Power system short-term load prediction method, device, equipment, medium and product

Abstract

The embodiment of the invention provides a power system short-term load prediction method and device, equipment, a medium and a product. The method comprises the following steps: acquiring historical power load data and historical meteorological parameter data in a historical time period and meteorological parameter data of a to-be-predicted day; according to the historical meteorological parameter data and the meteorological parameter data of the to-be-predicted day, determining a time period corresponding to a target similar day of each time period in the to-be-predicted day, and forming a historical day sample set based on the historical power load data and the historical meteorological parameter data of the time period corresponding to the target similar day; training an improved multi-layer feedforward neural network model based on the historical day sample set to obtain a trained target load prediction model; and inputting the meteorological parameter data of the to-be-predicted day into the target load prediction model to obtain a power load prediction value of each time period in the to-be-predicted day, and accumulating the power load prediction values of all time periods in the to-be-predicted day to obtain a power load prediction value of the to-be-predicted day. By using the method, the power load prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method, device, equipment, medium and product for short-term load forecasting of a power system. Background Art

[0002] With the development of smart grids, the economic dispatch, energy storage dispatch and power market transactions of power systems rely heavily on the accuracy of load forecasting. In order to reduce resource waste, accurate short-term load forecasting methods are particularly important for the balance between power supply and load demand. In the context of smart grids that obtain massive real-time meteorological data and power load data, meteorological factors, as one of the input data samples, are inherently uncertain and coupled, and the accuracy of power load forecasting depends largely on the regularity of the impact of meteorological factors on power load and the accuracy of meteorological forecasts.

[0003] At present, a large number of forecasting technologies use similar day theory to carry out load forecasting. By selecting historically similar power generation days with high similarity to the day to be predicted to train the forecasting model, the accuracy of power forecasting can be effectively improved. However, current research only selects relatively similar dates from the perspective of meteorological characteristics. However, this method ignores weather changes. For example, the weather changes in some areas during the day may show large changes. This will result in the best similar day selected being similar to the weather of the day to be predicted only in terms of statistical mean, while the weather may be completely different at different times of the day, which will result in lower prediction accuracy for this part of the time.

[0004] In addition, a lot of research has been conducted on power load forecasting of power grids, such as the Long Short-Term Memory (LSTM) model and the Least Squares Support Vector Machine (LSSVM) model. However, although the LSTM model can mine the information hidden in massive data, its training speed is slow; the LSSVM model has strong generalization performance, but the accuracy of the prediction results is greatly affected by the parameters; therefore, when using a single prediction model to predict power load, its prediction results are highly random and not universal. Summary of the invention

[0005] The present invention provides a method, device, equipment, medium and product for short-term load forecasting of an electric power system, which achieves the effect of improving the accuracy of electric power load forecasting.

[0006] In a first aspect, an embodiment of the present invention provides a method for short-term load forecasting of a power system, comprising:

[0007] Obtain historical power load data, historical meteorological parameter data within a historical time period, and meteorological parameter data for the day to be predicted;

[0008] According to the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, the target similar day corresponding time period of each time period in the day to be predicted is determined, and the historical day sample set is formed based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period;

[0009] Based on the historical daily sample set, the improved feedforward neural network model is trained to obtain a trained target load forecasting model;

[0010] The meteorological parameter data of the day to be predicted is input into the target load prediction model to obtain the power load prediction value of each time period in the day to be predicted, and the power load prediction values ​​of all time periods in the day to be predicted are accumulated to obtain the power load prediction value of the day to be predicted.

[0011] In a second aspect, an embodiment of the present invention provides a short-term load forecasting device for a power system, comprising:

[0012] The data acquisition module is used to acquire the historical power load data, historical meteorological parameter data and meteorological parameter data of the day to be predicted within the historical time period;

[0013] A sample set determination module, for determining the target similar day corresponding time period of each time period in the day to be predicted according to the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, and forming a historical day sample set based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period;

[0014] A model training module, used for training an improved feedforward neural network model based on the historical daily sample set to obtain a trained target load forecasting model;

[0015] The load forecasting module is used to input the meteorological parameter data of the day to be predicted into the target load forecasting model, obtain the power load forecast value of each time period in the day to be predicted, and accumulate the power load forecast values ​​of all time periods in the day to be predicted to obtain the power load forecast value of the day to be predicted.

[0016] The embodiment of the present invention provides a method, device, equipment, medium and product for short-term load forecasting of a power system, the method comprising: firstly obtaining historical power load data, historical meteorological parameter data and meteorological parameter data of a day to be predicted within a historical time period; secondly determining the target similar day corresponding time period of each time period in the day to be predicted according to the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, and forming a historical day sample set based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period; then training an improved feedforward neural network model based on the historical day sample set to obtain a trained target load forecasting model; finally inputting the meteorological parameter data of the day to be predicted into the target load forecasting model to obtain the power load forecast value of each time period in the day to be predicted, and accumulating the power load forecast values ​​of all time periods in the day to be predicted to obtain the power load forecast value of the day to be predicted. The above technical scheme can improve the prediction accuracy of the power load of the subsequent day to be predicted by selecting similar moments of historical days similar to the day to be predicted as subsequent prediction model training samples. In addition, the improved feedforward neural network model can enhance the global optimization capability of the algorithm, prevent the neural network from falling into a local optimal state, and enhance the load forecasting capability of the neural network, thereby accurately predicting the trend of short-term power load and having superiority in load forecasting problems.

[0017] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A schematic diagram of a flow chart of a method for short-term load forecasting of a power system provided in Embodiment 1 of the present invention;

[0020] Figure 2 A schematic diagram of a flow chart of a method for short-term load forecasting of a power system provided in Embodiment 2 of the present invention;

[0021] Figure 3 A schematic diagram of the structure of a short-term load forecasting device for a power system provided in Embodiment 3 of the present invention;

[0022] Figure 4A schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "original", "target", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] Embodiment 1

[0026] Figure 1 This is a flow chart of a method for short-term load forecasting of a power system provided in the first embodiment of the present invention. The method can be applied to the case of short-term load forecasting of a power system. The method can be executed by a short-term load forecasting device for a power system. The device can be implemented in the form of hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the power system short-term load forecasting method provided in this embodiment may specifically include the following steps:

[0027] S101. Obtain historical power load data, historical meteorological parameter data, and meteorological parameter data of a day to be predicted within a historical time period.

[0028] Considering that a large number of forecasting technologies use similar day theory to perform load forecasting, the accuracy of power forecasting can be effectively improved by selecting historically similar power generation days that are highly similar to the day to be predicted to train the forecasting model. However, current research only selects relatively similar dates from the perspective of meteorological characteristics. This method ignores weather changes. For example, the weather in some areas may change greatly during the day. This will result in the best similar day selected being similar to the weather of the day to be predicted only in terms of the statistical mean, while the weather may be completely different at different times of the day, which will result in lower prediction accuracy for this part of the time.

[0029] In order to solve the above problems, unlike the traditional similar day prediction method, after calculating the similar days, this embodiment divides the day to be predicted into several time periods. For example, twelve time periods can be used. The time period with the closest meteorological characteristics to the time period of the day to be predicted is selected from the similar days for subsequent prediction, which can obtain more accurate time period power load prediction data compared to the similar day prediction.

[0030] In this embodiment, it is necessary to first obtain some basic data, that is, collect historical power load data, historical meteorological parameter data, and meteorological parameter data of the day to be predicted, and pre-process the historical power load data. It should be noted that the load data in the power system is usually relatively smooth before and after, but when there are unexpected impacts such as equipment failure in the power system, it will cause the recorded data to be greatly distorted. When the data at a certain moment is significantly different from the data at adjacent moments, the data at that moment can be considered to be abnormal data. And the abnormal data needs to be corrected. In actual application, the correction can be made by the method of the before and after mean or the method of polynomial fitting, which will not be repeated here.

[0031] S102. Determine the target similar day corresponding time period in each time period of the day to be predicted based on the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, and form a historical day sample set based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period.

[0032] In this embodiment, in order to better predict the power load of the forecast day based on the meteorological parameter data, it can be divided and combined based on season and weather. Specifically, according to different season types: spring, summer, autumn, winter, and different weather types: sunny, cloudy, rainy, and snowy, each subdivided type is combined to divide into 16 groups of data sets. After the season and weather are divided and combined, based on the historical meteorological parameter data and the meteorological parameter data of the forecast day, similar historical days with similar meteorological characteristics to the forecast day are screened.

[0033] It should be noted that although the power load is affected by temperature, humidity, weather and other conditions and shows obvious fluctuations, it is actually a curve that changes over time, that is, the power load is data with time series properties. Moreover, not only is there a strong correlation between the power load and the load at the adjacent time, but the sequence of power load at a certain time is also related to the load sequence at similar times in the same period, such as the same time of the previous day, the same time of the previous few days, etc.

[0034] Therefore, in order to solve the problem of selecting relatively similar dates from the perspective of meteorological characteristics, but this method ignores the changes in weather, and the weather may be completely different in different periods of a day, which leads to low prediction accuracy during this part of time. After selecting similar days, the moments of similar days are also selected. By selecting similar moments of historical days similar to the day to be predicted as subsequent prediction model training samples, the subsequent prediction accuracy can be improved, and the time-sharing prediction of power load can be realized.

[0035] In order to solve the above problem, in this embodiment, the corresponding time periods similar to the meteorological characteristics of each time period of the day to be predicted are selected from similar historical days, that is, the meteorological characteristics of a certain time period of the day to be predicted and the meteorological characteristics of each time period of the similar historical days are calculated to select, so as to determine the corresponding time period in the similar historical days that is most similar to the meteorological characteristics of each time period of the day to be predicted. In this embodiment, the historical day with the most similar meteorological characteristics to the day to be predicted is recorded as the target historical day. The corresponding historical time period in the target historical day with the most similar meteorological characteristics to each time period of the day to be predicted is recorded as the corresponding time period.

[0036] Following the above description, after determining the target similar day corresponding time period of each time period in the forecasted day, the historical day sample set is formed based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period for subsequent forecast model training. Among them, the historical meteorological parameter data can be used as the input of the model, and the historical power load data is the reference value of the output of the model.

[0037] S103. Train an improved feedforward neural network model based on the historical daily sample set to obtain a trained target load forecasting model.

[0038] In this embodiment, the feedforward (BP) neural network, as the basic model of the artificial neural network, has strong self-learning, self-adaptation, generalization and fault tolerance capabilities, but it also has defects such as slow overall network operation speed, easy to fall into the local optimal solution, and difficulty in adjusting the network to the optimal weight threshold when the data volume is large. Therefore, it is necessary to improve the BP neural network so that the BP neural network can be better applied to power load forecasting with large-scale data.

[0039] The improved BP neural network model is as follows: set the number of network input and output neurons, the number of hidden layers, the number of training times t, the learning rate, the gradient descent algorithm and the number of samples for each training; determine the number of parameters m in the recurrent neural network, encode the parameters in sequence and set the parameter vector to be λ = (λ1,λ2,...,λ m ); initialize the ant population, set the number of ants to n, the dimension of the ant position vector to m, and set the ant population to be, i = 1, 2, ..., n; update the ant position, and when calculating the fitness function of the i-th ant or the optimal ant, assign the j-th parameter values ​​of the ant to the j-th weight in the neural network, that is, j = 1, 2, ..., m.

[0040] Following the above description, after the weight assignment is completed, the historical daily sample set is input into the neural network, and the load forecast value is given through the feedforward process of the recurrent neural network, and the fitness function of the ant is calculated; the position is iteratively updated until the fitness function converges, and the optimal ant position at the last iteration is recorded; the j parameters of the optimal ant at the last iteration are respectively assigned to the jth weight in the neural network to obtain the initial weight of the neural network; the recurrent neural network is trained using the gradient descent algorithm in the neural network, and after training to a preset number of times, if the loss function converges, the neural network training is completed; if the loss function does not converge, the number of training times is increased until the loss function converges.

[0041] S104, inputting the meteorological parameter data of the day to be predicted into the target load prediction model, obtaining the power load prediction value of each time period in the day to be predicted, and accumulating the power load prediction values ​​of all time periods in the day to be predicted to obtain the power load prediction value of the day to be predicted.

[0042] In this embodiment, the meteorological parameter data of each time period of the day to be predicted is input into the improved BP neural network model to obtain the predicted value of the power load of each time period of the day to be predicted, and finally all time periods are accumulated to obtain the predicted value of the power load of the day to be predicted.

[0043] The above technical solution can improve the prediction accuracy of the power load on the subsequent days to be predicted by selecting similar moments of historical days similar to the day to be predicted as subsequent prediction model training samples. In addition, the improved feedforward neural network model can improve the global optimization ability of the algorithm, prevent the neural network from falling into a local optimal state, improve the load prediction ability of the neural network, and then accurately predict the trend of short-term power load, which is superior in load prediction problems.

[0044] Embodiment 2

[0045] Figure 2A flow chart of another method for short-term load forecasting of an electric power system provided in Embodiment 2 of the present invention. This embodiment is a further optimization of the above-mentioned embodiment. In this embodiment, "determining the target similar day corresponding time period of each time period in the day to be predicted based on the historical meteorological parameter data and the meteorological parameter data of the day to be predicted" is further limited and optimized, and "obtaining a trained target load forecasting model based on the improved feedforward neural network model trained on the historical day sample set" is further limited and optimized.

[0046] like Figure 2 As shown, this embodiment 2 provides a method for short-term load forecasting of a power system, which specifically includes the following steps:

[0047] S201. Obtain historical power load data, historical meteorological parameter data, and meteorological parameter data of a day to be predicted within a historical time period.

[0048] S202: Based on the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, select a target similar day having meteorological characteristics similar to the day to be predicted.

[0049] In this embodiment, in order to better predict the power load of the forecast day based on meteorological data, the data are divided and combined based on seasons and weather. Specifically, according to different season types: spring, summer, autumn, and winter, and different weather types: sunny, cloudy, rainy, and snowy, each subdivided type is combined to divide into sixteen data sets. After the seasons and weather are divided and combined, based on the historical meteorological parameter data and the meteorological parameter data of the forecast day, similar historical days with similar meteorological characteristics to the forecast day are selected and recorded as target similar days with similar meteorological characteristics to the forecast day.

[0050] As a specific implementation method, the steps of selecting a target similar day with similar meteorological characteristics to the day to be predicted based on historical meteorological parameter data and meteorological parameter data of the day to be predicted can be optimized, including:

[0051] a1) Obtain the target seasonal weather combination type to which the day to be predicted belongs, and determine all historical days under the target seasonal weather combination type as candidate similar days.

[0052] In this embodiment, in order to better predict the power load of the day to be predicted based on meteorological data, division and combination are performed based on season and weather. Specifically, according to different season types: spring, summer, autumn, winter, and different weather types: sunny, cloudy, rainy, and snowy, each subdivided type is combined to form a data set of sixteen seasonal weather combination types. After the season and weather are divided and combined, the meteorological information of the day to be predicted is obtained from the weather forecast, and the specific type of the day to be predicted in the sixteen data sets is determined according to the season and weather type of the day to be predicted. In this embodiment, the specific type is recorded as the target seasonal weather combination type. After determining the target seasonal weather combination type to which the day to be predicted belongs, all historical days in the data set under the specific type are determined as candidate similar days.

[0053] b1) constructing features of the historical meteorological parameter data of the candidate similar days and the meteorological parameter data of the day to be predicted according to the set meteorological factors, and obtaining the meteorological feature vector of the historical day and the meteorological feature vector of the day to be predicted.

[0054] The set meteorological factors at least include average temperature, maximum temperature, minimum temperature, average humidity, maximum humidity, average rainfall, rainfall duration, average snowfall, and snowfall duration. Specifically, the average temperature, maximum temperature, minimum temperature, average humidity, maximum humidity, average rainfall, rainfall duration, average snowfall, and snowfall duration are selected to construct a daily meteorological feature vector, and the meteorological feature vector corresponding to the historical meteorological parameter data of the candidate similar day is recorded as the historical day meteorological feature vector, and the meteorological feature vector of the day to be predicted is recorded as the meteorological feature vector of the day to be predicted.

[0055] c1) Normalize the historical daily meteorological feature vector and the meteorological feature vector of the day to be predicted respectively.

[0056] Specifically, the historical daily meteorological feature vectors and the meteorological feature vectors of the day to be predicted are normalized respectively. Normalization can make subsequent model training more efficient. By adjusting the scale of the features to a similar range, the model can converge to the optimal solution faster during the training process.

[0057] d1) Traverse each candidate similar day and determine the weighted similarity between the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted.

[0058] In this embodiment, each candidate similar day is traversed, and the weighted similarity between the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted is calculated.

[0059] Furthermore, each candidate similar day is traversed to determine the weighted similarity between the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted, including:

[0060] d11) For each candidate similar day, based on the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted, combined with a preset correlation determination formula, determine the correlation between the meteorological feature vectors of the historical day and the meteorological feature vector of the day to be predicted.

[0061] Specifically, for each candidate similar day, the correlation between its historical day meteorological feature vector and the meteorological feature vector of the day to be predicted is calculated. The correlation determination formula is expressed as:

[0062]

[0063] Among them, ψ i (k) is the correlation between the historical daily meteorological feature vector and the meteorological feature vector of the day to be predicted, X p (k) is the component of the kth meteorological characteristic vector of the day to be predicted, X i (k) is the kth meteorological characteristic vector component of the historical day, i represents the i-th day, and ρ is a coefficient ranging from 0 to 1, which can be selected as 0.5.

[0064] d12) Accumulating the correlation degree of each meteorological factor in the correlation degree to obtain the first similarity between the candidate similar day and the day to be predicted.

[0065] In this embodiment, after determining the correlation between the meteorological feature vectors of the historical days and the meteorological feature vectors of the day to be predicted, the correlation between the meteorological feature vectors of the day to be predicted and the historical days is accumulated, and the similarity between the meteorological factors of the i-th historical day and the day to be predicted is recorded as the first similarity.

[0066] The first similarity formula can be expressed as: Among them, ξ 1-i is the first similarity between the candidate similar day and the day to be predicted, ψ i (k) is the correlation between the historical daily meteorological feature vector of the kth meteorological factor on the i-th day and the meteorological feature vector of the day to be predicted, and 9 is the type of meteorological factor, namely, average temperature, maximum temperature, minimum temperature, average humidity, maximum humidity, average rainfall, rainfall duration, average snowfall, and snowfall duration.

[0067] d13) determining the second similarity between the candidate similar day and the day to be predicted based on the meteorological feature vectors of the historical day and the meteorological feature vectors of the day to be predicted in combination with a preset second similarity determination formula.

[0068] This step is used to calculate the second similarity ξ between the historical day meteorological feature vector and the day meteorological feature vector to be predicted. 2-i ; The second similarity determination formula is expressed as:

[0069]

[0070] Among them, X p (k) is the component of the kth meteorological characteristic vector of the day to be predicted, X i (k) is the kth meteorological characteristic vector component of the i-th historical day.

[0071] d14) performing weighted processing on the first similarity and the second similarity to obtain a normalized weighted similarity between the meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted.

[0072] In this embodiment, the first similarity and the second similarity are used to calculate the weighted similarity ξ i ,ξ i =σξ 1-i +(1-σ)ξ 2-i , where σ is a weight coefficient ranging from 0 to 1, which is selected in combination with the specific weather type of the day to be predicted. When the temperature, humidity, rainfall, etc. fluctuate violently, that is, the weather type changes significantly, σ can be taken close to 0. When the temperature, humidity, rainfall, etc. fluctuate slightly, that is, the weather type basically does not change, σ can be taken close to 1.

[0073] e1) The candidate similar days whose weighted similarity is greater than a set similarity threshold are arranged in descending order of weighted similarity, and the first set number of candidate similar days with the highest weighted similarity are selected as target similar days.

[0074] In this embodiment, the similarity threshold can be set according to the actual situation. The historical days with weighted similarity greater than the preset threshold are arranged in descending order of weighted similarity, and the historical days with the highest weighted similarity in the previous several days are selected as the target similar days. Exemplarily, the historical days with the highest weighted similarity in the previous fifteen days are selected as the target similar days, and the sample set of the target similar days is used as the training data set of the BP neural network.

[0075] It should be noted that, in the process of determining the above-mentioned target similar days, if there are some days with the same weighted similarity, the historical day closest to the predicted day will be selected; if the historical days with a weighted similarity greater than the preset threshold are less than fifteen days, then in addition to incorporating all the historical days with a weighted similarity greater than the preset threshold into the target similar days, the missing days will be supplemented in the remaining candidate similar days in descending order of weighted similarity.

[0076] The above technical solution, by first calculating the first similarity and the second similarity, and then calculating the weighted similarity using the first similarity and the second similarity, can select historical days from historical days that have strong similarity with the meteorological factors of the day to be predicted, thereby improving the accuracy of subsequent model training samples.

[0077] S203, selecting corresponding time periods having meteorological characteristics similar to those of each time period of the day to be predicted from the target similar days.

[0078] This step is used to select corresponding time periods from the target similar days that have meteorological characteristics similar to those of the time periods of the day to be predicted, that is, the selection is made by calculating the correlation coefficient between the meteorological characteristics of a certain time period of the day to be predicted and the meteorological characteristics of each time period of the target similar day.

[0079] As a specific implementation method, the step of selecting corresponding time periods having similar meteorological characteristics to each time period of the day to be predicted from the target similar day can be optimized, including:

[0080] a2) For each time period of the day to be predicted, the correlation coefficient between the meteorological characteristic vector of each historical time period and the meteorological characteristic vector of the time period is determined based on the meteorological characteristic vector of the time period and the historical meteorological characteristic vector of each historical time period in the target similar day, combined with a preset coefficient determination formula.

[0081] This step is used to calculate the correlation coefficient between the meteorological characteristic vector of the time period of the day to be predicted and the meteorological characteristic vector of the corresponding time period of the historical day in the target similar day. The coefficient determination formula can be expressed as:

[0082]

[0083] Where I is the correlation coefficient between the meteorological feature vector X of the time period t of the day to be predicted and the meteorological feature vector Y of the corresponding time period of the historical day in the second similar day sample set, M is the sum of the number of all values ​​of the meteorological feature vector X and the meteorological feature vector Y, N i is the number of intervals of the meteorological characteristic vector X, M i is the number of values ​​of the meteorological characteristic vector Y falling in the i-th interval, N j is the number of intervals of the meteorological characteristic vector Y, P(y u ) is the probability that the meteorological characteristic vector Y falls in the uth interval, M uv It is the number of values ​​of meteorological characteristic vector X falling in the vth interval when meteorological characteristic vector Y falls in the uth interval.

[0084] b2) The historical time period corresponding to the maximum value is selected from each correlation coefficient as the corresponding time period with similar meteorological characteristics to the time period.

[0085] Through the above calculation, the correlation coefficient between the meteorological characteristics of a certain period of the day to be predicted and the meteorological characteristics of each period of the target similar day can be obtained, and the time period of the historical day with the most similar meteorological characteristics to the target time period of the day to be predicted can be selected from the target similar day.

[0086] S204. Construct a historical day sample set based on the historical power load data and historical meteorological parameter data of the corresponding time period of the target similar day.

[0087] Specifically, historical power load data and historical meteorological parameter data of a time period corresponding to a target similar day are obtained as historical day samples for training, and each historical day sample constitutes a historical day sample set.

[0088] S205, initializing the configuration of the improved feedforward neural network model.

[0089] The configured information includes at least: the number of network input and output neurons, the number of hidden layers, the number of training times, the learning rate, the gradient descent algorithm and the number of samples for each training, the number of parameters in the recurrent neural network, encoding the parameters in sequence into a parameter vector, initializing the ant population, updating the ant positions, and calculating the fitness function of the i-th ant or the optimal ant, respectively assigning the j-th weight in the neural network to the j-th parameter values ​​of the ant, where i and j are both positive integers.

[0090] It should be noted that, as the basic model of artificial neural network, BP neural network has strong self-learning, self-adaptation, generalization and fault tolerance capabilities, but it also has defects such as slow overall network operation speed, easy to fall into local optimal solution, and difficulty in adjusting the network to the optimal weight threshold when the data volume is large. Therefore, it is necessary to improve the BP neural network so that the BP neural network can be better applied to power load forecasting with large-scale data.

[0091] In this embodiment, the number of network input and output neurons, the number of hidden layers, the number of training times t, the learning rate, the gradient descent algorithm and the number of samples for each training are set; the number of parameters m in the recurrent neural network is determined, and the parameters are encoded in sequence. Let the parameter vector be λ = (λ1, λ2, ..., λ m );

[0092] Initialize the ant population, set the number of ants to n, the dimension of the ant position vector to m, and the ant population to A i (t) = [A i,1 (t),A i,2 (t),...,A i,m (t)], i = 1, 2, ..., n; when updating the ant position and calculating the fitness function of the i-th ant or the best ant, the j-th parameter value of the ant is assigned to the j-th weight in the neural network, that is, λ j =A i,j (t), j = 1, 2,…, m.

[0093] S206. Input the historical meteorological parameter data in the historical daily sample set into the improved feedforward neural network model, obtain the output load forecast value through the feedforward process of the recurrent neural network, and calculate the fitness function of the ant.

[0094] In this embodiment, after the weight assignment is completed, the historical meteorological parameter data in the historical daily sample set is input into the neural network for training. Through the feedforward process of the recurrent neural network, the load prediction value is given as the output load prediction value, and the fitness function of the ant is calculated.

[0095] S207, iteratively update the position until the fitness function converges, and record the optimal ant position at the last iteration.

[0096] Specifically, iteratively update the position until the fitness function converges, and record the optimal ant position at the last iteration

[0097] S208, assigning the optimal ant j parameters in the last iteration to the jth weight in the neural network respectively, to obtain the initial weight of the neural network.

[0098] In this embodiment, the optimal ant j parameters in the last iteration are assigned to the jth weight in the neural network to obtain the initial weight of the neural network.

[0099] S209, using the gradient descent algorithm in the neural network to train the improved feedforward neural network model until the loss function converges, thereby obtaining a trained target load prediction model.

[0100] In this embodiment, the gradient descent algorithm in the neural network is used to train the recurrent neural network. After training to a preset number of times, if the loss function converges, the neural network training is completed; if the loss function does not converge, the number of training is increased until the loss function converges. It can be understood that the historical power load data and the output load forecast value can be used to calculate the loss function value, thereby determining whether the loss function converges. The trained prediction model is recorded as the target load prediction model.

[0101] S210, input the meteorological parameter data of the day to be predicted into the target load prediction model to obtain the power load prediction value of each time period in the day to be predicted, and accumulate the power load prediction values ​​of all time periods in the day to be predicted to obtain the power load prediction value of the day to be predicted.

[0102] The above technical solution specifies the steps of determining the target similar day corresponding to each time period in the day to be predicted, as well as the training steps of the target load forecasting model. After calculating the similar days, the day to be predicted is divided into several time periods, and the time periods with the closest meteorological characteristics to the time period of the day to be predicted are selected from the similar days for subsequent predictions, which can obtain more accurate time period power load forecasting data than similar day predictions. In addition, the improved BP neural network model used can improve the global optimization ability of the algorithm, prevent the neural network from falling into a local optimal state, improve the load forecasting ability of the neural network, and then accurately predict the trend of short-term power load, which is superior in load forecasting problems.

[0103] Embodiment 3

[0104] Figure 3 The present invention provides a schematic diagram of a short-term load forecasting device for a power system according to a third embodiment of the present invention. The device can be used for short-term load forecasting of a power system. The short-term load forecasting device for a power system can be configured in an electronic device, such as Figure 3 As shown, the device includes: a data acquisition module 31, a sample set determination module 32, a model training module 33 and a load prediction module 34; wherein,

[0105] The data acquisition module 31 is used to acquire historical power load data, historical meteorological parameter data and meteorological parameter data of the day to be predicted within a historical period;

[0106] The sample set determination module 32 is used to determine the target similar day corresponding time period of each time period in the day to be predicted based on the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, and to form a historical day sample set based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period;

[0107] A model training module 33 is used to train an improved feedforward neural network model based on a historical daily sample set to obtain a trained target load forecasting model;

[0108] The load forecasting module 34 is used to input the meteorological parameter data of the day to be predicted into the target load forecasting model, obtain the power load forecast value of each time period in the day to be predicted, and accumulate the power load forecast values ​​of all time periods in the day to be predicted to obtain the power load forecast value of the day to be predicted.

[0109] The above technical solution can improve the prediction accuracy of the power load on the subsequent days to be predicted by selecting similar moments of historical days similar to the day to be predicted as subsequent prediction model training samples. In addition, the improved feedforward neural network model can improve the global optimization ability of the algorithm, prevent the neural network from falling into a local optimal state, improve the load prediction ability of the neural network, and then accurately predict the trend of short-term power load, which is superior in load prediction problems.

[0110] Optionally, the sample set determination module 32 includes:

[0111] A similar day determination unit, used for selecting a target similar day having similar meteorological characteristics to the day to be predicted based on historical meteorological parameter data and meteorological parameter data of the day to be predicted;

[0112] The time period determination unit is used to select corresponding time periods with similar meteorological characteristics to those of each time period of the day to be predicted from the target similar days.

[0113] Optionally, the similar day determination unit is specifically configured to:

[0114] Obtain the target seasonal weather combination type to which the day to be predicted belongs, and determine all historical days under the target seasonal weather combination type as candidate similar days;

[0115] The historical meteorological parameter data of the candidate similar days and the meteorological parameter data of the day to be predicted are respectively constructed according to the set meteorological factors to obtain the meteorological feature vector of the historical day and the meteorological feature vector of the day to be predicted;

[0116] Normalize the historical day meteorological feature vector and the day meteorological feature vector to be predicted respectively;

[0117] Traverse each candidate similar day and determine the weighted similarity between the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted;

[0118] The candidate similar days whose weighted similarity is greater than the set similarity threshold are arranged in descending order of weighted similarity, and the first set number of candidate similar days with the highest weighted similarity are selected as target similar days.

[0119] Optionally, the similar day determination unit is used to perform the step of traversing each candidate similar day and determining the weighted similarity between the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted, including:

[0120] For each candidate similar day, the correlation between the meteorological feature vectors of the historical days and the meteorological feature vectors of the day to be predicted is determined based on the normalized meteorological feature vectors of the historical days and the meteorological feature vectors of the day to be predicted, in combination with a preset correlation determination formula;

[0121] Accumulate the correlation degree of each meteorological factor in the correlation degree to obtain the first similarity between the candidate similar day and the day to be predicted;

[0122] According to the meteorological feature vectors of the historical days and the meteorological feature vectors of the day to be predicted, combined with a preset second similarity determination formula, the second similarity between the candidate similar day and the day to be predicted is determined;

[0123] The first similarity and the second similarity are weighted to obtain the normalized weighted similarity between the meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted.

[0124] Optionally, the time period determination unit is specifically configured to:

[0125] For each time period of the day to be predicted, the correlation coefficient between the meteorological characteristic vector of each historical time period and the meteorological characteristic vector of the time period is determined based on the meteorological characteristic vector of the time period and the historical meteorological characteristic vector of each historical time period on the target similar day, combined with a preset coefficient determination formula;

[0126] The historical period corresponding to the maximum value is selected from each correlation coefficient as the corresponding period with similar meteorological characteristics of the period.

[0127] Optionally, the model training module 33 is specifically used to:

[0128] Initialize and configure the improved feedforward neural network model, where the configured information at least includes: the number of network input and output neurons, the number of hidden layers, the number of training times, the learning rate, the gradient descent algorithm and the number of samples for each training, the number of parameters in the recurrent neural network, encode the parameters in sequence and set the parameter vector as , initialize the ant population, update the ant positions, and when calculating the fitness function of the i-th ant or the optimal ant, assign the j-th weight in the neural network to the j-th parameter values ​​of the ant, where i and j are both positive integers;

[0129] The historical meteorological parameter data in the historical daily sample set is input into the improved feedforward neural network model, and the output load forecast value is obtained through the feedforward process of the recurrent neural network, and the fitness function of the ants is calculated;

[0130] Iterate and update the position until the fitness function converges, and record the optimal ant position at the last iteration;

[0131] Assign the optimal ant j parameters in the last iteration to the jth weight in the neural network to obtain the initial weight of the neural network;

[0132] The improved feedforward neural network model is trained using the gradient descent algorithm in the neural network until the loss function converges, and then a trained target load forecasting model is obtained.

[0133] The power system short-term load forecasting system provided by the embodiment of the present invention can execute the power system short-term load forecasting method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0134] Embodiment 4

[0135] Figure 4 A schematic diagram of the structure of an electronic device provided for Embodiment 4 of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0136] like Figure 4 As shown, the electronic device 40 includes at least one processor 41, and a memory connected to the at least one processor 41, such as a read-only memory (ROM) 42, a random access memory (RAM) 43, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 41 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 42 or the computer program loaded from the storage unit 48 to the random access memory (RAM) 43. In the RAM 43, various programs and data required for the operation of the electronic device 40 can also be stored. The processor 41, the ROM 42, and the RAM 43 are connected to each other through a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0137] A number of components in the electronic device 40 are connected to the I / O interface 45, including: an input unit 46, such as a keyboard, a mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a disk, an optical disk, etc.; and a communication unit 49, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 49 allows the electronic device 40 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0138] The processor 41 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 41 executes the various methods and processes described above, such as a power system short-term load forecasting method.

[0139] In some embodiments, the power system short-term load forecasting method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 48. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 40 via the ROM 42 and / or the communication unit 49. When the computer program is loaded into the RAM 43 and executed by the processor 41, one or more steps of the power system short-term load forecasting method described above may be performed. Alternatively, in other embodiments, the processor 41 may be configured to execute the power system short-term load forecasting method in any other appropriate manner (e.g., by means of firmware).

[0140] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0141] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0142] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0143] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0144] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0145] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0146] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the method for short-term load forecasting of a power system provided by any embodiment of the present invention.

[0147] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code for the operation of the present disclosure, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0148] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0149] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for short-term load forecasting of a power system, characterized in that: include: Obtain historical power load data, historical meteorological parameter data within a historical time period, and meteorological parameter data for the day to be predicted; According to the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, the target similar day corresponding time period of each time period in the day to be predicted is determined, and the historical day sample set is formed based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period; Based on the historical daily sample set, the improved feedforward neural network model is trained to obtain a trained target load forecasting model; The meteorological parameter data of the day to be predicted is input into the target load prediction model to obtain the power load prediction value of each time period in the day to be predicted, and the power load prediction values ​​of all time periods in the day to be predicted are accumulated to obtain the power load prediction value of the day to be predicted.

2. The method according to claim 1, characterized in that The determining, based on the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, the target similar day corresponding time period of each time period in the day to be predicted comprises: Based on the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, selecting a target similar day having meteorological characteristics similar to the day to be predicted; Select corresponding time periods having meteorological characteristics similar to those of each time period of the day to be predicted from the target similar days.

3. The method according to claim 2, characterized in that The step of selecting a target similar day having meteorological characteristics similar to the day to be predicted based on the historical meteorological parameter data and the meteorological parameter data of the day to be predicted includes: Obtaining the target seasonal weather combination type to which the day to be predicted belongs, and determining all historical days under the target seasonal weather combination type as candidate similar days; Constructing features of the historical meteorological parameter data of the candidate similar day and the meteorological parameter data of the day to be predicted according to the set meteorological factors to obtain the meteorological feature vector of the historical day and the meteorological feature vector of the day to be predicted; Normalizing the historical day meteorological feature vector and the to-be-forecast day meteorological feature vector respectively; Traversing each of the candidate similar days, and determining the weighted similarity between the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted; The candidate similar days whose weighted similarity is greater than a set similarity threshold are arranged in descending order according to the weighted similarity, and the first set number of candidate similar days with the highest weighted similarity are selected as target similar days.

4. The method according to claim 3, characterized in that: The traversing of the candidate similar days to determine the weighted similarity between the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted includes: For each candidate similar day, according to the normalized meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted, in combination with a preset correlation determination formula, the correlation between the meteorological feature vector of the historical day and the meteorological feature vector of the day to be predicted is determined; Accumulating the correlation degree of each meteorological factor in the correlation degree to obtain a first similarity between the candidate similar day and the day to be predicted; Determine the second similarity between the candidate similar day and the day to be predicted based on the historical day meteorological feature vector and the day to be predicted meteorological feature vector, in combination with a preset second similarity determination formula; The first similarity and the second similarity are weighted to obtain a normalized weighted similarity between the meteorological feature vectors of each historical day and the meteorological feature vector of the day to be predicted.

5. The method according to claim 3, characterized in that: The step of selecting corresponding time periods having similar meteorological characteristics to those of each time period of the day to be predicted from the target similar day includes: For each time period of the day to be predicted, the correlation coefficient between the meteorological characteristic vectors of each historical time period and the meteorological characteristic vector of the time period is determined based on the meteorological characteristic vector of the time period and the historical meteorological characteristic vectors of each historical time period in the target similar day, in combination with a preset coefficient determination formula; The historical time period corresponding to the maximum value is selected from each of the correlation coefficients as the corresponding time period having meteorological characteristics similar to the time period.

6. The method according to claim 1, characterized in that The improved feedforward neural network model trained based on the historical daily sample set is used to obtain a trained target load forecasting model, including: Initialize and configure the improved feedforward neural network model, where the configured information at least includes: the number of network input and output neurons, the number of hidden layers, the number of training times, the learning rate, the gradient descent algorithm and the number of samples for each training, the number of parameters in the recurrent neural network, encode the parameters in order as the parameter vector, initialize the ant population, update the ant positions, and when calculating the fitness function of the i-th ant or the optimal ant, assign the j-th weight in the neural network to the j-th parameter values ​​of the ant, where i and j are both positive integers; Inputting the historical meteorological parameter data in the historical daily sample set into the improved feedforward neural network model, obtaining the output load forecast value through the feedforward process of the recurrent neural network, and calculating the fitness function of the ants; Iteratively update the position until the fitness function converges, and record the optimal ant position at the last iteration; Assign the optimal ant j parameters in the last iteration to the jth weight in the neural network to obtain the initial weight of the neural network; The improved feedforward neural network model is trained by using a gradient descent algorithm in a neural network until the loss function converges, thereby obtaining a trained target load forecasting model.

7. A short-term load forecasting device for a power system, characterized in that: include: The data acquisition module is used to acquire the historical power load data, historical meteorological parameter data and meteorological parameter data of the day to be predicted within the historical time period; A sample set determination module, for determining the target similar day corresponding time period of each time period in the day to be predicted according to the historical meteorological parameter data and the meteorological parameter data of the day to be predicted, and forming a historical day sample set based on the historical power load data and historical meteorological parameter data of the target similar day corresponding time period; A model training module, used for training an improved feedforward neural network model based on the historical daily sample set to obtain a trained target load forecasting model; The load forecasting module is used to input the meteorological parameter data of the day to be predicted into the target load forecasting model, obtain the power load forecast value of each time period in the day to be predicted, and accumulate the power load forecast values ​​of all time periods in the day to be predicted to obtain the power load forecast value of the day to be predicted.

8. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for short-term load forecasting of a power system according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for short-term load forecasting of a power system according to any one of claims 1 to 6 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements the method for short-term load forecasting of a power system according to any one of claims 1 to 6.

Images