumption within the forecast period to dynamically correct the load decomposition process, significantly suppressing the cumulative amplification effect of errors in long-sequence forecasts, thereby providing a reliable data foundation for medium-term power system dispatch planning and power trading decisions.

[0103] Example 3

[0104] This invention provides a medium-term load forecasting system based on power feedback correction, including a storage medium and a processor;

[0105] The storage medium is used to store instructions;

[0106] The processor is configured to operate according to the instructions to execute the method according to any one of Embodiment 1.

[0107] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0111] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0112] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A medium-term load forecasting method based on power feedback correction, characterized in that, include: Input the selected first historical daily electricity consumption data into the pre-selected daily electricity consumption prediction model to obtain the daily electricity consumption prediction value for the period to be predicted. Based on the selected second historical day's electricity consumption data, as well as its corresponding historical day's meteorological data and historical holiday data, identify historically similar days that correspond to each day within the forecast period. Historical meteorological data and historical holiday data corresponding to each historically similar day are input into a pre-trained load-electricity consumption ratio prediction model to obtain the load-electricity consumption ratio for each time period corresponding to each day within the prediction period. Based on the predicted daily electricity consumption for each day within the forecast period, and the ratio of time-based load to daily electricity consumption for each day within the forecast period, the predicted time-based load for each day within the forecast period is calculated.

2. The medium-term load forecasting method based on power feedback correction according to claim 1, characterized in that: The forecast period is one month, and the first historical daily electricity consumption data includes all historical daily electricity consumption data for the two months preceding the forecast period.

3. The medium-term load forecasting method based on power feedback correction according to claim 2, characterized in that: The second historical daily electricity consumption data includes all historical daily electricity consumption data for the month preceding the forecast period and the month corresponding to the forecast period in the previous year.

4. A medium-term load forecasting method based on power feedback correction according to claim 1 or 3, characterized in that: The historical similarity dates were obtained through the following methods: Based on the historical daily meteorological data and historical holiday data corresponding to the second historical day's electricity consumption data, the correlation coefficient between the historical daily electricity consumption data and the historical daily meteorological data and historical holiday data is calculated. Based on historical daily meteorological data and historical holiday data corresponding to the second historical day's electricity consumption data, a historical multidimensional feature vector is constructed. The historical multidimensional feature vector is standardized, and the standardized historical multidimensional feature vector is weighted according to the correlation coefficient between historical daily electricity consumption data, historical daily meteorological data, and historical holiday data to obtain the historical daily normalized weighted feature vector. Based on the forecast day meteorological data and holiday data of each day within the forecast period, construct the daily multidimensional feature vector corresponding to each day within the forecast period; Calculate the Euclidean distance between the daily multidimensional feature vector and the normalized weighted feature vector of the historical days for each day in the period to be predicted, and select the N historical days with the smallest Euclidean distance to form the historical similar days of the day to be predicted.

5. The medium-term load forecasting method based on power feedback correction according to claim 4, characterized in that: The correlation coefficient between the historical daily electricity consumption data and the historical daily meteorological data is the Pearson correlation coefficient, which is calculated using the following formula: The correlation coefficient between the historical daily electricity consumption data and the historical holiday data is the Spearman correlation coefficient, which is calculated using the following formula: ， In the formula, This represents the Pearson correlation coefficient. For the first Daily weather data for the day, For the first Daily electricity consumption data Total number of days This is the average of daily meteorological data. This represents the average daily electricity consumption. This represents the Spearman correlation coefficient. Indicates the first The rank difference between the historical daily electricity consumption data series and the historical holiday data series.

6. The medium-term load forecasting method based on power feedback correction according to claim 4, characterized in that: The standardization process for the historical multidimensional feature vector specifically involves: The historical multidimensional feature vectors are normalized.

7. The medium-term load forecasting method based on power feedback correction according to claim 1, characterized in that: The daily electricity consumption prediction model is an LSTM neural network; the load-electricity consumption ratio prediction model is an SVR model. During the training process, the input data of this model includes historical meteorological data and historical holiday data corresponding to historical similar days, and the label data is the load-daily electricity consumption ratio at the time corresponding to historical similar days.

8. The medium-term load forecasting method based on power feedback correction according to claim 1, characterized in that: The medium-term load forecasting method further includes preprocessing the first historical daily electricity consumption data and the second historical daily electricity consumption data. The preprocessing includes: The quartile method was used to detect outliers in the electricity consumption data of the first and second historical days. Outliers are treated as missing values, and a linear interpolation algorithm is used to fill in the missing values.

9. A medium-term load forecasting device based on power feedback correction, characterized in that, include: The daily electricity consumption forecasting module is used to input the selected first historical daily electricity consumption data into the pre-selected daily electricity consumption forecasting model to obtain the daily electricity consumption forecast value within the forecast period. The historical similar day determination module is used to find historical similar days corresponding to each day in the forecast period based on the selected second historical day's electricity consumption data, as well as its corresponding historical day's meteorological data and historical holiday data. The time-load-daily electricity consumption ratio calculation module is used to input historical daily meteorological data and historical holiday data corresponding to each historical similar day into the pre-trained load-electricity consumption ratio prediction model to obtain the time-load-daily electricity consumption ratio corresponding to each day in the prediction period. The time-based load forecast calculation module is used to calculate the time-based load forecast value for each day in the forecast period based on the daily electricity consumption forecast value for each day in the forecast period and the time-based load-daily electricity consumption ratio corresponding to each day in the forecast period.

10. A medium-term load forecasting system based on power feedback correction, characterized in that, Including storage media and processor; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1-8.

Images