Load prediction method based on weather forecast correction and temperature accumulation index

Abstract

The invention discloses a load prediction method based on weather forecast correction and a temperature accumulation index, and relates to the technical field of load prediction, and the method comprises the steps: S1, obtaining a first forecast total error of historical weather forecast; s2, calculating a current forecast error rate based on current weather information and the first forecast total error, and performing primary correction on weather forecast data according to the error rate; s3, calculating a temperature accumulation index according to the influence of the temperature accumulation effect on the load; s4, building a temperature correction model based on the temperature accumulation index, and carrying out the secondary correction of the meteorological prediction data; s5, constructing a BP neural network prediction model, and carrying out load prediction according to the secondarily corrected meteorological prediction data; the problem of low load prediction accuracy caused by lack of accurate prediction of future weather conditions in the prior art is solved, and the accuracy and stability of load prediction are remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of load forecasting, and in particular to a load forecasting method based on weather forecast correction and temperature accumulation index. Background Art

[0002] With the surge in air conditioning use, especially in urban areas, this has significantly increased electricity demand in the summer. These changes have brought new challenges to the planning, operation and reliability of the power system. Therefore, urban planners and energy policy makers need to take these factors into account and incorporate the impact of meteorological factors on actual load usage into load forecasts to ensure the sustainability and resilience of the power system.

[0003] The current load forecasting methods are mainly divided into three categories: mathematical models, statistical models and artificial intelligence models. Among them, mathematical models include regression models, Gaussian processes, gray prediction, Markov chains, etc. They are usually based on mathematical formulas to predict loads. For example, panel data regression models can be used for aggregate load forecasting in specific areas. Deep learning methods; statistical models mainly include autoregression (AR), moving average (MA), autoregressive moving average (ARMA), etc., which predict future load demand by analyzing historical load data; artificial intelligence models can be divided into two categories: machine learning and deep learning. Machine learning techniques include decision trees, random forests, and support vector machines. For example, prediction models based on multi-task learning and least squares support vector machines can significantly improve prediction accuracy and effectively shorten training time. Deep learning models mainly include convolutional neural networks (CNN), gated recurrent units (GRU), and long short-term memory networks (LSTM). These models improve prediction accuracy by automatically extracting relevant features from data sets. However, existing methods have problems such as insufficient nonlinear processing, insufficient generalization and adaptability, strong data dependence, and drastic fluctuations in load curves, especially under extreme weather conditions, which limits the accuracy and robustness of current load forecasting.

[0004] Chinese patent, publication number: CN111461463A, publication date: July 28, 2020, discloses a short-term load forecasting method, system and equipment based on TCN-BP, which collects load data and main meteorological factor data, pre-processes the data and performs grey correlation analysis, and uses a discretized temperature correction model to correct the daily maximum temperature, daily average temperature and daily minimum temperature; divides the meteorological factor data into time series data and non-time series data, and performs TCN dimension reduction processing on the historical time series data; uses the data after dimension reduction processing and non-time series data as input, and the load data as output, and substitutes them into the BP neural network for training until the network converges; uses the trained back propagation network to complete the load forecast and output the load forecast data. The invention corrects the temperature based on special temperature values, such as the daily maximum and minimum and daily average temperatures, but the meteorological data changes dynamically. Although the average value may eliminate certain errors, it cannot be used as the data for future meteorological forecasts for load forecasting, and cannot accurately reflect the changes in the load curve under extreme weather, and has poor adaptability. Summary of the invention

[0005] The purpose of the present invention is to address the problem that the prior art lacks accurate prediction of future weather conditions, resulting in low load prediction accuracy; a load prediction method based on weather forecast correction and temperature accumulation index is proposed, by making a primary correction and a secondary correction to the current weather forecast, a BP neural network prediction model is constructed to perform load prediction on the meteorological forecast data after the secondary correction, thereby avoiding load prediction errors caused by inaccurate weather forecasts, and achieving more efficient fitting of the complex relationship between load and influencing factors, making load prediction more intelligent and automated, and significantly improving the accuracy and stability of load prediction, and overcoming the problem that the prior art lacks accurate prediction of future weather conditions, resulting in low load prediction accuracy.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a load forecasting method based on weather forecast correction and temperature accumulation index, comprising the following steps: S1, obtaining the first forecast total error of historical weather forecast; S2. Calculating a current forecast error rate based on the current meteorological information and the first forecast total error, and correcting the meteorological forecast data once according to the error rate; S3. Calculate the temperature accumulation index based on the impact of the temperature accumulation effect on the load; S4, constructing a temperature correction model based on the temperature accumulation index, and performing secondary correction on the meteorological forecast data; S5. Construct a BP neural network prediction model and perform load prediction based on the secondary corrected meteorological forecast data.

[0007] In this scheme, by calculating the total error of historical weather forecasts, statistical characteristics are provided for the correction of current weather forecast errors, which can ensure the accuracy of the current weather forecast correction; the error rate of the current weather forecast (i.e., the current forecast error rate) is calculated using the historical weather forecast errors and current meteorological information, so as to determine whether the current weather forecast data needs to be corrected. If the error rate is within the threshold range, it means that the current weather forecast data does not need to be corrected and can be used directly for load forecasting. If the error rate is not within the threshold range, the current weather forecast data is corrected once, which reduces the load forecast error caused by inaccurate weather forecasts, and at the same time improves the adaptability of load forecasts to weather changes, making the forecast more accurate. The results are closer to reality; calculating the temperature accumulation index of the load according to the changing characteristics of the load under the temperature accumulation effect can make the load forecast more in line with the actual situation, so as to facilitate the secondary correction of the current weather forecast data, and further reduce the load forecast error caused by temperature changes; by constructing a BP neural network prediction model, the advantages of the BP neural network with powerful nonlinear mapping ability and self-learning ability, and the ability to process a large amount of nonlinear data, are used to perform load forecasting on the current weather forecast data with secondary correction, so that the model can more efficiently fit the complex relationship between the load and the influencing factors, realize more intelligent and automated load forecasting, and improve the accuracy and stability of load forecasting.

[0008] Preferably, S1 comprises the following sub-steps: Acquire future weather information through weather forecasts, and collect historical weather data based on the future weather information, wherein the historical weather data includes historical measured weather data of weather stations and historical network weather forecast data within a corresponding period of time; Calculating the historical time error of the historical weather forecast according to the historical measured weather data and the historical network weather forecast data; The first total forecast error of the historical weather forecast is calculated based on the weight coefficient corresponding to the historical moment error.

[0009] Preferably, the first total forecast error of the historical weather forecast is calculated based on the weight coefficient corresponding to the historical moment error, including: Based on the time distance between the historical moment error and the predicted day, a weighted average method is used to determine the weight coefficient corresponding to each historical moment error; The sum of the products of the historical moment errors and the corresponding weight coefficients is taken as the first forecast total error.

[0010] Preferably, S2 comprises the following sub-steps: Obtaining current weather forecast data and measured meteorological data of the target area, and calculating a current forecast error of the current weather forecast; calculating a current forecast error rate of the current weather forecast by the ratio of the current forecast error to the first forecast total error at the corresponding historical moment; The current forecast error rate is compared with a threshold value, and the first forecast total error is proportionally converted according to the comparison result, and then the short-term meteorological forecast data of the target area is corrected to obtain meteorological forecast correction data.

[0011] Preferably, the method of performing a correction on the short-term meteorological forecast data of the target area after proportionally converting the first forecast total error according to the comparison result to obtain a meteorological forecast correction data includes: If the current forecast error rate is not within the threshold range, multiplying the first forecast total error by the current forecast error rate to proportionally convert the first forecast total error to obtain a second forecast total error of the historical weather forecast; The meteorological forecast data is subtracted from the second total forecast error to obtain primary meteorological forecast correction data, thereby completing a primary correction of the meteorological forecast data.

[0012] Preferably, the temperature accumulation index includes a limit temperature index, a maximum accumulation days index and a cumulative effect index; wherein: By fitting the time variation curve of temperature and load, the limit temperature index of load variation is obtained; Calculate the correlation between the maximum temperature and the maximum load, and use the cumulative number of days with the maximum correlation as the maximum cumulative number of days index; The cumulative effect index is calculated based on the limit temperature index and the maximum cumulative days index.

[0013] Preferably, the calculation of the correlation between the maximum temperature and the maximum load includes: Calculating the correlation degree based on the limit temperature index and the second total forecast error; The calculation formula of the correlation degree is as follows: In the formula, R represents the degree of correlation, T i represents the limit temperature index value of the i-th cumulative day, Indicates the average value of the limit temperature index, L i represents the total error of the second forecast on the ith cumulative day, represents the average value of the total error of the second forecast, and n represents the cumulative number of days.

[0014] Preferably, the cumulative effect index is calculated based on the limit temperature index and the maximum cumulative days index, including: Establishing a cumulative effect coefficient of loads at different temperatures based on the limit temperature index and the maximum cumulative days index; Discretizing the cumulative effect coefficient to obtain a cumulative effect coefficient sequence; Calculating a corrected correlation coefficient between temperature and load based on the primary weather forecast correction data and the limit temperature index; It is determined whether the cumulative effect coefficient sequence meets the cumulative effect specification by using the modified correlation coefficient. If so, the cumulative effect index is determined based on the cumulative effect coefficient sequence.

[0015] Preferably, S4 includes the following sub-steps: A temperature correction model is constructed based on the limit temperature index, the maximum cumulative days index and the cumulative effect index; the primary meteorological forecast correction data is calculated based on the temperature correction model to obtain secondary meteorological forecast correction data, thereby completing secondary correction of meteorological forecast data.

[0016] Preferably, S5 comprises the following sub-steps: Acquire the historical load data of the target area, use the historical load data and the historical meteorological data as inputs of a BP neural network model for model training, and construct a BP neural network prediction model; The secondary meteorological forecast correction data is predicted based on the BP neural network prediction model, and the load prediction value of the target area is output.

[0017] In this scheme, by establishing a temperature correction model, the cumulative impact of continued high temperature on load growth is quantified, and a load forecasting model with BP neural network as the structural feature is established by combining historical load data, historical meteorological data and corrected weather forecast data. This enables the load forecasting model to have nonlinear mapping and self-learning capabilities, and to be able to process a large amount of nonlinear data, thereby achieving more efficient fitting of the complex relationship between load and influencing factors, and effectively improving the robustness and accuracy of the load forecasting model.

[0018] Beneficial effects of the present invention: 1. By correcting the current weather forecast data once and twice, the load forecast error caused by inaccurate weather forecast is reduced, the adaptability of load forecast to weather changes is improved, and the forecast results are closer to reality; 2. According to the correlation between temperature and load and the fitting curve of temperature and load, the limit temperature index, the maximum cumulative days index and the cumulative effect index are calculated, and the cumulative effect of temperature on load is taken into account, so that the load forecast is more in line with the actual situation and the load forecast error caused by temperature changes is further reduced; 3. By building a load forecasting model through BP neural network and performing load forecasting on the secondary corrected meteorological forecast data, it can fit the complex relationship between load and influencing factors more efficiently and accurately, thereby improving the accuracy and stability of load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0020] Figure 1 The present invention is a flow chart of a load forecasting method based on weather forecast correction and temperature accumulation index according to an embodiment.

[0021] Figure 2 The figure is a schematic diagram of power mutation detection according to a specific embodiment.

[0022] Figure 3 It is a schematic diagram of the forecast results before and after the weather correction of a specific embodiment.

[0023] Figure 4 It is a schematic diagram of prediction results of a different modeling method according to a specific embodiment. DETAILED DESCRIPTION

[0024] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] Example 1: Figure 1 As shown, a load forecasting method based on weather forecast correction and temperature accumulation index includes steps S1-S5, wherein: S1. Obtain the total error of the first forecast of historical weather forecasts.

[0026] Preferably, S1 comprises the following sub-steps: Acquire future weather information through weather forecasts, and collect historical weather data based on the future weather information, wherein the historical weather data includes historical measured weather data of weather stations and historical network weather forecast data within a corresponding period of time; Calculating the historical time error of the historical weather forecast according to the historical measured weather data and the historical network weather forecast data; The first total forecast error of the historical weather forecast is calculated based on the weight coefficient corresponding to the historical moment error.

[0027] Preferably, the first total forecast error of the historical weather forecast is calculated based on the weight coefficient corresponding to the historical moment error, including: Based on the time distance between the historical moment error and the predicted day, a weighted average method is used to determine the weight coefficient corresponding to each historical moment error; The sum of the products of the historical moment errors and the corresponding weight coefficients is taken as the first forecast total error.

[0028] As an implementation method, by obtaining historical weather forecasts and meteorological data, the historical error of the weather forecast at each moment is calculated. (historical time error), determine The corresponding weight coefficient υ D ; The details are as follows: Determine the current date as day d, the current time as i, and obtain the future weather conditions through weather forecast as in Represents different weather conditions such as sunny, rainy, and cloudy. Select historical data based on the acquired weather conditions. Error at each moment in history The calculation formula is as follows: Where: is the weather data measured by the historical meteorological station at the corresponding time I; is the historical network weather forecast data corresponding to time I; I∈{i+1,i+2,...,i+q}; q is the weather forecast time domain; D∈{d-1,d-2,...,dn}; determine The corresponding weight coefficient υ D (Weight coefficient υ D The selection of adopts the weighted average method, which is determined according to the distance of historical error from the prediction date, such as The weight coefficient should be greater than υ d-1 The calculation method is as follows: According to the historical error at each moment and its weight coefficient to calculate the historical total error data EI (i.e., the total error of the first forecast), the calculation formula is as follows: S2. Calculate the current forecast error rate based on the current meteorological information and the first forecast total error, and make a correction to the meteorological forecast data according to the error rate.

[0029] Preferably, S2 comprises the following sub-steps: Obtaining current weather forecast data and measured meteorological data of the target area, and calculating a current forecast error of the current weather forecast; calculating a current forecast error rate of the current weather forecast by the ratio of the current forecast error to the first forecast total error at the corresponding historical moment; The current forecast error rate is compared with a threshold value, and the first forecast total error is proportionally converted according to the comparison result, and then the short-term meteorological forecast data of the target area is corrected to obtain meteorological forecast correction data.

[0030] Preferably, the method of performing a correction on the short-term meteorological forecast data of the target area after proportionally converting the first forecast total error according to the comparison result to obtain a meteorological forecast correction data includes: If the current forecast error rate is not within the threshold range, multiplying the first forecast total error by the current forecast error rate to proportionally convert the first forecast total error to obtain a second forecast total error of the historical weather forecast; The meteorological forecast data is subtracted from the second total forecast error to obtain primary meteorological forecast correction data, thereby completing a primary correction of the meteorological forecast data.

[0031] As an implementation method, after calculating the historical error, the measured weather data is obtained and the measured weather error E is calculated. i (current forecast error) and weather forecast accuracy ξ(current forecast error rate), the calculation formulas are as follows: E i =|W i -P i |(4) ξ=E i / E I·i (5) Where: P i Real-time meteorological data for weather forecast; W i Real-time monitoring data for weather stations; E I·i is the total error at historical moment i.

[0032] At present, the accuracy of the weather forecast of the China Meteorological Administration can generally reach 80%. Therefore, in this embodiment, the threshold range is (0.8, 1.25); if the weather forecast is inaccurate, it will be out of the threshold range, which also means that there is a certain deviation in correcting the measured value using the historical error. Therefore, it is necessary to multiply the historical error by ξ for proportional conversion. The conversion formula is as follows: L=E I ·ξ(7) Where: γ is the upper threshold of the accuracy coefficient; χ is the lower threshold of the accuracy coefficient, E correct represents the second forecast total error, L is one of the values ​​of the second forecast total error; Finally, the regional short-term weather forecast data (first meteorological forecast correction data) is obtained: W FOR·I W FOR·I =W CET·I -E correct (8) Where W CET·I Indicates the current weather forecast data, W FOR·I W FOR·I Indicates the weather forecast correction data after a correction.

[0033] S3. Calculate the temperature accumulation index based on the impact of the temperature accumulation effect on the load.

[0034] Preferably, the temperature accumulation index includes a limit temperature index, a maximum accumulation days index and a cumulative effect index; wherein: By fitting the time variation curve of temperature and load, the limit temperature index of load variation is obtained; Calculate the correlation between the maximum temperature and the maximum load, and use the cumulative number of days with the maximum correlation as the maximum cumulative number of days index; The cumulative effect index is calculated based on the limit temperature index and the maximum cumulative days index.

[0035] It can be understood that the limit temperature can generally be given by statistical laws or research experience, or by fitting the temperature and load curve. The point with the largest load change rate is the limit temperature.

[0036] In this embodiment, the limit temperature index, as a specific temperature value that affects the load change, can be used as a basis for dividing the load change interval, so as to more accurately predict the load change trend; when the temperature reaches or exceeds a certain limit temperature, the load may increase or decrease significantly. For example, in summer, when the temperature exceeds a certain value, the air conditioning load will increase significantly; and in winter, when the temperature is below a certain value, the heating load will increase significantly. By analyzing the load changes at different limit temperatures, the sensitivity of the load to temperature changes can be understood, so as to formulate a more accurate load forecasting strategy. The maximum cumulative days index can reflect the continuity and cumulative effect of the temperature on the load; for example, in continuous high or low temperature weather, the load will continue to accumulate until it reaches the maximum value. The maximum cumulative days index can measure this cumulative effect, thereby predicting the changing trend of the load under extreme weather conditions. By analyzing the maximum cumulative days in different temperature ranges, we can understand the accumulation speed and changing characteristics of the load under different temperature conditions, providing an important basis for load prediction and key data support for weather forecast corrections. The cumulative effect index takes into account the long-term impact of the persistence and accumulation of temperature changes on the load, and accurately reflects the dynamic relationship between temperature and load. Since the cumulative effect of temperature will have a continuous impact on the load, even if the temperature has not changed significantly at the current moment, the load may change due to the previous temperature accumulation. Therefore, by analyzing the cumulative effect index, we can understand the long-term impact of temperature on the load, thereby more accurately predicting the future trend of load changes.

[0037] Preferably, the calculation of the correlation between the maximum temperature and the maximum load includes: Calculating the correlation degree based on the limit temperature index and the second total forecast error; The calculation formula of the correlation degree is as follows: In the formula, R represents the degree of correlation, T i represents the limit temperature index value of the i-th cumulative day, Indicates the average value of the limit temperature index, L i represents the total error of the second forecast on the ith cumulative day, represents the average value of the total error of the second forecast, and n represents the cumulative number of days.

[0038] In this embodiment, for the maximum cumulative number of days p, a trial method is used to calculate the correlation between the maximum temperature and the maximum load under different cumulative numbers of days, and finally the cumulative day with the largest correlation is selected as the maximum cumulative number of days p.

[0039] Preferably, the cumulative effect index is calculated based on the limit temperature index and the maximum cumulative days index, including: Establishing a cumulative effect coefficient of loads at different temperatures based on the limit temperature index and the maximum cumulative days index; Discretizing the cumulative effect coefficient to obtain a cumulative effect coefficient sequence; Calculating a corrected correlation coefficient between temperature and load based on the primary weather forecast correction data and the limit temperature index; It is determined whether the cumulative effect coefficient sequence meets the cumulative effect specification by using the modified correlation coefficient. If so, the cumulative effect index is determined based on the cumulative effect coefficient sequence.

[0040] In this embodiment, for the cumulative effect coefficient k ij The degree of influence of temperature accumulation effect will be different at different temperatures, that is, it is related to the temperature of the day to be predicted. The influence of the temperature on the day to be predicted is also different. Therefore, the cumulative effect coefficient k ij The discretized cumulative effect coefficient sequence is used, as shown in Table 1; and the revised correlation between temperature and load (i.e., revised correlation coefficient) is calculated through an iterative method to confirm whether it is the best k value within this range (i.e., whether it meets the cumulative effect specification); Table 1 Discretized cumulative effect coefficient sequence <![CDATA[T i / ℃]]> <![CDATA[k i1 ]]> <![CDATA[k i2 ]]> … <![CDATA[k ip ]]> <![CDATA[<T min ]]> 0 0 0 0 <![CDATA[[T min ,T min +1)]]> <![CDATA[k 11 ]]> <![CDATA[k 12 ]]> … <![CDATA[k 1p <!-- 7 -->]]> <![CDATA[[T min +1,T min +2)]]> <![CDATA[k 21 ]]> <![CDATA[k 22 ]]> … <![CDATA[k 2p ]]> <![CDATA[[T min +2,T min +3)]]> <![CDATA[k 31 ]]> <![CDATA[k 32 ]]> … <![CDATA[k 3p ]]> <![CDATA[[T min +3,T min +4)]]> <![CDATA[k 41 ]]> <![CDATA[k 42 ]]> … <![CDATA[k 4p ]]> … … … … … <![CDATA[≥T max ]]> 0 / 0 0 0 The calculation formula of the modified correlation coefficient R is as follows: k i1 >k i2 >... >k ip st0≤k ij ≤1 S4. Constructing a temperature correction model based on the temperature accumulation index to perform secondary correction on the meteorological forecast data.

[0041] Preferably, S4 includes the following sub-steps: A temperature correction model is constructed based on the limit temperature index, the maximum cumulative days index and the cumulative effect index; the primary meteorological forecast correction data is calculated based on the temperature correction model to obtain secondary meteorological forecast correction data, thereby completing secondary correction of meteorological forecast data.

[0042] Furthermore, the temperature correction model is as follows: Where, T i ' is the cumulative corrected temperature at the i-th moment of the predicted day; T i is the temperature at the i-th moment of the day to be measured; T minis the critical temperature for the temperature accumulation effect; T ij is the temperature at the i-th moment on the j-th day before the day to be predicted; p is the maximum cumulative number of days (p≥1); k ij ∈[0,1] is the temperature accumulation coefficient, and according to the principle that the closer to the predicted day, the greater the temperature accumulation effect, the k should be satisfied. i1 >k i2 >k i3 …>k ip .

[0043] In this embodiment, the weather forecast (meteorological forecast data) is corrected twice through the temperature correction model, which helps to more accurately reflect the impact of temperature on load, thereby improving the accuracy of load prediction. The cumulative effect of temperature will have a long-term impact on the load. Even if the temperature has not changed significantly at the current moment, the load may change due to the previous temperature accumulation. Through the temperature correction model, the cumulative effect of temperature change can be comprehensively considered, so that the load prediction sample data is more in line with the actual situation, which further helps to achieve efficient use of energy and reduce energy waste.

[0044] S5. Construct a BP neural network prediction model and perform load prediction based on the secondary corrected meteorological forecast data.

[0045] Preferably, S5 comprises the following sub-steps: Acquire the historical load data of the target area, use the historical load data and the historical meteorological data as inputs of a BP neural network model for model training, and construct a BP neural network prediction model; The secondary meteorological forecast correction data is predicted based on the BP neural network prediction model, and the load prediction value of the target area is output.

[0046] It can be understood that the BP neural network consists of an input layer, a hidden layer (also called an intermediate layer) and an output layer, where the hidden layer has one or more layers. Each layer can have several nodes. The connection status of the nodes between layers is reflected by weights; After the input is input, it is multiplied by the corresponding weight and then added along the direction of the network. The result is then used as input to calculate in the activation function, and the calculated result is passed as input to the next node. Calculations are performed in sequence until the final result is obtained; The output result is compared with the expected output result, and the error generated by the comparison is back-propagated through the network. This is essentially a "negative feedback" process. Through multiple iterations, the weights between the nodes on the network are continuously adjusted (updated), and the weight adjustment (update) adopts the gradient descent method.

[0047] In the process of forward propagation, there is a link to compare with the expected result to see if it is satisfactory. In this link, there will be an error between the actual output result and the expected output result. In order to reduce this error, it can be converted into an optimization process. For any optimization problem, there is always an objective function, which is called the loss function. Through the loss function, each time it moves in the direction of the negative gradient of the loss function until the loss function reaches the minimum value, in order to find the optimal parameters of the neural network, that is, the back propagation process, and its calculation formula is: In the formula, y i represents the actual value of each i-th sample, y represents the predicted value of each i-th sample, w represents the weight vector, and b represents the bias constant.

[0048] The beneficial effects of the embodiment are as follows: by calculating the total error of historical weather forecasts, statistical characteristics are provided for the correction of the current weather forecast error, which can ensure the accuracy of the current weather forecast correction; by using the historical weather forecast errors and the current meteorological information, the error rate of the current weather forecast (i.e., the current forecast error rate) is calculated to facilitate the judgment of whether the current weather forecast data needs to be corrected. If the error rate is within the threshold range, it means that the current weather forecast data does not need to be corrected and can be used directly for load forecasting. If the error rate is not within the threshold range, the current weather forecast data is corrected once, which reduces the load forecast error caused by inaccurate weather forecasts and improves the adaptability of load forecasting to weather changes. The prediction results are closer to reality; the temperature accumulation index of the load is calculated according to the changing characteristics of the load under the temperature accumulation effect, which can make the load prediction more in line with the actual situation, so as to facilitate the secondary correction of the current weather forecast data, and further reduce the load prediction error caused by temperature changes; by constructing a BP neural network prediction model, the BP neural network has the advantages of strong nonlinear mapping ability and self-learning ability, and can process a large amount of nonlinear data, and the load prediction is carried out on the secondary corrected current weather forecast data, so that the model can more efficiently fit the complex relationship between the load and the influencing factors, and realize more intelligent and automated load prediction, which improves the accuracy and stability of load prediction.

[0049] As a specific example analysis, load forecasting is performed for a target area in Zhejiang Province. First, a correlation analysis is performed on the meteorological data. By analyzing the correlation, the three meteorological factors with the highest correlation with the power load can be identified and corrected and analyzed. The data collection frequency is once every 15 minutes, and the number of single data sets is 23,328. The first 21,888 points are selected as the training set, and the following 96 points are selected as the test set. The weather forecast data comes from the website of the China Meteorological Administration. The main meteorological factors examined include key indicators such as temperature (degrees Celsius), humidity (percentage), rainfall (mm / hour), and wind speed (m / second). The results of the correlation analysis are shown in Table 2. Table 2 Correlation analysis between weather and load Temp Rh_2 Wspd_10 Wspd_30 0.91 -0.87 0.73 0.71 Ghi_sfc Wspd_50 P_sfc Rain_sfc 0.70 0.63 0.59 0.23 As can be seen from Table 2, the weather variables to be studied are determined to be temperature, humidity, 10-meter wind speed, 30-meter wind speed, and irradiance, and the temperature, humidity, and 10-meter wind speed are subjected to error correction. The R2 and RMSE error prediction evaluation indicators are used as the basis for judging the effectiveness of the method.

[0050] The simulation of weather forecast takes the current time as 11:00 am and the prediction of temperature and humidity for 12 to 15 hours as an example. The temperature and humidity for the next 4 hours are obtained from the Meteorological Bureau.

[0051] The weather forecast is set to start at 11:00 am on the same day, focusing on predicting the temperature and humidity for the next 12 to 14 hours. The weather forecast for the next 3 hours is provided by the meteorological agency. At 11:00 am, the local meteorological station reported a temperature of 33.787℃ and a humidity of 53.113%. The historical weather forecast data is corrected with the historical and real-time data of the local meteorological station, and the corrected historical total error is subtracted from the weather forecast data to obtain the corrected weather forecast data.

[0052] Subtract the corrected historical total error from the weather forecast data to get the actual forecast data. The comparison results between the corrected weather and the actual weather are shown in Table 3. The correlation analysis with the load is shown in Table 4. Table 3 Comparison results between predicted temperature and actual temperature time Corrected temperature / (℃) Actual temperature / (℃) Corrected humidity / (%) Actual humidity / (%) 12:00 33.312 34.475 54.396 48.840 13:00 34.562 35.662 47.490 43.184 14:00 34.475 35.975 48.697 42.123 Table 4 Correlation analysis between meteorological data and actual load before and after correction Temp Rh_2 Wspd_10 Before correction 0.91 -0.87 0.73 After correction 0.93 -0.92 0.77 It can be seen from Tables 3 and 4 that after an in-depth analysis of historical weather data and considering the deviation between the predicted value and the actual observed value on the same weather day, this embodiment effectively narrows the gap between the future weather forecast value and the actual value. This shows that by accurately correcting the meteorological data, the impact of weather forecast uncertainty can be compensated to a large extent.

[0053] According to the analysis of Zhejiang's regional characteristics and statistics, when the temperature exceeds 33°C, human comfort is affected and cooling loads such as air conditioners will start to start. When the temperature exceeds 38°C, most of these loads will be enabled, and the temperature accumulation effect reaches the saturation point and can be ignored.

[0054] Therefore, the study defined the temperature boundary range as 33℃~38℃. Through correlation analysis, it can be seen that the maximum accumulation period is 2 days. The cumulative effect index at 33℃~38℃ was determined by discretization method.

[0055] According to the temperature accumulation correction model, the three highest temperature days before the forecast date in August 2023 in a city in Zhejiang Province were corrected. The correction results are shown in Table 5. Table 5 Temperature accumulation model correction results date Actual load / MW Original temperature / ℃ Corrected temperature / ℃ 8 / 1413:00 13857 35.5 40.354 8 / 1513:00 12859 35.1 38.993 8 / 1613:00 13027 34.3 37.544 like Figure 2 As shown in the figure, it can be seen that the addition of temperature accumulation effect significantly improves the correlation between high temperature load and corrected temperature, and the correlation coefficient rises to 0.8518. Therefore, using the corrected temperature instead of the original temperature for load forecasting can improve the accuracy of load forecasting. When conducting the experiment, the above contents were organized into an experimental data set. The last day's data of the experimental data set was used as test data, and the first 180 days' data was used as training data.

[0056] In the comparative experiment, this embodiment uses the following methods for comparison: (1) using uncorrected meteorological data as input, which only contains historical weather forecast errors (correction method 1); (2) using weather data that contains both historical weather forecast errors and temperature accumulation effect corrections (correction method 2). The load forecast results of the two correction methods are compared with the actual load data. The load forecast result curves of the two correction methods are shown in Figure 2. Figure 3 The error evaluation indicators are shown in Table 6. Table 6 Error evaluation indicators before and after correction <![CDATA[R 2 ]]> RMSE Predicted value 0.96218 3.3601％ Correction method 1 0.96965 2.2816％ Correction Method 2 0.97157 2.1859％ according to Figure 3 As shown in Table 5, compared with the correction method 1 and the uncorrected meteorological data, the load curve generated by the correction method 2 is closer to the actual load curve. The error evaluation index of the correction method 2 is R2 = 0.97157, RMSE = 2.1859%. This shows that the accuracy has been significantly improved compared with the uncorrected forecast method and the simpler weather forecast error correction method.

[0057] In order to compare the effectiveness of different methods, this embodiment evaluates the long short-term memory (LSTM) network, the gated recurrent unit (GRU) network, the convolutional neural network (CNN) and the prediction method proposed in this embodiment. Figure 4 The load forecasting result curves of various weather forecasting methods are shown in Table 7. The corresponding evaluation indicators are given. Table 7 Comparison of evaluation indicators of different prediction models <![CDATA[R 2 ]]> RMSE GRU 0.96056 2.1189％ LSTM 0.96381 1.7213％ CNN 0.96418 2.2501％ Correction Method 2 0.97249 1.5126％ according to Figure 4 Table 6 and Table 6 show that the forecasting method (including the weather adjustment method) proposed in this embodiment effectively improves the correlation with the load data, performs best in all indicators, and has the lowest RMSE (1.5126%) and the highest R2 (0.97249). In contrast, the RMSE and R2 indicators of the model CNN are 0.7375% higher and 0.00831 lower than those of the model of this embodiment, respectively. The models GRU and LSTM perform relatively well in all indicators, but their RMSE values ​​increase by 0.006063% and 0.002087% and their R2 values ​​decrease by 0.01193 and 0.00868, respectively, relative to the model proposed in this embodiment. Therefore, in terms of prediction accuracy, the model proposed in this embodiment performs best.

[0058] In summary, the load forecasting method based on weather forecast correction and temperature accumulation index proposed in this embodiment can improve the overall accuracy of load forecasting, has high robustness, and can provide data support for power system operation and management.

[0059] The above specific embodiments are preferred embodiments of the present invention, and are not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to the present specific embodiments. All equivalent changes made in accordance with the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. A load forecasting method based on weather forecast correction and temperature accumulation index, characterized in that: The steps include: S1, obtaining the first forecast total error of historical weather forecast; S2. Calculating a current forecast error rate based on the current meteorological information and the first forecast total error, and correcting the meteorological forecast data once according to the error rate; S3. Calculate the temperature accumulation index based on the impact of the temperature accumulation effect on the load; S4, constructing a temperature correction model based on the temperature accumulation index, and performing secondary correction on the meteorological forecast data; S5. Construct a BP neural network prediction model and perform load prediction based on the secondary corrected meteorological forecast data.

2. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 1, characterized in that: The S1 comprises the following sub-steps: Acquire future weather information through weather forecasts, and collect historical weather data based on the future weather information, wherein the historical weather data includes historical measured weather data of weather stations and historical network weather forecast data within a corresponding period of time; Calculating the historical time error of the historical weather forecast according to the historical measured weather data and the historical network weather forecast data; The first total forecast error of the historical weather forecast is calculated based on the weight coefficient corresponding to the historical moment error.

3. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 2, characterized in that: The first total forecast error of the historical weather forecast is calculated based on the weight coefficient corresponding to the historical moment error, including: Based on the time distance between the historical moment error and the predicted day, a weighted average method is used to determine the weight coefficient corresponding to each historical moment error; The sum of the products of the historical moment errors and the corresponding weight coefficients is taken as the first forecast total error.

4. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 1, characterized in that: The S2 comprises the following sub-steps: Obtain the current weather forecast data and measured meteorological data of the target area, and calculate the current forecast error of the current weather forecast; The current forecast error rate of the current weather forecast is calculated by the ratio of the current forecast error to the first forecast total error at the corresponding historical moment; The current forecast error rate is compared with a threshold value, and the first forecast total error is proportionally converted according to the comparison result, and then the short-term meteorological forecast data of the target area is corrected to obtain meteorological forecast correction data.

5. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 4, characterized in that: The method of performing a correction on the short-term meteorological forecast data of the target area after proportionally converting the first forecast total error according to the comparison result to obtain a meteorological forecast correction data comprises: If the current forecast error rate is not within the threshold range, multiplying the first forecast total error by the current forecast error rate to proportionally convert the first forecast total error to obtain a second forecast total error of the historical weather forecast; The meteorological forecast data is subtracted from the second total forecast error to obtain primary meteorological forecast correction data, thereby completing a primary correction of the meteorological forecast data.

6. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 5, characterized in that: The temperature accumulation index includes a limit temperature index, a maximum accumulation days index and a cumulative effect index; wherein: By fitting the time variation curve of temperature and load, the limit temperature index of load variation is obtained; Calculate the correlation between the maximum temperature and the maximum load, and use the cumulative number of days with the maximum correlation as the maximum cumulative number of days index; The cumulative effect index is calculated based on the limit temperature index and the maximum cumulative days index.

7. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 6, characterized in that: The calculation of the correlation between the maximum temperature and the maximum load includes: Calculating the correlation degree based on the limit temperature index and the second total forecast error; The calculation formula of the correlation degree is as follows: In the formula, R represents the degree of correlation, T i represents the critical temperature index value of the i-th cumulative day, T represents the average critical temperature index value, L i represents the total error of the second forecast on the ith cumulative day, L represents the average value of the total error of the second forecast, and n represents the number of cumulative days.

8. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 6, characterized in that: The calculating the cumulative effect index based on the limit temperature index and the maximum cumulative days index comprises: establishing a cumulative effect coefficient of loads at different temperatures based on the limit temperature index and the maximum cumulative days index; Discretizing the cumulative effect coefficient to obtain a cumulative effect coefficient sequence; Calculating a corrected correlation coefficient between temperature and load based on the primary weather forecast correction data and the limit temperature index; It is determined whether the cumulative effect coefficient sequence meets the cumulative effect specification by using the modified correlation coefficient. If so, the cumulative effect index is determined based on the cumulative effect coefficient sequence.

9. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 4, characterized in that: The S4 comprises the following sub-steps: A temperature correction model is constructed based on the limit temperature index, the maximum cumulative days index and the cumulative effect index; the primary meteorological forecast correction data is calculated based on the temperature correction model to obtain secondary meteorological forecast correction data, thereby completing secondary correction of meteorological forecast data.

10. A load forecasting method based on weather forecast correction and temperature accumulation index according to claim 2 or 9, characterized in that: The S5 comprises the following sub-steps: Acquire the historical load data of the target area, use the historical load data and the historical meteorological data as inputs of a BP neural network model for model training, and construct a BP neural network prediction model; The secondary meteorological forecast correction data is predicted based on the BP neural network prediction model, and the load prediction value of the target area is output.

Images