Deep learning-based power distribution network load prediction method and system

Abstract

Disclosed in the present invention is a deep learning-based power distribution network load prediction method, comprising: acquiring regional load data and renewable energy power generation data to form a data set, and preprocessing the data set to obtain a first data set; using a convolutional neural network to extract a time series feature in the first data set, and converting the time series feature into a data form of a deep learning model by means of an embedding layer to obtain one-dimensional time series data; performing fast Fourier transform on the one-dimensional time series data to obtain a frequency curve, extracting amplitude values on the frequency curve to calculate corresponding periods, and selecting the corresponding periods to slice the one-dimensional time series data and form same into two-dimensional matrixes; using a two-dimensional convolutional network to perform feature extraction, reshaping the two-dimensional matrixes that have undergone feature extraction into one-dimensional arrays, and performing adaptive fusion on the one-dimensional arrays to obtain a first time series feature; and inputting the first time series feature into a fully connected layer for weight calculation and linear transformation to obtain a power distribution network load prediction result, thereby improving the stability and reliability of electric power supply.

Description

A power distribution network load prediction method and system based on deep learning TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution networks, in particular to a power distribution network load prediction method and system based on deep learning. BACKGROUND

[0002] As an important part of the power system, the reliability and efficiency of the power distribution network directly affect the stable supply of electricity. With the access of new energy and the opening of the electricity market, traditional load prediction methods have been difficult to meet the needs of modern power distribution networks. The introduction of wind and photovoltaic energy provides a new way for the optimal operation of power distribution networks, but also increases the complexity of load prediction.

[0003] Traditional load prediction methods, such as time series analysis and regression analysis based on statistics, often fail to fully utilize feature data related to power distribution network load (such as weather data, user behavior data, and distributed energy data), resulting in low prediction accuracy. In addition, these methods are prone to overfitting or underfitting when dealing with long time spans and multi-scale data, making it difficult to accurately capture the dynamic characteristics of load changes. In order to solve these problems, in recent years, deep learning methods have been introduced into the field of load prediction. In particular, deep learning models based on the Transformer architecture have become a hot topic in load prediction research due to their excellent performance in sequence data processing. Among them, the TimesNet network has unique advantages: strong long-time sequence modeling and multi-scale feature extraction capabilities, and good robustness, showing great potential in power distribution network load prediction. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] Therefore, the technical problem solved by the present application is that the present application solves the limitations of existing power distribution network load prediction methods, and the modeling of flexible resource influencing factors is insufficient, and the multi-scale feature extraction is insufficient.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a power distribution network load prediction method based on deep learning, comprising:

[0008] Obtaining regional load data and renewable energy generation data, grouping the regional load data and renewable energy generation data into a data set, preprocessing the data set to obtain a first data set;

[0009] The time sequence features in the first data set are extracted by using a convolutional neural network, the time sequence features are converted into a data form of a deep learning model through an embedding layer, and one-dimensional time sequence data is obtained;

[0010] The one-dimensional time sequence data is subjected to fast Fourier transform to obtain a frequency curve, the amplitude values on the frequency curve are extracted, corresponding periods are calculated according to the amplitude values, and the one-dimensional time sequence data is sliced and divided into a two-dimensional matrix by selecting the corresponding periods;

[0011] The two-dimensional matrix is subjected to feature extraction by using a two-dimensional convolutional network, the two-dimensional matrix subjected to the feature extraction is restored into a one-dimensional array, and the one-dimensional array is subjected to adaptive fusion to obtain first time sequence features;

[0012] The first time sequence features are input into a full connection layer to perform weight calculation and linear transformation, and a power distribution network load prediction result is obtained.

[0013] As a preferred scheme of the power distribution network load prediction method based on deep learning, wherein the data set is preprocessed, including:

[0014] The first quartile and the third quartile in the data set are calculated, the third quartile is subtracted from the first quartile to obtain a quartile range;

[0015] The normal value range is that the lower limit is obtained by extending the first quartile downward by 1.5 times the quartile range, and the upper limit is obtained by extending the third quartile upward by 1.5 times the quartile range;

[0016] The values exceeding the normal range are abnormal values, which are removed from the data set;

[0017] For the missing part in the data set, a K value is selected, the Euclidean distance between each missing value sample and other samples is calculated, the K nearest neighbor samples are selected, and the missing value is filled by using the weighted average value of the K nearest neighbor samples.

[0018] As a preferred scheme of the power distribution network load prediction method based on deep learning, wherein the one-dimensional time sequence data includes:

[0019] The time sequence features of each column in the data set are converted into a data form processed by TimesBlock through an embedding layer, data is transmitted between each layer of TimesBlock through a residual operation, and the input data form of the kth layer of TimesBlock, i.e., one-dimensional time sequence data, is obtained.

[0020] As a preferred scheme of the power distribution network load prediction method based on deep learning, wherein the fast Fourier transform includes:

[0021] The one-dimensional time series data is converted into a frequency spectrum curve, the amplitude value corresponding to each positive frequency on the frequency spectrum curve is extracted, according to the setting of the hyperparameter, the first n frequency components with average amplitude are selected, and the corresponding period is calculated and expressed as:

[0022] wherein, is one-dimensional time series data, FFT(·) is fast Fourier transform, Amp(·) is an amplitude function, Avg(·) is an average value function, {f1,...,f n} is n frequency components, {p1,...,p n} is the corresponding period, W is the average amplitude corresponding to the frequency component, and T is the data length.

[0023] As a preferred scheme of the power distribution network load prediction method based on deep learning provided by the application, wherein: the one-dimensional time series data is divided into a two-dimensional matrix by slicing according to the corresponding period, including:

[0024] According to the size of each period, the one-dimensional time series data is divided into multiple slices, and each slice corresponds to sequence data in a period;

[0025] 0 is supplemented at the end of the one-dimensional time series data until the length of the one-dimensional time series data can be divided by the period, and the one-dimensional time series data is reorganized to obtain a two-dimensional matrix.

[0026] As a preferred scheme of the power distribution network load prediction method based on deep learning provided by the application, wherein: the two-dimensional matrix is subjected to feature extraction by using a two-dimensional convolution network, and the two-dimensional matrix subjected to feature extraction is restored to a one-dimensional array, including:

[0027] The two-dimensional matrix is taken as input data, and an Inception convolution network is used to perform convolution processing on the two-dimensional matrix to obtain a two-dimensional matrix subjected to feature extraction;

[0028] The two-dimensional matrix subjected to feature extraction is reshaped into a one-dimensional vector according to the number of rows and columns of the matrix, and the padding 0 added in the reshaping process is removed to obtain one-dimensional time series data, that is, a one-dimensional array.

[0029] As a preferred scheme of the power distribution network load prediction method based on deep learning provided by the application, wherein: the one-dimensional array is adaptively fused, including:

[0030] The weight of each frequency component at k-1 step is calculated, and the weight is converted into a probability distribution by using a Softmax function;

[0031] The weight of each frequency component is multiplied by the corresponding one-dimensional array, and the multiplication result is summed to obtain a one-dimensional time array after weighting.

[0032] In a second aspect, the present application provides a system for power distribution network load forecasting based on deep learning, comprising:

[0033] a preprocessing module configured to obtain regional load data and renewable energy generation data, form a data set from the regional load data and the renewable energy generation data, and preprocess the data set to obtain a first data set;

[0034] a data conversion module configured to extract time series features in the first data set using a convolutional neural network, convert the time series features into a data form of a deep learning model through an embedding layer to obtain one-dimensional time series data;

[0035] a slicing module configured to perform fast Fourier transform on the one-dimensional time series data to obtain a frequency curve, extract amplitude values on the frequency curve, calculate corresponding periods according to the amplitude values, and select the corresponding periods to slice and divide the one-dimensional time series data into a two-dimensional matrix;

[0036] a feature extraction module configured to extract features of the two-dimensional matrix using a two-dimensional convolutional network, restore the two-dimensional matrix after feature extraction to a one-dimensional array, and perform adaptive fusion on the one-dimensional array to obtain first time series features;

[0037] a calculation module configured to input the first time series features into a fully connected layer to perform weight calculation and linear transformation, and obtain a power distribution network load forecasting result.

[0038] In a third aspect, the present application provides a computing device, comprising:

[0039] a memory and a processor;

[0040] The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, so as to realize the steps of the power distribution network load forecasting method based on deep learning.

[0041] In a fourth aspect, the present application provides a computer readable storage medium storing computer executable instructions, which are executed by a processor to realize the steps of the power distribution network load forecasting method based on deep learning.

[0042] The power distribution network load prediction method based on deep learning provided by the present application adopts a TimesNet network to construct a load prediction model, combines the advantages of deep learning technology, improves the accuracy of load prediction, and ensures the accuracy and integrity of the data set through data screening, outlier detection, missing value filling and data normalization processing, thereby providing high-quality input data for the model, effectively managing the load of the power distribution network, and improving the stability and reliability of power supply. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0044] Fig. 1 is a whole flow chart of the power distribution network load prediction method based on deep learning provided by the present application;

[0045] Fig. 2 is a wind power generation load prediction result chart of the power distribution network load prediction method based on deep learning provided by the present application;

[0046] Fig. 3 is a photovoltaic power generation load prediction result chart of the power distribution network load prediction method based on deep learning provided by the present application;

[0047] Fig. 4 is a regional load prediction result chart of the power distribution network load prediction method based on deep learning provided by the present application;

[0048] Fig. 5 is a power distribution network load prediction result chart considering wind energy-photovoltaic access of the power distribution network load prediction method based on deep learning provided by the present application. DETAILED DESCRIPTION

[0049] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0050] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0051] Secondly, the "one embodiment" or "an embodiment" referred to herein means a specific feature, structure, characteristic, or combination of features and / or characteristics described herein that can be included in at least one implementation of the present application. The various appearances of "in one embodiment" or "an embodiment" in the specification do not all refer to the same embodiment, although they can.

[0052] The application is described in detail in conjunction with the schematic drawings, and in the detailed description of the embodiments of the application, the sectional view of the device structure is partially enlarged without the general proportion for the convenience of illustration, and the schematic drawings are only examples, which should not limit the scope of protection of the application herein. In addition, the three-dimensional spatial dimensions of length, width and depth should be included in actual production.

[0053] Meanwhile, in the description of the application, it should be noted that the terms "upper, lower, inner and outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application. In addition, the terms "first, second or third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0054] Unless otherwise specifically defined and limited in the present application, the terms "mounting, connecting, connecting" should be broadly understood, for example: it can be fixedly connected, detachably connected or integrally connected; it can also be mechanically connected, electrically connected or directly connected, it can also be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0055] Embodiment 1

[0056] Referring to FIG. 1, according to an embodiment of the present application, a power distribution network load prediction method based on deep learning is provided, comprising:

[0057] S100: Obtain regional load data and renewable energy power generation data, group the regional load data and renewable energy power generation data into a data set, preprocess the data set to obtain a first data set;

[0058] In the application embodiment, the first quartile and the third quartile in the data set are calculated, and the third quartile is subtracted from the first quartile to obtain the interquartile range;

[0059] The normal value range is 1.5 times the interquartile range extending downward from the first quartile to obtain the lower limit, and 1.5 times the interquartile range extending upward from the third quartile to obtain the upper limit;

[0060] Values beyond the normal range are outliers, removed from the dataset;

[0061] For missing parts in the dataset, select the value of K, calculate the Euclidean distance between each missing value sample and other samples, select the nearest K samples, and fill in the missing values using the weighted average of the K nearest neighbor samples.

[0062] For example, obtaining regional load data and renewable energy generation data includes collecting relevant characteristic factors that affect wind power generation, including wind speed, wind direction, air temperature, and air pressure, to represent wind power generation as much as possible. Wind speed and wind direction directly affect the power generation efficiency of wind turbines, while air temperature and air pressure affect air density, which directly affects the kinetic energy content of wind energy and the efficiency of wind energy conversion. At the same time, collect photovoltaic power generation related data, including solar irradiance, total cloud cover, air temperature, and air pressure, to ensure that the environmental conditions of photovoltaic power generation are fully considered. Solar irradiance and total cloud cover directly affect the power generation capacity of photovoltaic panels, while air temperature and air pressure affect the overall performance of the photovoltaic system. In addition, collect regional load related data, including historical load data and user electricity consumption behavior, to provide background information on regional power distribution network load changes. Historical load data helps understand past electricity consumption patterns and load peak-valley changes, while user electricity consumption behavior reflects the periodicity and suddenness of electricity demand, allowing for more accurate prediction of future load changes.

[0063] For the obtained data, through correlation analysis calculation, the characteristics that significantly affect wind energy, photovoltaic and regional load are selected respectively to form a dataset, ensuring that the model only uses the most relevant data, reducing data redundancy and unnecessary computing power consumption; data cleaning is performed on the dataset, and the interquartile range (IQR) method is used to detect and remove outliers in the dataset, including calculating the first quartile (Q1) and the third quartile (Q3) of the dataset, calculating the interquartile range (IQR): IQR = Q3 - Q1; define the normal value range: Q = [Q1 - 1.5 * IQR, Q3 + 1.5 * IQR];

[0064] Use K-Nearest Neighbor Interpolation (KNN) to fill in missing parts of the dataset to ensure the accuracy and integrity of the dataset, including selecting the value of K, which can be 5, calculating the Euclidean distance between each missing value sample and other samples, represented as:

[0065] where d(x, y) is the Euclidean distance between sample x and sample y;

[0066] Fill in the missing values using the weighted average of the K nearest neighbor samples, represented as:

[0067] where, is the filled missing value, xi is the value of the i-th nearest neighbor sample, and K is the number of selected neighboring samples.

[0068] The data in the data set of wind energy, photovoltaic and regional load is normalized by using the min-max normalization method, and complex nonlinear time series features in the data set are extracted by a convolutional neural network (CNN).

[0069] Specifically, the missing value processing methods include mean / median filling, linear interpolation and K nearest neighbor interpolation method, the abnormal value detection methods include standard deviation method, model-based method and interquartile range method, and the application prefers the interquartile range method and the K nearest neighbor interpolation method.

[0070] It should be noted that there may be abnormal values in the power distribution network load data and renewable energy generation data, which may interfere with the training of the deep learning model and affect the prediction accuracy. The interquartile range method can effectively identify and remove abnormal data points, ensuring the stability and consistency of the data. Compared with other abnormal value detection methods, the interquartile range method can flexibly cope with the fluctuations in the power distribution network load data, ensuring the robustness of data preprocessing.

[0071] In the actual data collection process, there may be some missing data. If not handled, these missing values will also affect the training of the model. Compared with other missing value processing methods, the K nearest neighbor method (KNN) can use the existing sample information in the data set to fill in the missing values by calculating the similarity, maintain the integrity and continuity of the data, and ensure that the filled data is reasonable and accurate.

[0072] By removing abnormal values using the IQR method, the noise and redundancy in the data set can be reduced, improving the training efficiency and accuracy of the model. KNN fills in the missing values, ensuring the integrity of the data set and avoiding information loss due to missing data. The interquartile range method and the K nearest neighbor interpolation method have obvious advantages in data preprocessing, which can improve data quality, reduce the consumption of computing resources, and provide more accurate and reliable data basis for subsequent model training and prediction.

[0073] S200: Extracting time series features in the first data set using a convolutional neural network, converting the time series features into a data form of a deep learning model through an embedding layer to obtain one-dimensional time series data;

[0074] In the embodiments of the present application, the time series features of each column in the data set are converted into the data form of TimesBlock processing through an embedding layer, and the data is transmitted between each layer of TimesBlock through a residual operation to obtain the input data form of the kth layer of TimesBlock, i.e. one-dimensional time series data.

[0075] Exemplarily, each column of time series data in the wind power, photovoltaic and regional load data set is transformed to represent as: wherein, Z 1D is each column of time series data in the wind power, photovoltaic and regional load data set, F(·) is a function of transforming Z 1D is input into the embedding layer for processing, is the input data of the first layer TimeBlock in the model, and the input data form of the kth layer TimeBlock module is represented as:

[0076] wherein, is the input data form of the (k-1)th layer TimeBlock module, and TimesBlock(·) is the conversion form of the TimeBlock module.

[0077] It should be noted that TimesBlock, as a deep learning structure, can capture complex patterns and long-term dependencies in time series data. Through the embedding layer, the original data can be transformed into more rich feature representation. The residual connection helps to alleviate the gradient vanishing problem in deep learning models, so that the deep network can be trained more effectively, which helps to maintain the transmission of long-term dependence information in time series models. Through the combination of embedding layer and TimesBlock, the model can learn more abstract feature representation, which helps to improve the generalization ability of the model to unseen data.

[0078] S300: Fast Fourier transform is performed on the one-dimensional time series data to obtain a frequency curve, the amplitude value on the frequency curve is extracted, the corresponding period is calculated according to the amplitude value, and the one-dimensional time series data is sliced and divided into a two-dimensional matrix by selecting the corresponding period;

[0079] In the embodiment of the present application, the one-dimensional time series data is converted into a frequency spectrum curve, the amplitude value corresponding to each positive frequency on the frequency spectrum curve is extracted, according to the setting of the hyperparameter, the average amplitude of the first n frequency components is selected, and the corresponding period is calculated and represented as:

[0080] wherein, is one-dimensional time series data, FFT(·) is fast Fourier transform, Amp(·) is an amplitude function, Avg(·) is an average function, {f1,...,f n} is n frequency components, {p1,...,p n} is the corresponding period, W is the average amplitude corresponding to the frequency component, and T is the data length.

[0081] It should be noted that FFT can transform time series data from the time domain to the frequency domain, making the periodic features in the data more obvious and helping to identify and analyze the periodic components in the data. By selecting frequency components with larger amplitudes, the features that contribute the most to the data pattern can be extracted, thereby simplifying the model and potentially improving prediction accuracy. By selecting only the n frequency components with larger average amplitudes, the data can be compressed, reducing the computational load of subsequent processing. Applying FFT in the TimesBlock module can effectively process and analyze time series data, providing strong support for subsequent feature learning and model training.

[0082] In this embodiment of the application, one-dimensional time series data is divided into multiple slices according to the size of each period, and each slice corresponds to the sequence data within one period;

[0083] Add zeros to the end of the one-dimensional time series data until the length of the one-dimensional time series is divisible by the period, and then reorganize the one-dimensional time series to obtain a two-dimensional matrix.

[0084] Specifically, based on the selected n periods, respectively The slice is divided into n groups of two-dimensional matrices, each of which has a dimension of (p... i ,f i );

[0085] in, To pair based on the size of the i-th period The two-dimensional matrix for slicing, i∈{1,…,n}; Padding(·) is the pagination of the slices. The sequence is padded with zeros so that the sequence length can be p. i Divisibility yields a set of two-dimensional matrices. Corresponding to a period of p i Two-dimensional temporal changes;

[0086] S400: Use a two-dimensional convolutional network to extract features from a two-dimensional matrix, restore the two-dimensional matrix after feature extraction to a one-dimensional array, and adaptively fuse the one-dimensional array to obtain the first time series features;

[0087] In this embodiment of the application, a two-dimensional matrix is ​​used as input data, and the Inception convolutional network is used to perform convolution processing on the two-dimensional matrix to obtain a two-dimensional matrix after feature extraction.

[0088] The two-dimensional matrix after feature extraction is reshaped into a one-dimensional vector according to the number of rows and columns of the matrix. The padding zeros added during the reshaping process are removed to obtain one-dimensional time series data, i.e., a one-dimensional array.

[0089] Specifically, a two-dimensional convolutional network is used to process n two-dimensional matrices. The feature extraction is represented as: Wherein, is the two-dimensional matrix after feature extraction according to the i th period, The two-dimensional matrix after feature extraction, f(·) is a two-dimensional matrix Convolution processing is performed; the convolution network can be an Inception convolution network, a ResNet network, and an Xception network, and the Inception convolution network is preferred in the present application.

[0090] It should be noted that, compared with other convolution networks, the Inception network can capture the features of the image at different scales at the same time by applying different sizes of convolution kernels (1x1, 3x3, 5x5) and maximum pooling, so that the network can more comprehensively understand the image content, the 1x1 convolution kernel in the Inception module is used to reduce the dimension of the feature map before performing large size convolution operation, thereby reducing the amount of calculation and preventing overfitting, and multiple different size convolution kernels are used to capture feature information at different scales. This multi-scale convolution operation is beneficial to extract rich feature information.

[0091] Specifically, the two-dimensional matrix after feature extraction is restored to a one-dimensional representation as:

[0092] Wherein, The dimension of is (1, (p i *f i )), and Trunc(·) is to remove the 0 supplemented by Padding(·).

[0093] In the embodiments of the present application, the weight of each frequency component at the k-1 step is calculated, and the weight is converted into a probability distribution by a Softmax function;

[0094] The weight of each frequency component is multiplied by the corresponding one-dimensional array, and the multiplication result is summed to obtain a weighted one-dimensional time array;

[0095] In the embodiments of the present application, the obtained one-dimensional array is weighted and summed with the average amplitude of its corresponding frequency to obtain a representation as:

[0096] Wherein, is the weight of the frequency component {f1, f 2,…, f n}, and Softmax(·) is an activation function.

[0097] Specifically, TimesNet completes multiple cycles of extracting two-dimensional time series changes, and then performs adaptive fusion of the time series change modeling process. The final time series features are flattened and input into the fully connected layer. The fully connected layer performs weight calculation and linear transformation on the flattened features to generate load prediction results. In addition to the TimesNet model, the Informer model and the LSTM model can also be used.

[0098] It should be noted that the TimesNet model is based on the Transformer architecture and is good at processing long-time span data, which can capture long-term dependencies in time series, i.e., it can more accurately capture time series features in the data. The key indicator that affects the accuracy of load prediction is whether it can accurately extract the time series features in the time series data. This makes it have higher prediction accuracy and robustness in distribution network load prediction, thereby better meeting the prediction needs of modern distribution networks.

[0099] Compared with other deep learning models, the TimesNet is preferred to build a load prediction model. TimesNet can capture both intra-period and inter-period changes in time series data by converting one-dimensional time series data into a two-dimensional tensor, thereby more accurately modeling time series data with multiple periods. TimesNet has shown superior performance to other models in multiple time series tasks, including short-term and long-term prediction, classification, anomaly detection, and missing value processing.

[0100] The above is a schematic scheme of the power distribution network load prediction method based on deep learning of the embodiment. It should be noted that the technical scheme of the power distribution network load prediction device based on deep learning belongs to the same concept as the technical scheme of the power distribution network load prediction method based on deep learning described above. The technical scheme of the power distribution network load prediction device based on deep learning in the embodiment is not described in detail, and can be referred to the description of the technical scheme of the power distribution network load prediction method based on deep learning described above.

[0101] The power distribution network load prediction device based on deep learning in the embodiment comprises:

[0102] The preprocessing module is configured to obtain regional load data and renewable energy generation data, group the regional load data and renewable energy generation data into a data set, and preprocess the data set to obtain a first data set.

[0103] The data conversion module is configured to extract time series features in the first data set using a convolutional neural network, convert the time series features into a data form of a deep learning model through an embedding layer, and obtain one-dimensional time series data.

[0104] The slice module is configured to perform fast Fourier transform on the one-dimensional time series data to obtain a frequency curve, extract an amplitude value on the frequency curve, calculate a corresponding period according to the amplitude value, and select the corresponding period to slice and divide the one-dimensional time series data into a two-dimensional matrix;

[0105] The feature extraction module is configured to perform feature extraction on the two-dimensional matrix by using a two-dimensional convolution network, restore the two-dimensional matrix after the feature extraction to a one-dimensional array, and perform adaptive fusion on the one-dimensional array to obtain a first time series feature.

[0106] The calculation module is configured to input the first time series feature into a full connection layer to perform weight calculation and linear transformation, and obtain a power grid load prediction result.

[0107] The embodiment also provides a computing device suitable for power grid load prediction based on deep learning, including:

[0108] The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the power grid load prediction method based on deep learning.

[0109] The embodiment also provides a storage medium having a computer program stored thereon, and the program is executed by a processor to implement the power grid load prediction method based on deep learning.

[0110] The storage medium provided by the embodiment belongs to the same inventive concept as the power grid load prediction method based on deep learning provided by the above embodiment, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk or an optical disk, and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0112] Embodiment 2

[0113] Referring to FIGS. 2-5, for one embodiment of the present application, a deep learning-based load forecasting method for power distribution networks is provided to verify the technical effects used in this method.

[0114] Taking a power distribution network system in a coastal city as an example, the system comprehensively utilizes wind energy and photovoltaic power generation to meet the electricity demand of local residents and industries.

[0115] The voltage level of this power distribution network system is 10kV, and there are 10 wind turbine generators and one photovoltaic power station installed in the covered area. The rated power of each wind turbine generator is 500kW, and the total installed capacity reaches 5000kW, while the total installed capacity of the photovoltaic power station is 3000kW;

[0116] The system collects environmental data such as wind speed, wind direction, solar irradiance, total cloud cover, air temperature, and air pressure, and combines historical load data and user electricity behavior data to build a high-precision load forecasting model, thereby achieving accurate prediction and effective management of the power distribution network load.

[0117] As shown in FIGS. 2, 3 and 4, the wind power generation load, photovoltaic power generation load and regional load prediction results, respectively, overall, the predicted value (circle marker) and the actual value (diamond marker) have good fitting degree;

[0118] As can be seen from FIG. 2, the wind power generation load power rapidly decreases from about 350kW at 1 hour to about 150kW at 5 hours, then shows obvious fluctuations at 10 hours and 15 hours, and finally tends to zero after 20 hours. The change trend within 1-24 hours is accurately captured by the model;

[0119] As shown in FIG. 3, with the natural law of sunrise and sunset, the photovoltaic power generation power gradually increases in the morning, reaches a peak of about 700kW at noon, then gradually decreases, and tends to zero at night. The model's prediction of this change trend is very accurate;

[0120] As shown in FIG. 4, the regional load gradually increases from about 1000kW in the early morning to about 1800kW at 6 hours, and maintains a high level in the small fluctuations thereafter, and finally rapidly decreases after 20 hours. The predicted value and the actual value are basically consistent in overall trend and detailed changes, showing the high precision of the model in regional load prediction.

[0121] Figure 5 is a load forecasting result chart of a distribution network region with flexible resource access. Due to the access of wind energy and photovoltaic, the original load curve of the distribution network region is significantly affected. Photovoltaic power generation provides power supplement for the distribution network during the day, which significantly reduces the load peak-valley difference. Wind power generation provides power support for the distribution network at different time periods throughout the day. Overall, the access of these two renewable energy sources effectively improves the stability and reliability of power supply.

[0122] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A power distribution network load forecasting method based on deep learning, characterized in that, The method comprises the following steps: obtaining regional load data and renewable energy power generation data, grouping the regional load data and renewable energy power generation data into a data set, preprocessing the data set to obtain a first data set; extracting time series features in the first data set by using a convolutional neural network, converting the time series features into a data form of a deep learning model through an embedding layer to obtain one-dimensional time series data; performing fast Fourier transform on the one-dimensional time series data to obtain a frequency curve, extracting an amplitude value on the frequency curve, calculating a corresponding period according to the amplitude value, and slicing the one-dimensional time series data into a two-dimensional matrix by selecting the corresponding period; extracting features of the two-dimensional matrix by using a two-dimensional convolutional network, restoring the two-dimensional matrix after feature extraction to a one-dimensional array, and adaptively fusing the one-dimensional array to obtain first time series features; inputting the first time series features into a fully connected layer to perform weight calculation and linear transformation, and obtaining a power distribution network load prediction result.

2. The power distribution network load forecasting method based on deep learning according to claim 1, wherein, The preprocessing of the data set comprises the following steps: calculating a first quartile and a third quartile in the data set, and obtaining a quartile range by subtracting the first quartile from the third quartile; the normal value range is from the first quartile to the lower limit of the normal value range, which is obtained by extending 1.5 times the quartile range downward, and from the third quartile to the upper limit of the normal value range, which is obtained by extending 1.5 times the quartile range upward; values exceeding the normal range are abnormal values, which are removed from the data set; for the missing part in the data set, a K value is selected, the Euclidean distance between each missing value sample and other samples is calculated, the K nearest neighbor samples are selected, and the missing value is filled by using the weighted average value of the K nearest neighbor samples.

3. The power distribution network load forecasting method based on deep learning according to claim 1 or 2, characterized in that, The one-dimensional time series data comprises the following steps: the time series features in each column of the data set are converted into a data form processed by TimesBlock through an embedding layer, data is transmitted between each layer of TimesBlock through a residual operation, and the input data form of the kth layer of TimesBlock, i.e. one-dimensional time series data, is obtained.

4. The power distribution network load forecasting method based on deep learning according to claim 3, wherein, The fast Fourier transform comprises the following steps: The one-dimensional time series data is converted into a frequency spectrum curve, the amplitude value corresponding to each positive frequency on the frequency spectrum curve is extracted, according to the setting of the hyperparameters, the first n frequency components with the average amplitude are selected, and the corresponding period is calculated and expressed as: wherein For one-dimensional time series data, FFT(·) is the fast Fourier transform, Amp(·) is the amplitude function, Avg(·) is the average function, {f1,...,f n} are n frequency components, {p1,...,p n} are corresponding periods, W is the average amplitude corresponding to the frequency component, and T is the data length.

5. The power distribution network load forecasting method based on deep learning according to claim 4, wherein, slicing the one-dimensional time series data into a two-dimensional matrix by selecting a corresponding period comprises the following steps: according to the size of each period, the one-dimensional time series data is divided into multiple slices, each slice corresponds to sequence data within a period; 0 is supplemented at the end of the one-dimensional time series data until the length of the one-dimensional time series data can be divided by the period, and the one-dimensional time series data is reorganized to obtain a two-dimensional matrix.

6. The power distribution network load forecasting method based on deep learning according to claim 5, wherein, extracting features of the two-dimensional matrix by using a two-dimensional convolutional network, restoring the two-dimensional matrix after feature extraction to a one-dimensional array, and adaptively fusing the one-dimensional array to obtain first time series features comprises the following steps: the two-dimensional matrix is taken as input data, an Inception convolutional network is used to convolve the two-dimensional matrix to obtain a two-dimensional matrix after feature extraction; the two-dimensional matrix after feature extraction is reshaped into a one-dimensional vector according to the number of rows and columns of the matrix, the padding 0 added in the reshaping process is removed, and a one-dimensional time series data, i.e. a one-dimensional array, is obtained.

7. The power distribution network load forecasting method based on deep learning according to claim 6, wherein, The adaptive fusion of the one-dimensional array comprises the following steps: calculating the weight of each frequency component at step k-1, and converting the weight into a probability distribution by using a Softmax function; The weight of each frequency component is multiplied by a corresponding one-dimensional array, and the multiplication results are summed to obtain a weighted one-dimensional time array.

8. A system for power distribution network load forecasting based on deep learning, characterized in that, The method comprises the following steps: The pre-processing module is configured to obtain regional load data and renewable energy power generation data, combine the regional load data and the renewable energy power generation data into a data set, and pre-process the data set to obtain a first data set. The data conversion module is configured to extract time sequence features in the first data set by using a convolutional neural network, convert the time sequence features into a data form of a deep learning model through an embedding layer, and obtain one-dimensional time sequence data. The slicing module is configured to perform fast Fourier transform on the one-dimensional time sequence data to obtain a frequency curve, extract amplitude values on the frequency curve, calculate corresponding periods according to the amplitude values, and slice and divide the one-dimensional time sequence data into a two-dimensional matrix according to the corresponding periods. The feature extraction module is configured to extract features of the two-dimensional matrix by using a two-dimensional convolutional network, restore the two-dimensional matrix after feature extraction to a one-dimensional array, and adaptively fuse the one-dimensional array to obtain first time sequence features. The calculation module is configured to input the first time sequence features into a full connection layer to perform weight calculation and linear transformation, and obtain a power distribution network load prediction result.

9. An electronic device, comprising: The device comprises: a processor; a memory for storing processor-executable instructions; the processor is configured to invoke the instructions stored in the memory to execute the method of any one of claims 1-7.

10. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions are executed by the processor to implement the method of any one of claims 1-7.

Images