Power load prediction method and related device

Abstract

The invention discloses a power load prediction method and a related device, and the method comprises the steps: determining a target future date for carrying out the power load prediction, determining d impact factors affecting the power load prediction, selecting a date similar to the target future date from a preset historical time period according to the d impact factors, and carrying out the prediction of the power load. The method comprises the steps of obtaining p similar dates, obtaining a load sequence in a preset historical time period, obtaining a target load sequence, decomposing the target load sequence into a high-frequency component, an intermediate-frequency component and a low-frequency component through a variational mode decomposition method, and determining a predicted load sequence of a target future date according to the p similar dates, the high-frequency component, the intermediate-frequency component and the low-frequency component. According to the invention, the method achieves the accurate prediction of the power load.

Description

Technical Field

[0001] The present application relates to the technical field of load forecasting, and in particular to a method for forecasting electric load and related devices. Background Art

[0002] Power load forecasting is one of the important tasks in power system operation and planning. The accuracy of its forecast results is directly related to the economy, safety and stability of the power system. Existing load forecasting methods are mainly divided into two categories: traditional statistical models and machine learning-based models.

[0003] First, traditional statistical models, such as the autoregressive integrated moving average model and exponential smoothing, perform poorly when dealing with complex, nonlinear data features. These models often assume that load data has a linear trend and fixed periodicity, and are difficult to cope with the complexity of various factors that affect load fluctuations (such as weather changes, socioeconomic activities, and emergencies), resulting in limited prediction accuracy. Secondly, although machine learning and deep learning models, such as long short-term memory networks and gated recurrent units, can better handle complex nonlinear relationships, they are highly data-dependent. These models usually require a large amount of historical data for training, and when the data is incomplete or of low quality, the model's predictive ability will be significantly reduced.

[0004] In short, the existing load forecasting technology is difficult to fully handle the impact of non-periodic factors such as holidays and emergencies. These events will significantly change the power demand pattern. The existing models often produce large forecasting deviations when dealing with such irregular changes. Therefore, how to accurately forecast power load has become an urgent problem to be solved. Summary of the invention

[0005] The embodiment of the present application provides a method for predicting electric power load and related devices, which predicts the electric power load of future dates by combining similar days similar to the future dates to be predicted and a variational mode decomposition method, thereby achieving accurate prediction of the electric power load.

[0006] In a first aspect, an embodiment of the present application provides a power load prediction method, which is applied to a server, comprising:

[0007] Determine target future dates;

[0008] Determine d influencing factors; d is a positive integer;

[0009] According to the d influencing factors, a date similar to the target future date is selected from a preset historical time period to obtain p similar dates; p is a positive integer;

[0010] Obtaining a load sequence within the preset historical time period to obtain a target load sequence;

[0011] Decomposing the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by a variational mode decomposition method; the high-frequency component, the medium-frequency component and the low-frequency component are all time series;

[0012] A forecast load sequence for the target future date is determined according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component.

[0013] In a second aspect, an embodiment of the present application provides a power load forecasting device, which is applied to a server and includes: a data initialization module, a date selection module, a sequence decomposition module, and a load forecasting module, wherein:

[0014] The data initialization module is used to determine the target future date;

[0015] The data initialization module is also used to determine d influencing factors; d is a positive integer;

[0016] The date selection module is used to select a date similar to the target future date from a preset historical time period according to the d influencing factors, to obtain p similar dates; p is a positive integer;

[0017] The data initialization module is also used to obtain the load sequence within the preset historical time period to obtain the target load sequence;

[0018] The sequence decomposition module is used to decompose the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by using a variational mode decomposition method; the high-frequency component, the medium-frequency component and the low-frequency component are all time series;

[0019] The load forecasting module is used to determine the forecast load sequence of the target future date according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component.

[0020] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the program includes instructions for executing the steps in the first aspect of the embodiment of the present application.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps described in the first aspect of the embodiment of the present application.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps described in the first aspect of the embodiment of the present application. The computer program product may be a software installation package.

[0023] It can be seen that the following beneficial effects are achieved by using the embodiments of the present application:

[0024] By implementing the embodiment of the present application, the target future date for power load forecasting is determined, and d influencing factors affecting power load forecasting are determined. According to the d influencing factors, a date similar to the target future date is selected from a preset historical time period to obtain p similar dates, and a load sequence within the preset historical time period is obtained to obtain a target load sequence. The target load sequence is decomposed into high-frequency components, medium-frequency components and low-frequency components by a variational mode decomposition method, and the predicted load sequence of the target future date is determined according to p similar dates, high-frequency components, medium-frequency components and low-frequency components. It realizes the prediction of power load of complex nonlinear data and non-periodic factors by comprehensively considering multiple influencing factors, improves the accuracy of power load prediction, overcomes the shortcomings of traditional statistical models and machine learning models in dealing with complex situations, provides more accurate decision-making basis for power system operation and planning, and enhances the economy, safety and stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background technology, the drawings required for use in the embodiments of the present application or the background technology will be described below.

[0026] Figure 1 It is a flow chart of a method for predicting power load provided in an embodiment of the present application;

[0027] Figure 2 It is a structural schematic diagram of a power load prediction device provided in an embodiment of the present application;

[0028] Figure 3 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0030] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. 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 limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0031] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0032] The following is an explanation of the relevant contents, concepts, meanings, technical issues, technical solutions, beneficial effects, etc. involved in the embodiments of the present application.

[0033] First, some terms involved in this application are explained:

[0034] Variational Mode Decomposition Method: Variational Mode Decomposition (VMD) method is a signal processing method used to decompose a complex original signal into multiple modal components with different center frequencies.

[0035] Long Short-Term Memory Network: Long-Short Term Memory (LSTM) is a neural network model suitable for classification, processing and prediction based on time series data. The LSTM unit consists of a memory unit, an input gate, an output gate and a forget gate. These mechanisms enable LSTM to better handle long-term dependencies in the sequence.

[0036] See also Figure 1 , Figure 1 : is a flow chart of a power load forecasting method provided in an embodiment of the present application, the method is applied to a server, and the method includes but is not limited to the following steps:

[0037] S101. Determine a target future date.

[0038] In the embodiment of the present application, the target future date refers to a specific future time point or time period for which the power load conditions need to be estimated in advance. By predicting the power load on the target future date, power companies, grid operators and other relevant entities can formulate power generation plans, arrange power resource allocation and ensure power supply reliability.

[0039] In a possible embodiment, the server provides a user interface, and the user can directly specify the target future date through a calendar selector on the interface. For example, May 1 can be selected on the calendar selector and determined as the target future date for performing power load forecasting.

[0040] S102. Determine d influencing factors.

[0041] Wherein, d is a positive integer. Wherein, the influencing factor refers to a factor related to the load, that is, various factors that can affect the size and change trend of the power load, for example, weather factors that affect the power demand of air conditioners, heating equipment, etc., are related to people's life and work rules, and thus affect the time factors of the periodic change of the power load, reflecting the economic activity factors of different industries and commercial behaviors on power consumption, or special event factors such as large-scale sports events and concerts. In the embodiment of the present application, the influencing factor may include one of the following: maximum temperature, average temperature, humidity, wind speed, electricity price, atmospheric pressure, working day index, holidays, seasons, etc., which are not limited here. It should be noted that for some influencing factors without fixed units, such as the working day index, reasonable assignments are required to be made in order to accurately calculate and analyze in the model. For example, for working days from Monday to Friday, they can be assigned to 0.8, 0.9, 1, 0.9, and 0.8 respectively, and for Saturday and Sunday, they can be assigned to 0.3 and 0.2. The larger the value, the closer it is to the working day, and the greater the impact on the power load.

[0042] In a specific embodiment, the original data of the influencing factors that may affect the power load can be obtained from multiple data sources, and the data source can be a data platform such as an Internet data platform. Then, the original data is standardized, and the standardization method can use Max-Min standardization. Then, the correlation analysis method is used to calculate the correlation coefficient between each influencing factor and the historical power load data. For example, the Pearson correlation coefficient between the average temperature and the power load is calculated. If the correlation coefficient is greater than 0.5, it indicates that the average temperature has a strong influence on the power load, and the average temperature is used as the main influencing factor for power load prediction. Then, d influencing factors can be determined as reference factors for power load prediction.

[0043] S103. Selecting a date similar to the target future date from a preset historical time period according to the d influencing factors to obtain p similar dates.

[0044] Wherein, p is a positive integer. The preset historical time period refers to the time interval preset for obtaining historical data when conducting power load forecasting. The selection of this time period is usually determined according to the specific power load forecasting requirements. For example, the preset historical time period can be the past year or two years. The similar date refers to a date within the preset historical time period that has a high similarity with the target future date in terms of key factors affecting power load.

[0045] In a specific embodiment, a date similar to the target future date can be selected from a preset historical time period to obtain p similar dates, wherein a specific selection method can be to compare the data of d influencing factors of each date with the data of d influencing factors of the target future date to screen out the date that is most similar to the target future date in terms of each influencing factor. Since the load change trend of similar dates may be similar to the load change trend of the target future date under the same influencing factor condition, the load data of similar dates can provide a reference for predicting the power load of the target future date, so as to more accurately predict the power load of the target future date.

[0046] Optionally, the above step S103, selecting a date similar to the target future date from a preset historical time period according to the d influencing factors to obtain p similar dates, may include the following steps:

[0047] A301. Divide the preset historical time period into f dates based on a preset date granularity; f is an integer greater than p;

[0048] A302, obtaining samples corresponding to each of the f dates to obtain a target sample set; the samples include load data and impact factor data;

[0049] A303, determining the importance of each of the d influencing factors according to the target sample set, and obtaining d importances;

[0050] A304, determining the similarity between each of the f dates and the target future date according to the d importances, to obtain f similarities;

[0051] A305, selecting a similarity greater than or equal to a preset similarity threshold from the f similarities to obtain p similarities;

[0052] A306. Determine the p similarity corresponding dates as the p similar dates.

[0053] The preset date granularity is the basic unit for dividing the historical time period. Generally, the preset historical time period can be divided by day. The preset similarity threshold is the critical value for measuring the similarity between two dates in terms of influencing factors. For example, the preset similarity threshold can be set to 0.8. When the similarity calculation result is greater than or equal to 0.8, it can be determined that the historical date corresponding to the similarity is similar to the target future date.

[0054] In a specific embodiment, the preset historical time period is divided into f dates in units of days, where f is an integer greater than p. Then, samples corresponding to each of the f dates are obtained to construct a target sample set, wherein each sample includes load data and influencing factor data, the load data is the power load value, which can be obtained from the monitoring equipment records of the power system, and the influencing factor data is the data corresponding to each of the d influencing factors.

[0055] Further, each of the d influencing factors is analyzed according to the target sample set to determine the importance of each influencing factor, and d importances are obtained. The importance is used as a weight factor when calculating the date similarity, which represents the degree of influence of different influencing factors on the similarity calculation. After determining the importance of each of the g influencing factors, the similarity between each of the f dates and the target future date can be determined according to the d importances, and f similarities are obtained. From the f similarities, a similarity greater than or equal to a preset similarity threshold is selected to obtain p similarities. Then, the date corresponding to each similarity can be determined as a similar date according to the p similarities, and p similar dates are obtained. Each similar date has a high degree of similarity with the target future date in terms of the influencing factor. Using the load data of these similar dates for power load forecasting can more accurately predict the power load situation of the target future date.

[0056] Optionally, the above step A303, determining the importance of each of the d influencing factors according to the target sample set to obtain d importances, may include the following steps:

[0057] B301. Construct a decision tree according to the target sample set and the d influencing factors to obtain w decision trees; each decision tree includes g influencing factors; w is a positive integer; g is a positive integer less than or equal to d;

[0058] B302. Determine the importance of each of the d influencing factors in the w decision trees to obtain the d importances.

[0059] In an embodiment of the present application, in order to quantify the impact of different influencing factors among d influencing factors on the similarity between similar dates and target future dates, a decision tree can be constructed based on d influencing factors according to historical load data and influencing factor data. The decision tree is used to repeatedly divide the samples, and based on the structure of the decision tree and the feature division, the feature importance can be calculated. Therefore, the importance of each of the d influencing factors can be determined according to the decision tree.

[0060] In a specific embodiment, a sample can be randomly selected from the target sample set as a training set by a sampling replacement method to generate a decision tree, and the training set is feature-divided using d influencing factors to generate a decision tree. Therefore, by constructing a decision tree based on the target sample set and d influencing factors, w decision trees can be obtained, where w is a positive integer. In order to avoid the decision tree being too complicated or overfitting, g influencing factors can be selected to simplify the generation operation of the decision tree, so that each decision tree includes g influencing factors, that is, each decision tree is feature-divided according to g influencing factors, which improves the interpretability of the decision tree, where g is a positive integer less than or equal to d. When constructing a decision tree, d influencing factors are used for feature partitioning operations, so the importance of each of the d influencing factors can be determined based on w decision trees, and d importances are obtained, and d importances are used to reflect the influence of different influencing factors on the similarity between similar dates and target future dates.

[0061] Optionally, the above step B301, constructing a decision tree according to the target sample set and the d influencing factors to obtain w decision trees, may include the following steps:

[0062] C301. Randomly and non-repeatedly select the g influencing factors corresponding to the first decision tree from the d influencing factors; the first decision tree is any decision tree among the w decision trees;

[0063] C302, determining the Gini coefficient when each of the g influencing factors divides the target sample set, and obtaining g Gini coefficients;

[0064] C303. Select the smallest Gini coefficient among the g Gini coefficients to obtain a first Gini coefficient;

[0065] C304. Determine an impact factor corresponding to the first Gini coefficient to obtain a first impact factor;

[0066] C305. Divide the target sample set according to the first influencing factor to obtain a first node and a second node;

[0067] C306. When the first node and the second node do not meet the preset first condition, divide the first node and the second node respectively based on the g impact factors;

[0068] C307. When the first node and the second node meet the preset first condition, construct the first decision tree according to the first node and the second node.

[0069] The preset first condition refers to that the number of samples contained in the node is less than the sample threshold, or the purity of the node is higher than the purity threshold, wherein the sample threshold and the purity threshold can be set according to the actual situation. When the node meets the preset first condition, it means that the node is a leaf node of the decision tree, and then a decision tree can be obtained.

[0070] In a specific embodiment, g influencing factors corresponding to the first decision tree are randomly selected from the d influencing factors without duplication, wherein the g influencing factors are used to divide the sample set to construct a decision tree, wherein the first decision tree is any decision tree among the w decision trees. By randomly selecting some influencing factors for feature division from the d influencing factors, the constructed decision tree can maintain diversity, avoid each decision tree relying on the same combination of influencing factors, simplify the decision tree, and improve the interpretability of the decision tree.

[0071] Furthermore, during a feature partitioning, it is necessary to partition the nodes according to each of the g influencing factors, and calculate the Gini coefficient at each partitioning, and then obtain g Gini coefficients, where the data of the root node of a decision tree is the target sample set. By partitioning the nodes based on the influencing factors, multiple subsets can be obtained, and a node of a decision tree can be constructed based on the subsets. A subset corresponds to a category, and a category refers to a different type of label. For example, in a binary classification problem, the categories may be "yes" and "no", while in a multi-classification problem, the categories may be "category A", "category B", "category C", etc. The formula for calculating the Gini coefficient is as follows:

[0072]

[0073] In the formula, GI q represents the Gini coefficient of node q. The smaller the Gini coefficient, the higher the purity of node q, that is, the higher the probability that the samples in the node belong to the same category. S represents the total number of categories that the samples in node q may belong to. Represents the estimated probability that the sample in node q belongs to category s.

[0074] For example, in a certain node, the target sample set is divided into categories A (average temperature is higher than or equal to 25°C) and B (average temperature is less than 25°C). In the current node, the estimated probability that the target sample set belongs to category A is 0.6, and the estimated probability that the target sample set belongs to category B is 0.4. Therefore, it can be calculated that when the target sample set is divided into categories based on the average temperature, the Gini coefficient of the current node is 0.48.

[0075] Further, the smallest Gini coefficient among the g Gini coefficients is selected to obtain the first Gini coefficient, and the influence factor corresponding to the first Gini coefficient is determined to obtain the first influence factor. The first influence factor is the best partition influence factor, that is, the best partition feature, which means that in the current sample set, the first influence factor can maximize the purity of the subset when the sample set is partitioned, that is, it can better distinguish samples under different power load modes, so the first influence factor is used as the key feature for partitioning the sample set. The target sample set is divided according to the first influence factor to obtain the first node and the second node. Among them, the data of the first node and the second node are both relative samples in the target sample set. For example, the data of the first node is 60% of the samples in the target sample set belonging to category A (average temperature higher than or equal to 25°C), and the data of the second node is 40% of the samples in the target sample set belonging to category B (average temperature less than 25°C). Of course, when the sample set is divided, multiple subsets can be obtained, that is, multiple nodes are obtained.

[0076] Then, when the first node and the second node do not meet the preset first condition, the first node and the second node are divided based on the g influencing factors respectively, and the division process is repeated continuously, so that the first node and the second node continue to grow more child nodes until all child nodes meet the preset first condition, thereby forming a complete decision tree structure. When the first node and the second node meet the preset first condition, a complete first decision tree can be constructed based on the first node and the second node.

[0077] Optionally, when constructing a decision tree based on influencing factors, the Gini coefficient of each of the g influencing factors when dividing the target sample set is usually calculated to determine the best dividing influencing factor. The gain rate or information gain can also be introduced as a reference indicator to determine the best dividing influencing factor.

[0078] Optionally, the above step B302, determining the importance of each of the d influencing factors in the w decision trees to obtain the d importances, may include the following steps:

[0079] D301, obtaining a second impact factor; the second impact factor is any impact factor among the d impact factors;

[0080] D302, obtaining a target node; the target node is any node in the w decision trees that is divided based on the second influencing factor;

[0081] D303, obtain the child nodes of the target node, and obtain the third node and the fourth node;

[0082] D304, determining the Gini coefficient of the target node to obtain a second Gini coefficient;

[0083] D305, determining the Gini coefficients of the third node and the fourth node, and obtaining a third Gini coefficient and a fourth Gini coefficient respectively;

[0084] D306, determining a change amount of the Gini coefficient according to the second Gini coefficient, the third Gini coefficient and the fourth Gini coefficient;

[0085] D307. Determine the importance of the second influencing factor according to the change in the Gini coefficient.

[0086] In a specific embodiment, a second influencing factor is obtained, wherein the second influencing factor represents any influencing factor among d influencing factors, and a target node is obtained, wherein the target node represents any node in w decision trees that is divided based on the second influencing factor. Specifically, when there is a decision tree among the w decision trees, and a node in the decision tree is feature-divided based on the second influencing factor, then the node is the target node. Then the child nodes of the target node are obtained, and the third node and the fourth node are obtained respectively. The Gini coefficient of the target node is determined to obtain the second Gini coefficient, and the Gini coefficients of the third node and the fourth node are determined to obtain the third Gini coefficient and the fourth Gini coefficient respectively. The change in the Gini coefficient can be determined based on the second Gini coefficient, the third Gini coefficient and the fourth Gini coefficient. If the change in the Gini coefficient is a positive value, it means that the purity of the data set after segmentation is improved, indicating that the segmentation is beneficial and reduces uncertainty. The importance of the second influencing factor can be determined based on the change in the Gini coefficient, wherein the calculation formula for the change in the Gini coefficient is as follows:

[0087]

[0088] In the formula, It represents the change in the Gini coefficient of the target node q of the i-th decision tree before and after branching based on the second influencing factor j, that is, the importance of the second influencing factor on the target node q of the i-th decision tree; Represents the second Gini coefficient of the target node q of the i-th decision tree; represents the third Gini coefficient of the third node l of the i-th decision tree; Represents the fourth Gini coefficient of the fourth node r of the i-th decision tree.

[0089] For example, if the second influencing factor is the average temperature, the third node of the target node's child node is a high temperature group node (average temperature higher than or equal to 25°C), and the fourth node is a low temperature group node (average temperature lower than 25°C). For the target node, its second Gini coefficient is 0.45, the third Gini coefficient of the third node is 0.35, and the fourth Gini coefficient of the fourth node is 0.3. The change in the Gini coefficient is -0.2. When the change in the Gini coefficient is negative, it means that after the decision tree is segmented according to the second influencing factor, the overall purity of the data set has not improved. Therefore, the segmentation effect of this second influencing factor in this tree is not very ideal.

[0090] If the current decision tree is divided multiple times based on the second influencing factor, the node set where the second influencing factor appears in the current decision tree is Q. Then the importance of the second influencing factor in the current decision tree can be calculated by the following formula:

[0091]

[0092] In the formula, It represents the change in the Gini coefficient of all target nodes of the i-th decision tree before and after branching based on the second influencing factor j, that is, the importance of the second influencing factor in the i-th decision tree; Q represents the set of nodes where the second influencing factor appears in the i-th decision tree.

[0093] Generally speaking, the leaf nodes of a decision tree are generally not further split. Therefore, the Gini coefficient of a leaf node can be 0, indicating that all samples of the node belong to the same category, that is, the purity is the highest.

[0094] Furthermore, since there are w decision trees, the importance of the second influencing factor in the w decision trees can be calculated, that is, it can be calculated according to the following formula:

[0095]

[0096] In the formula, w j is the importance of the second influencing factor j; w means there are w decision trees.

[0097] In a specific embodiment, by comprehensively considering the situations of w decision trees, the importance of each of the d influencing factors can be calculated to obtain d importances, thereby making the importance index more comprehensive and objective, and avoiding the impact of the one-sidedness or overfitting problems that may exist in a single decision tree on the feature importance evaluation.

[0098] Optionally, the above step A304, determining the similarity between each of the f dates and the target future date according to the d importances to obtain f similarities, may include the following steps:

[0099] E301. Obtain a target date; the target date is any one of the f dates;

[0100] E302. Determine the sample corresponding to the target date to obtain the target sample;

[0101] E303, obtaining the impact factor data corresponding to each of the d impact factors of the target sample, and obtaining d third impact factor data;

[0102] E304. Standardize the d third impact factor data to obtain d third standard impact factor data;

[0103] E305, determining the sample corresponding to the target future date, and obtaining the target sample to be tested;

[0104] E306. Obtain the impact factor data corresponding to each of the d impact factors of the target sample to be tested, and obtain d fourth impact factor data;

[0105] E307, standardizing the d fourth impact factor data to obtain d fourth standard impact factor data;

[0106] E308. Determine the similarity between the target date and the target future date based on the following formula according to the d importances, the d third standard influencing factor data and the d fourth standard influencing factor data:

[0107]

[0108] Among them, dis mn represents the weighted Euclidean distance between the target sample m and the target sample n; d represents the number of influencing factors; w j represents the importance of the jth impact factor; x mj represents the data of the jth influencing factor of the target sample m, that is, the third standard influencing factor data; x nj The data representing the j-th influencing factor of the target sample n to be tested is the fourth standard influencing factor data.

[0109] In a specific embodiment, a target date is obtained, wherein the target date is any date among f dates, a sample corresponding to the target date is determined to obtain a target sample, influence factor data corresponding to each of d influence factors of the target sample is obtained, d third influence factor data are obtained, and the d third influence factor data are standardized to obtain d third standard influence factor data, wherein the standardization method can use Max-Min standardization. A sample corresponding to a target future date is determined to obtain a target sample to be tested, influence factor data corresponding to each of d influence factors of the target sample to be tested is obtained, d fourth influence factor data are obtained, and the d fourth influence factor data are standardized to obtain d fourth standard influence factor data.

[0110] Furthermore, the similarity between the target date and the target future date can be determined based on d importances, d third standard influencing factor data and d fourth standard influencing factor data, wherein the smaller the calculated weighted Euclidean distance is, the higher the similarity between the target date and the target future date is. In order to quantify the similarity, the weighted Euclidean distance can be inversely calculated. By accurately quantifying the similarity between the target date and the target future date, dates with high similarity to the target future date can be screened out from historical data as similar dates.

[0111] S104: Obtain a load sequence within the preset historical time period to obtain a target load sequence.

[0112] In a specific embodiment, the load data at each time of each day in a preset historical time period can be obtained from the database of the power company or the data query website to obtain a historical load sequence, i.e., a target load sequence, wherein the horizontal axis of the target load sequence is time and the vertical axis is the load value. The target load sequence can be used to analyze the changing pattern of the power load, explore the factors affecting the load, and predict the future power load.

[0113] S105. Decomposing the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by using a variational mode decomposition method.

[0114] The high-frequency component, the medium-frequency component and the low-frequency component are all time series, and the time series takes time as the horizontal axis and the load size as the vertical axis.

[0115] Among them, the variational modal decomposition method is a signal processing method, which decomposes the original signal into modal components with different center frequencies based on the variational principle, and performs well in processing non-stationary and nonlinear signals. In the embodiment of the present application, the power load sequence can be decomposed by the variational modal decomposition method.

[0116] In a specific embodiment, initial parameters of the variational modal decomposition method are set, such as the number of decomposed modes, penalty factor, and convergence criterion. In the embodiment of the present application, the variational modal decomposition method is used to decompose the target load sequence into three modal components within a specific frequency range.

[0117] It should be noted that the variational modal decomposition method constructs a constrained variational problem for the target load sequence, and introduces an extended Lagrangian expression to transform the constrained variational problem into an unconstrained variational problem. Then, the multiplication operator alternating direction method is used to solve it. After continuous iterations to meet the iteration conditions, multiple modal components are obtained. By classifying multiple modal components, high-frequency components, medium-frequency components and low-frequency components can be obtained respectively.

[0118] Among them, the high-frequency component reflects the rapid fluctuation information in the target load sequence, such as rapid changes in load caused by factors such as the random opening or closing of a large number of small electrical equipment in a short period of time and small-scale disturbances in the local power grid. Although these high-frequency fluctuations account for a relatively small proportion of the overall load, they may have a certain impact on the stability and power quality of the power system, such as causing voltage flicker, harmonics and other problems.

[0119] Among them, the medium-frequency component reflects the load change factors with a certain periodicity but relatively short period, such as the daily business peak and valley in the commercial area, the production shift switching of some medium-sized industrial enterprises, etc. By analyzing the characteristics of the medium-frequency component, it can help the relevant power dispatching departments to reasonably arrange the allocation of power resources in different areas and different time periods, and improve the operating efficiency of the power system. For example, before the commercial business peak arrives, power resources can be deployed to the commercial area in advance to ensure the adequacy and stability of power supply.

[0120] Among them, the low-frequency component reflects the long-term trend and major periodic changes in the target load sequence, such as seasonal changes and the impact of annual economic development trends on power load. The low-frequency component usually accounts for a large proportion of the overall load, and its changes are relatively slow but large in amplitude.

[0121] Optionally, the above step S105, decomposing the target load sequence into high-frequency components, medium-frequency components and low-frequency components by using a variational mode decomposition method, may include the following steps:

[0122] A501. Decomposing the target load sequence into k modal components by using the variational modal decomposition method; k is an integer greater than 3;

[0123] A502, obtaining the center frequency of each of the k modal components to obtain k center frequencies;

[0124] A503, multiplying the preset sampling frequency by each of the k center frequencies to obtain k component frequencies;

[0125] A504, determining that the modal component corresponding to the component frequency greater than or equal to the preset high-frequency threshold among the k component frequencies is the high-frequency component;

[0126] A505, determining that the modal component corresponding to the component frequency greater than or equal to the preset low-frequency threshold and less than the preset high-frequency threshold among the k component frequencies is the intermediate-frequency component;

[0127] A506. Determine that the modal component corresponding to the component frequency less than the preset low-frequency threshold among the k component frequencies is the low-frequency component.

[0128] The preset sampling frequency refers to the number of sampling points set per unit time when collecting power load data, that is, the number of times the power load value is recorded per second. For example, if load data is collected every 15 minutes, there are 96 data points in a day, and the preset sampling frequency is 96 times per day, that is, 1 / 900Hz.

[0129] The preset high-frequency threshold refers to the critical frequency value for distinguishing the high-frequency range to which the modal component belongs, and is used to determine whether the modal component is a high-frequency component. The high-frequency component of the high-frequency component usually corresponds to short-term fluctuations within a day. Therefore, the preset high-frequency threshold can be set to correspond to fluctuations within a few hours. For example, if the data shows that there is an obvious fluctuation every 2 hours, the preset high-frequency threshold can be set to correspond to the frequency of 2 hours. Specifically, f max =1 / (2×3600)=1.389×10 -4 Hz.

[0130] The preset low-frequency threshold refers to the critical frequency value for distinguishing the low-frequency range to which the modal component belongs, and is used to determine whether the modal component is a low-frequency component. The low-frequency component of the low-frequency component usually corresponds to the long-term trend within a day. For example, in order to extract the main trend of the daily load curve, the preset low-frequency threshold can be set to the frequency corresponding to 24 hours. Specifically, f min =1 / (24×3600)=1.157×10 -5 Hz.

[0131] In a specific embodiment, the target load sequence is decomposed into k modal components by a variational modal decomposition method, wherein k is an integer greater than 3, and the k modal components are evaluated and classified to determine three main components, namely, high-frequency components, medium-frequency components, and low-frequency components. Specifically, the center frequency of each modal component in the k modal components can be obtained to obtain k center frequencies, wherein the center frequency of each modal component can calculate the bandwidth of the modal component, i.e., the difference between the maximum frequency and the minimum frequency, and the bandwidth is divided by 2 to obtain the center frequency of the modal component. The center frequency of each modal component can also be solved by iteratively optimizing the variational problem. The variational modal decomposition method can decompose the target load sequence into multiple modal components with dimensionless center frequencies, and therefore, the center frequency of each modal component can also be directly obtained.

[0132] Furthermore, the preset sampling frequency is multiplied by each of the k center frequencies to obtain k component frequencies. The modal component corresponding to the component frequency greater than or equal to the preset high-frequency threshold among the k component frequencies can be determined as a high-frequency component, the modal component corresponding to the component frequency greater than or equal to the preset low-frequency threshold and less than the preset high-frequency threshold among the k component frequencies can be determined as a medium-frequency component, and the modal component corresponding to the component frequency less than the preset low-frequency threshold among the k component frequencies can be determined as a low-frequency component. By decomposing the target load sequence into high-frequency, medium-frequency and low-frequency components through variational modal decomposition, the changing characteristics of the power load can be accurately understood from different frequency levels, providing strong support for the stable operation, optimized scheduling and long-term planning of the power system.

[0133] S106. Determine a forecast load sequence for the target future date according to the p similar dates, the high-frequency component, the medium-frequency component, and the low-frequency component.

[0134] In a specific embodiment, different methods are used to predict the decomposed high-frequency component, medium-frequency component and low-frequency component. For example, the high-frequency component can be predicted by the similar day average method, the medium-frequency component and the low-frequency component can be predicted using the LSTM neural network, the medium-frequency component is predicted by taking the first few days of the target future date as the input of the LSTM neural network, and the low-frequency component is predicted by taking the similar dates of the target future date as the input of the LSTM neural network, and finally the prediction results of the three components are superimposed to obtain the power load prediction sequence of the target future date. Specifically, the predicted load sequence of the target future date can be determined according to p similar dates, high-frequency components, medium-frequency components and low-frequency components. Since the changing laws of high-frequency, medium-frequency and low-frequency components are very different, the use of a unified prediction model may not be able to adapt to all these laws well, so making full use of the time series characteristics of different components for prediction can improve the accuracy of the prediction.

[0135] Optionally, the above step S106, determining the forecast load sequence of the target future date according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component, may include the following steps:

[0136] A601, averaging the load data corresponding to each similar date of the p similar dates and the high frequency component to obtain a first load sequence;

[0137] A602. Determine a similar dates among the p similar dates; a is a positive integer less than or equal to p;

[0138] A603. Obtain the date b days before the target future date, and obtain b historical dates; b is an integer greater than 1;

[0139] A604, obtaining load data corresponding to each historical date of the b historical dates and the intermediate frequency component, to obtain a first load data set;

[0140] A605. Obtain the impact factor data corresponding to the d impact factors to obtain an impact factor data set;

[0141] A606. Input the first load data set and the influencing factor data set into a preset deep learning model to obtain a second load sequence;

[0142] A607, obtaining load data corresponding to each similar date of the a similar dates and the low-frequency component, to obtain a second load data set;

[0143] A608. Input the second load data set into the preset deep learning model to obtain a third load sequence;

[0144] A609. Determine the predicted load sequence based on the first load sequence, the second load sequence and the third load sequence.

[0145] Among them, the preset deep learning model can be an LSTM neural network model, and different modal components are predicted using preset deep learning models trained with different training sets. For example, when using the preset deep learning model to predict the low-frequency component, its training set is usually the load data of similar dates, and when using the preset deep learning model to predict the medium-frequency component, its training set is usually the historical dates of a few days before the target future date and the influencing factor data of the target future date.

[0146] In a specific embodiment, the load data corresponding to the high-frequency component of each similar date among p similar dates are averaged to obtain a first load sequence; a similar dates among p similar dates are determined; a is a positive integer less than or equal to p; the date b days before the target future date is obtained to obtain b historical dates; b is an integer greater than 1; the load data corresponding to the medium-frequency component of each historical date among the b historical dates is obtained to obtain a first load data set; the influencing factor data corresponding to d influencing factors are obtained to obtain an influencing factor data set; the first load data set and the influencing factor data set are input into a preset deep learning model to obtain a second load sequence; the load data corresponding to the low-frequency component of each similar date among a similar dates is obtained to obtain a second load data set; the second load data set is input into a preset deep learning model to obtain a third load sequence; and a predicted load sequence is determined according to the first load sequence, the second load sequence and the third load sequence. Using the similar day average method to predict the high-frequency components can avoid the overfitting problem of the prediction model and improve the prediction efficiency and stability of the high-frequency part. The medium-frequency and low-frequency components are predicted by constructing a prediction model, combining the modeling capabilities of the deep learning model and the high efficiency of the similar day method to effectively enhance the robustness and prediction accuracy of the model.

[0147] In one possible embodiment, when making predictions for high-frequency components, there are 10 similar dates, and the target future date for prediction is a holiday, then the 10 similar dates are basically holidays, and the load data of each of the 10 similar dates at each moment are averaged to obtain the first load sequence.

[0148] In one possible embodiment, when predicting the intermediate frequency component, the dates three days before the target future date are obtained to obtain three historical dates, and the influencing factor data of the target future date is obtained at the same time. The load data of the three historical dates and the influencing factor data of the target future date are used as inputs to the preset deep learning model, and the second load sequence can be obtained by inputting them into the preset deep learning model.

[0149] In a possible embodiment, when predicting the low-frequency component, 3 similar dates are selected from 10 similar dates, and the load data of the 3 similar dates are used as inputs of a preset deep learning model. The third load sequence can be obtained by inputting the load data into the preset deep learning model.

[0150] Optionally, the above step A608, determining the predicted load sequence according to the first load sequence, the second load sequence and the third load sequence, may include the following steps:

[0151] The first load sequence, the second load sequence and the third load sequence are superimposed to obtain the predicted load sequence.

[0152] In a specific embodiment, the first load sequence, the second load sequence and the third load sequence respectively represent load changes in different time periods. Therefore, by superimposing the first load sequence, the second load sequence and the third load sequence at corresponding moments, a predicted load sequence for a target future date can be obtained.

[0153] In summary, by implementing the embodiments of the present application, the target future date for power load forecasting is determined, and d influencing factors affecting power load forecasting are determined. According to the d influencing factors, a date similar to the target future date is selected from a preset historical time period to obtain p similar dates, and a load sequence within the preset historical time period is obtained to obtain a target load sequence. The target load sequence is decomposed into high-frequency components, medium-frequency components, and low-frequency components by a variational mode decomposition method, and the predicted load sequence of the target future date is determined according to p similar dates, high-frequency components, medium-frequency components, and low-frequency components. It realizes the prediction of power load of complex nonlinear data and non-periodic factors by comprehensively considering multiple influencing factors, improves the accuracy of power load prediction, overcomes the shortcomings of traditional statistical models and machine learning models in dealing with complex situations, provides more accurate decision-making basis for power system operation and planning, and enhances the economy, safety, and stability of the power system.

[0154] See also Figure 2 , Figure 2 2 is a schematic diagram of the structure of a power load forecasting device provided in an embodiment of the present application. The power load forecasting device 200 includes: a data initialization module 201, a date selection module 202, a sequence decomposition module 203, and a load forecasting module 204, wherein:

[0155] The data initialization module 201 is used to determine the target future date;

[0156] The data initialization module 201 is also used to determine d influencing factors; d is a positive integer;

[0157] The date selection module 202 is used to select a date similar to the target future date from a preset historical time period according to the d influencing factors, to obtain p similar dates; p is a positive integer;

[0158] The data initialization module 201 is also used to obtain the load sequence within the preset historical time period to obtain the target load sequence;

[0159] The sequence decomposition module 203 is used to decompose the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by using a variational mode decomposition method; the high-frequency component, the medium-frequency component and the low-frequency component are all time series;

[0160] The load forecasting module 204 is used to determine the forecast load sequence of the target future date according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component.

[0161] Optionally, in terms of selecting a date similar to the target future date from a preset historical time period according to the d influencing factors to obtain p similar dates, the date selection module 202 is further specifically configured to:

[0162] The preset historical time period is divided into f dates based on the preset date granularity; f is an integer greater than p;

[0163] Obtain samples corresponding to each of the f dates to obtain a target sample set; the samples include load data and impact factor data;

[0164] Determine the importance of each of the d influencing factors according to the target sample set, and obtain d importances;

[0165] Determine the similarity between each of the f dates and the target future date according to the d importances, and obtain f similarities;

[0166] Selecting similarities greater than or equal to a preset similarity threshold from the f similarities to obtain p similarities;

[0167] The p similarity corresponding dates are determined as the p similar dates.

[0168] Optionally, in determining the importance of each of the d influencing factors according to the target sample set to obtain d importances, the date selection module 202 is further specifically configured to:

[0169] A decision tree is constructed according to the target sample set and the d influencing factors to obtain w decision trees; each decision tree includes g influencing factors; w is a positive integer; g is a positive integer less than or equal to d;

[0170] Determine the importance of each of the d influencing factors in the w decision trees to obtain the d importances.

[0171] Optionally, in the aspect of constructing a decision tree according to the target sample set and the d influencing factors to obtain w decision trees, the date selection module 202 is further specifically used for:

[0172] Randomly and non-repetitively select the g influencing factors corresponding to the first decision tree from the d influencing factors; the first decision tree is any decision tree among the w decision trees;

[0173] Determine the Gini coefficient when each of the g influencing factors divides the target sample set, and obtain g Gini coefficients;

[0174] Select the smallest Gini coefficient among the g Gini coefficients to obtain a first Gini coefficient;

[0175] Determine an impact factor corresponding to the first Gini coefficient to obtain a first impact factor;

[0176] Dividing the target sample set according to the first influencing factor to obtain a first node and a second node;

[0177] When the first node and the second node do not meet the preset first condition, dividing the first node and the second node respectively based on the g impact factors;

[0178] When the first node and the second node meet the preset first condition, the first decision tree is constructed according to the first node and the second node.

[0179] Optionally, in determining the importance of each of the d influencing factors in the w decision trees to obtain the d importances, the date selection module 202 is further specifically configured to:

[0180] Obtain a second impact factor; the second impact factor is any impact factor among the d impact factors;

[0181] Obtain a target node; the target node is any node in the w decision trees that is divided based on the second influencing factor;

[0182] Get the child nodes of the target node and get the third and fourth nodes;

[0183] Determine the Gini coefficient of the target node to obtain a second Gini coefficient;

[0184] Determine the Gini coefficients of the third node and the fourth node to obtain a third Gini coefficient and a fourth Gini coefficient respectively;

[0185] Determine a change in the Gini coefficient according to the second Gini coefficient, the third Gini coefficient, and the fourth Gini coefficient;

[0186] The importance of the second influencing factor is determined according to the change in the Gini coefficient.

[0187] Optionally, in determining the similarity between each of the f dates and the target future date according to the d importances to obtain f similarities, the date selection module 202 is further specifically configured to:

[0188] Obtain a target date; the target date is any date among the f dates;

[0189] Determine the sample corresponding to the target date to obtain the target sample;

[0190] Obtaining the impact factor data corresponding to each of the d impact factors of the target sample to obtain d third impact factor data;

[0191] Standardizing the d third impact factor data to obtain d third standard impact factor data;

[0192] Determine the sample corresponding to the target future date to obtain the target sample to be tested;

[0193] Obtaining the impact factor data corresponding to each of the d impact factors of the target sample to be tested, and obtaining d fourth impact factor data;

[0194] Standardizing the d fourth impact factor data to obtain d fourth standard impact factor data;

[0195] The similarity between the target date and the target future date is determined based on the following formula according to the d importances, the d third standard influencing factor data and the d fourth standard influencing factor data:

[0196]

[0197] Among them, dis mn represents the weighted Euclidean distance between the target sample m and the target sample n; d represents the number of influencing factors; w j represents the importance of the jth impact factor; x mj represents the data of the jth influencing factor of the target sample m, that is, the third standard influencing factor data; x nj The data representing the j-th influencing factor of the target sample n to be tested is the fourth standard influencing factor data.

[0198] Optionally, in the aspect of decomposing the target load sequence into high-frequency components, medium-frequency components and low-frequency components by using a variational mode decomposition method, the sequence decomposition module 203 is further specifically used for:

[0199] Decomposing the target load sequence into k modal components by the variational modal decomposition method; k is an integer greater than 3;

[0200] Obtaining the center frequency of each of the k modal components to obtain k center frequencies;

[0201] Multiplying a preset sampling frequency by each of the k center frequencies to obtain k component frequencies;

[0202] Determine that the modal component corresponding to the component frequency greater than or equal to the preset high-frequency threshold among the k component frequencies is the high-frequency component;

[0203] Determine that the modal component corresponding to the component frequency greater than or equal to the preset low-frequency threshold and less than the preset high-frequency threshold among the k component frequencies is the intermediate-frequency component;

[0204] Determine that a modal component corresponding to a component frequency less than the preset low-frequency threshold among the k component frequencies is the low-frequency component.

[0205] Optionally, in determining the predicted load sequence of the target future date according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component, the load prediction module 204 is further specifically configured to:

[0206] Averaging the load data corresponding to each similar date of the p similar dates and the high frequency component to obtain a first load sequence;

[0207] Determine a similar dates among the p similar dates; a is a positive integer less than or equal to p;

[0208] Get the date b days before the target future date, and obtain b historical dates; b is an integer greater than 1;

[0209] Obtaining load data corresponding to each historical date of the b historical dates and the intermediate frequency component to obtain a first load data set;

[0210] Obtaining the impact factor data corresponding to the d impact factors to obtain an impact factor data set;

[0211] Inputting the first load data set and the influencing factor data set into a preset deep learning model to obtain a second load sequence;

[0212] Obtaining load data corresponding to the low-frequency component on each similar date of the a similar dates to obtain a second load data set;

[0213] Inputting the second load data set into the preset deep learning model to obtain a third load sequence;

[0214] The predicted load sequence is determined based on the first load sequence, the second load sequence and the third load sequence.

[0215] Optionally, in determining the predicted load sequence according to the first load sequence, the second load sequence and the third load sequence, the load prediction module 204 is further specifically configured to:

[0216] The first load sequence, the second load sequence and the third load sequence are superimposed to obtain the predicted load sequence.

[0217] The power load forecasting device 200 described in the present application can determine the target future date for power load forecasting, and determine d influencing factors affecting power load forecasting, select a date similar to the target future date from a preset historical time period according to the d influencing factors, obtain p similar dates, obtain the load sequence within the preset historical time period, obtain the target load sequence, decompose the target load sequence into high-frequency components, medium-frequency components and low-frequency components by the variational mode decomposition method, and determine the predicted load sequence of the target future date according to the p similar dates, high-frequency components, medium-frequency components and low-frequency components. It realizes the prediction of power load of complex nonlinear data and non-periodic factors by comprehensively considering multiple influencing factors, improves the accuracy of power load prediction, overcomes the shortcomings of traditional statistical models and machine learning models in dealing with complex situations, provides more accurate decision-making basis for power system operation and planning, and enhances the economy, safety and stability of the power system.

[0218] See also Figure 3 , Figure 3 : is a structural diagram of an electronic device provided in an embodiment of the present application, the electronic device may include a processor, a memory, a communication interface and one or more programs, the processor, the memory and the communication interface may be interconnected through a bus; the one or more programs are stored in the memory and configured to be executed by the processor; in the embodiment of the present application, the program includes instructions for executing the following steps:

[0219] Determine target future dates;

[0220] Determine d influencing factors; d is a positive integer;

[0221] According to the d influencing factors, a date similar to the target future date is selected from a preset historical time period to obtain p similar dates; p is a positive integer;

[0222] Obtaining a load sequence within the preset historical time period to obtain a target load sequence;

[0223] Decomposing the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by a variational mode decomposition method; the high-frequency component, the medium-frequency component and the low-frequency component are all time series;

[0224] A forecast load sequence for the target future date is determined according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component.

[0225] The electronic device described in the present application can determine the target future date for power load forecasting, and determine d influencing factors affecting power load forecasting, select a date similar to the target future date from a preset historical time period according to the d influencing factors, obtain p similar dates, obtain the load sequence within the preset historical time period, obtain the target load sequence, decompose the target load sequence into high-frequency components, medium-frequency components and low-frequency components by the variational mode decomposition method, and determine the predicted load sequence of the target future date according to the p similar dates, high-frequency components, medium-frequency components and low-frequency components. It realizes the prediction of power load of complex nonlinear data and non-periodic factors by comprehensively considering multiple influencing factors, improves the accuracy of power load forecasting, overcomes the shortcomings of traditional statistical models and machine learning models in dealing with complex situations, provides more accurate decision-making basis for power system operation and planning, and enhances the economy, safety and stability of the power system.

[0226] An embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps of any method described in the above method embodiments, and the above computer includes an electronic device.

[0227] The embodiment of the present application also provides a computer program product, the computer program product includes a non-transitory computer-readable storage medium storing a computer program, the computer program is operable to cause a computer to execute some or all of the steps of any method described in the method embodiment. The computer program product may be a software installation package, and the computer includes an electronic device.

[0228] Those skilled in the art can understand that to implement all or part of the processes in the above-mentioned embodiments, the processes can be completed by computer programs to instruct related hardware, and the programs can be stored in computer-readable storage media. When the programs are executed, they can include the processes of the above-mentioned method embodiments. The aforementioned storage media include: ROM or random access memory RAM, magnetic disk or optical disk and other media that can store program codes.

[0229] The steps of the method or algorithm described in the embodiments of the present application can be implemented in hardware or by executing software instructions by a processor. The software instructions can be composed of corresponding software modules, and the software modules can be stored in RAM, flash memory, ROM, EPROM, electrically erasable programmable read-only memory (electrically EPROM, EEPROM), registers, hard disks, mobile hard disks, read-only compact disks (CD-ROMs) or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC. In addition, the ASIC can be located in a terminal device or a management device. Of course, the processor and the storage medium can also be present in a terminal device or a management device as discrete components.

[0230] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server, or data center to another website site, computer, server, or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital video disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0231] The modules / units included in the devices and products described in the above embodiments may be software modules / units or hardware modules / units, or may be partially software modules / units and partially hardware modules / units. For example, for the devices and products applied to or integrated in the chip, the modules / units included therein may all be implemented in the form of hardware such as circuits, or at least some of the modules / units may be implemented in the form of software programs, which run on the processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits; for the devices and products applied to or integrated in the chip module, the modules / units included therein may all be implemented in the form of hardware such as circuits, and different modules / units may be located in the same component (such as a chip, circuit module, etc.) or in different components of the chip module, or at least some of the modules / units may be implemented in the form of software programs. The software programs run on the processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits. It is implemented in the form of a software program, which runs on a processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in hardware such as circuits; for various devices and products applied to or integrated in the terminal equipment, the various modules / units contained therein can be implemented in hardware such as circuits, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or in different components in the terminal equipment, or, at least some modules / units can be implemented in the form of a software program, which runs on a processor integrated inside the terminal equipment, and the remaining (if any) modules / units can be implemented in hardware such as circuits.

[0232] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above description is only the specific implementation method of the embodiments of the present application and is not intended to limit the protection scope of the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A method for predicting power load, characterized in that: Applied to a server, the method comprises: Determine target future dates; Determine d influencing factors; d is a positive integer; According to the d influencing factors, a date similar to the target future date is selected from a preset historical time period to obtain p similar dates; p is a positive integer; Obtaining a load sequence within the preset historical time period to obtain a target load sequence; Decomposing the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by a variational mode decomposition method; the high-frequency component, the medium-frequency component and the low-frequency component are all time series; A forecast load sequence for the target future date is determined according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component.

2. The method according to claim 1, characterized in that The step of selecting a date similar to the target future date from a preset historical time period according to the d influencing factors to obtain p similar dates includes: The preset historical time period is divided into f dates based on the preset date granularity; f is an integer greater than p; Obtain samples corresponding to each of the f dates to obtain a target sample set; the samples include load data and impact factor data; Determine the importance of each of the d influencing factors according to the target sample set, and obtain d importances; Determine the similarity between each of the f dates and the target future date according to the d importances, and obtain f similarities; Selecting similarities greater than or equal to a preset similarity threshold from the f similarities to obtain p similarities; The p similarity corresponding dates are determined as the p similar dates.

3. The method according to claim 2, characterized in that Determining the importance of each of the d influencing factors according to the target sample set to obtain d importances includes: A decision tree is constructed according to the target sample set and the d influencing factors to obtain w decision trees; each decision tree includes g influencing factors; w is a positive integer; g is a positive integer less than or equal to d; Determine the importance of each of the d influencing factors in the w decision trees to obtain the d importances.

4. The method according to claim 3, characterized in that The step of constructing a decision tree according to the target sample set and the d influencing factors to obtain w decision trees includes: Randomly and non-repetitively select the g influencing factors corresponding to the first decision tree from the d influencing factors; the first decision tree is any decision tree among the w decision trees; Determine the Gini coefficient when each of the g influencing factors divides the target sample set, and obtain g Gini coefficients; Select the smallest Gini coefficient among the g Gini coefficients to obtain a first Gini coefficient; Determine an impact factor corresponding to the first Gini coefficient to obtain a first impact factor; Dividing the target sample set according to the first influencing factor to obtain a first node and a second node; When the first node and the second node do not meet the preset first condition, dividing the first node and the second node respectively based on the g impact factors; When the first node and the second node meet the preset first condition, the first decision tree is constructed according to the first node and the second node.

5. The method according to claim 3 or 4, characterized in that The determining the importance of each of the d influencing factors in the w decision trees to obtain the d importances includes: Obtain a second impact factor; the second impact factor is any impact factor among the d impact factors; Obtain a target node; the target node is any node in the w decision trees that is divided based on the second influencing factor; Get the child nodes of the target node and get the third and fourth nodes; Determine the Gini coefficient of the target node to obtain a second Gini coefficient; Determine the Gini coefficients of the third node and the fourth node to obtain a third Gini coefficient and a fourth Gini coefficient respectively; Determine a change in the Gini coefficient according to the second Gini coefficient, the third Gini coefficient, and the fourth Gini coefficient; The importance of the second influencing factor is determined according to the change in the Gini coefficient.

6. The method according to claim 2, characterized in that Determining the similarity between each of the f dates and the target future date according to the d importances to obtain f similarities includes: Obtain a target date; the target date is any date among the f dates; Determine the sample corresponding to the target date to obtain the target sample; Obtaining the impact factor data corresponding to each of the d impact factors of the target sample to obtain d third impact factor data; Standardizing the d third impact factor data to obtain d third standard impact factor data; Determine the sample corresponding to the target future date to obtain the target sample to be tested; Obtaining the impact factor data corresponding to each of the d impact factors of the target sample to be tested, and obtaining d fourth impact factor data; Standardizing the d fourth impact factor data to obtain d fourth standard impact factor data; The similarity between the target date and the target future date is determined based on the following formula according to the d importances, the d third standard influencing factor data and the d fourth standard influencing factor data: Among them, dis mn represents the weighted Euclidean distance between the target sample m and the target sample n; d represents the number of influencing factors; w j represents the importance of the jth impact factor; x mj represents the data of the jth influencing factor of the target sample m, that is, the third standard influencing factor data; x nj The data representing the j-th influencing factor of the target sample n to be tested is the fourth standard influencing factor data.

7. The method according to any one of claims 1 to 3, characterized in that: Decomposing the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by a variational mode decomposition method includes: Decomposing the target load sequence into k modal components by the variational modal decomposition method; k is an integer greater than 3; Obtaining the center frequency of each of the k modal components to obtain k center frequencies; Multiplying a preset sampling frequency by each of the k center frequencies to obtain k component frequencies; Determine that the modal component corresponding to the component frequency greater than or equal to the preset high-frequency threshold among the k component frequencies is the high-frequency component; Determine that the modal component corresponding to the component frequency greater than or equal to the preset low-frequency threshold and less than the preset high-frequency threshold among the k component frequencies is the intermediate-frequency component; Determine that a modal component corresponding to a component frequency less than the preset low-frequency threshold among the k component frequencies is the low-frequency component.

8. The method according to any one of claims 1 to 7, characterized in that: The step of determining the forecast load sequence of the target future date according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component comprises: Averaging the load data corresponding to each similar date of the p similar dates and the high frequency component to obtain a first load sequence; Determine a similar dates among the p similar dates; a is a positive integer less than or equal to p; Get the date b days before the target future date, and obtain b historical dates; b is an integer greater than 1; Obtaining load data corresponding to each historical date of the b historical dates and the intermediate frequency component to obtain a first load data set; Obtaining the impact factor data corresponding to the d impact factors to obtain an impact factor data set; Inputting the first load data set and the influencing factor data set into a preset deep learning model to obtain a second load sequence; Obtaining load data corresponding to the low-frequency component on each similar date of the a similar dates to obtain a second load data set; Inputting the second load data set into the preset deep learning model to obtain a third load sequence; The predicted load sequence is determined based on the first load sequence, the second load sequence and the third load sequence.

9. The method according to claim 8, characterized in that The determining the predicted load sequence according to the first load sequence, the second load sequence and the third load sequence comprises: The first load sequence, the second load sequence and the third load sequence are superimposed to obtain the predicted load sequence.

10. A power load forecasting device, characterized in that: Applied to a server, the device comprises: a data initialization module, a date selection module, a sequence decomposition module, and a load prediction module, wherein: The data initialization module is used to determine the target future date; The data initialization module is also used to determine d influencing factors; d is a positive integer; The date selection module is used to select a date similar to the target future date from a preset historical time period according to the d influencing factors, to obtain p similar dates; p is a positive integer; The data initialization module is also used to obtain the load sequence within the preset historical time period to obtain the target load sequence; The sequence decomposition module is used to decompose the target load sequence into a high-frequency component, a medium-frequency component and a low-frequency component by using a variational mode decomposition method; the high-frequency component, the medium-frequency component and the low-frequency component are all time series; The load forecasting module is used to determine the forecast load sequence of the target future date according to the p similar dates, the high frequency component, the medium frequency component and the low frequency component.

Images