Method for predicting power load of power grid, storage medium, and program product

Abstract

The invention provides a prediction method of a power load of a power grid, a storage medium and a program product. The prediction method comprises the steps of obtaining period power load data of the power grid, period meteorological data of an area where the power grid is located and corresponding time information; obtaining data features based on the period power load data, the period meteorological data and the time information; constructing a plurality of power load prediction models based on the data features; and predicting the power load at the target time by using the power load prediction model corresponding to the target time in the plurality of power load prediction models to obtain the predicted power load at the target time. The problem that it is difficult to obtain an accurate power load prediction result is solved, more accurate power load prediction can be achieved, guidance can be provided for power generation prediction of a power grid, the effectiveness of power grid plan arrangement is improved, and safe and stable operation of the power grid is guaranteed.

Description

Technical Field

[0001] The present disclosure relates to the field of wind power generation, and more specifically, to a method for predicting power load of a power grid, a storage medium, and a program product. Background Art

[0002] With the continuous growth of electricity demand and the development of power technology, the sustainable development of power systems and energy management planning have become the focus of global attention. The increase in electricity consumption has brought a huge absorption burden to the power grid. In order to promote the safe and stable operation of the power grid, higher requirements are put forward for the accurate prediction of power load.

[0003] In the related art, the power load prediction method generally adopts a unified model, which does not consider the impact of factors such as weather and special dates during the predicted time period on the power grid's power generation load. Therefore, it is difficult to obtain accurate power load prediction results. Summary of the invention

[0004] In view of the problem that it is difficult to obtain accurate power load prediction results in the related art, the present disclosure provides a method for predicting the power load of a power grid, a storage medium, and a program product.

[0005] A first aspect of the present disclosure provides a method for predicting the power load of a power grid, the prediction method comprising: obtaining period power load data of the power grid to be predicted, period meteorological data of the area where the power grid to be predicted is located, and time information corresponding to the period power load data and the period meteorological data; obtaining data features based on the period power load data, the period meteorological data and the time information, wherein the data features include power load coding features, meteorological features and time features; constructing multiple power load prediction models based on the data features, wherein the multiple power load prediction models correspond to different prediction time intervals; using the power load prediction model corresponding to the target time among the multiple power load prediction models, predicting the power load at the target time, and obtaining the predicted power load at the target time.

[0006] Optionally, the period meteorological data includes period temperature data, wherein the power load coding feature is obtained in the following manner: based on the period power load data and the period temperature data, a first coding fitting relationship between power load and temperature is obtained; and the period temperature data is substituted into the first coding fitting relationship to obtain the power load coding feature.

[0007] Optionally, the period meteorological data includes period temperature data, wherein the power load coding characteristics are obtained in the following manner: according to a preset unit time, the period temperature data and the period power load data are respectively aggregated to obtain temperature aggregation characteristics and load aggregation characteristics; based on the temperature aggregation characteristics and the load aggregation characteristics, a second coding fitting relationship between power load and temperature is obtained; the temperature aggregation characteristics are substituted into the second coding fitting relationship to obtain the power load coding characteristics.

[0008] Optionally, the power load levels of the prediction time intervals corresponding to the multiple power load prediction models are different, and the multiple power load prediction models include a normal load model, a temperature load model and a node load model, wherein the temperature load model is used to predict the power load in a first time interval corresponding to a preset temperature range, the node load model is used to predict the power load in a second time interval including a preset time node, and the normal load model is used to predict the power load in other time intervals except the first time interval and the second time interval.

[0009] Optionally, the normal load model is constructed in the following manner: based on model parameters, a machine learning model to be trained is configured, wherein the model parameters are obtained by performing hyperparameter optimization on training results obtained in previous training of the machine learning model; based on the data features, the machine learning model is trained to obtain a trained machine learning model; and the trained machine learning model is used as the normal load model.

[0010] Optionally, the temperature load model is constructed in the following manner: based on the data features, a plurality of first sample features are screened out, wherein the plurality of first sample features include data features corresponding to the first time interval and data features corresponding to time intervals adjacent to the first time interval; based on the temperature features and power load coding features in each first sample feature, a first modeling fitting relationship between the temperature and the power load corresponding to each first sample feature is obtained respectively; based on the first modeling fitting relationship corresponding to each first sample feature, a temperature basic load part corresponding to each first sample feature and a temperature influence load part corresponding to any first sample feature are obtained, wherein the temperature basic load part is the part of the first modeling fitting relationship whose derivative is a first specific value, and the temperature influence load part is the difference between the derivative of the first modeling fitting relationship and the temperature basic load part; on the basis of the normal load model, the temperature influence load part and the temperature basic load part of each first sample feature are introduced to construct the temperature load model.

[0011] Optionally, the node load model is constructed in the following manner: based on the data characteristics, a second sample characteristic corresponding to the second time interval is screened out; based on the temperature characteristics and the power load coding characteristics in the second sample characteristics, a second modeling fitting relationship between temperature and power load is obtained; based on the second modeling fitting relationship at the target temperature, the node basic load part is determined, wherein the target temperature is the temperature that makes the derivative of the second modeling fitting relationship a second specific value; based on the derivative of the second modeling fitting relationship, the temperature load part is determined; based on the time characteristics corresponding to the preset time node in the second sample characteristics, the node influence load part is determined; and the node load model is constructed based on the node basic load part, the node influence load part and the temperature load part.

[0012] Optionally, before executing the step of obtaining data features based on the power load data for the period, the meteorological data for the period and the time information, the prediction method further includes: performing data repair on the power load data for the period or the meteorological data for the period, wherein the data repair is performed in the following manner: decomposing the time series of the data to be repaired to obtain multiple subsequences, wherein the multiple subsequences include basic trend part subsequences and period part subsequences; in response to a target subsequence having missing data among the multiple subsequences, generating derived features based on the target subsequence, wherein the derived features are data extracted from the subsequence based on the time features; predicting the missing data in the target subsequence based on the target subsequence and the derived features, and filling the target subsequence with the predicted data value; merging the filled target subsequence with a subsequence without missing data among the multiple subsequences to obtain repaired data.

[0013] A second aspect of the present disclosure provides a computer-readable storage medium. When instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute the method for predicting the power load of the power grid according to the present disclosure.

[0014] A third aspect of the present disclosure provides a computer program product, which includes computer instructions. When the computer instructions are executed by a processor, the method for predicting the power load of the power grid according to the present disclosure is implemented.

[0015] According to the power load prediction method of the power grid disclosed in the present invention, corresponding data characteristics can be obtained based on period power load data, period meteorological data and time information, so as to construct a power load prediction model corresponding to different prediction time intervals, so that the model can take into account the meteorological characteristics and time characteristics of the time to be predicted when predicting the power load of the power grid, thereby achieving more accurate power load prediction, providing guidance for the power generation prediction of the power grid, improving the effectiveness of the power grid planning and arrangement, and ensuring the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Detailed description is a flow chart showing a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0017] Figure 2 1 is a schematic diagram showing the overall flow of a prediction system of a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0018] Figure 3 1 is a flow chart showing data anomaly detection in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0019] Figure 4 is a schematic diagram illustrating steps of performing data repair in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0020] Figure 5 The present invention is a flowchart illustrating a method for filling missing values ​​in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0021] Figure 6 1 is a flow chart showing a time series modeling and repairing step in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0022] Figure 7 1 is a flow chart showing feature generation in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0023] Figure 8 1 is a flowchart illustrating an example of constructing a normal load model in a method for predicting electric power load of a power grid according to an exemplary embodiment of the present disclosure.

[0024] Fig. 9 is a schematic diagram illustrating a step of constructing a temperature load model in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0025] Fig.101 is a flowchart illustrating an example of constructing a temperature load model in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0026] Fig.11 is a schematic diagram illustrating the step of constructing a node load model in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0027] Fig.12 1 is a flowchart illustrating an example of constructing a node load model in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure.

[0028] Fig.13 1 is a flow chart showing base load decomposition in a method for predicting electric load of a power grid according to an exemplary embodiment of the present disclosure.

[0029] Fig.14 and Fig.15 is a schematic diagram showing prediction results obtained by using a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0030] The following specific embodiments are provided to help the reader obtain a comprehensive understanding of the methods, devices and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications and equivalents of the methods, devices and / or systems described herein will be clear. For example, the order of operations described herein is only an example and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art may be omitted.

[0031] The features described herein can be implemented in different forms and should not be construed as being limited to the examples described herein. Rather, the examples described herein have been provided to illustrate only some of the many possible ways to implement the methods, devices, and / or systems described herein, which will be clear after understanding the disclosure of the present application.

[0032] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more.

[0033] Although terms such as "first", "second", and "third" may be used herein to describe various members, components, regions, layers, or portions, these members, components, regions, layers, or portions should not be limited by these terms. Instead, these terms are only used to distinguish one member, component, region, layer, or portion from another member, component, region, layer, or portion. Therefore, without departing from the teachings of the examples described herein, the first member, first component, first region, first layer, or first portion referred to in the examples may also be referred to as the second member, second component, second region, second layer, or second portion.

[0034] In the specification, when an element (such as a layer, a region, or a substrate) is described as being “on”, “connected to”, or “coupled to” another element, the element may be directly “on”, “connected to”, or “coupled to” another element, or one or more other elements may be present therebetween. Conversely, when an element is described as being “directly on”, “directly connected to”, or “directly coupled to” another element, there may be no other elements present therebetween.

[0035] The terms used herein are only used to describe various examples and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. The terms "comprise", "include" and "have" indicate the presence of the described features, quantities, operations, components, elements and / or combinations thereof, but do not exclude the presence or addition of one or more other features, quantities, operations, components, elements and / or combinations thereof.

[0036] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by a person of ordinary skill in the art to which the present disclosure belongs after understanding the present disclosure. Unless explicitly defined as such herein, terms (such as those defined in a general dictionary) should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and should not be interpreted in an idealized or overly formal manner.

[0037] Furthermore, in the description of examples, when it is considered that a detailed description of a well-known related structure or function would cause vague interpretation of the present disclosure, such a detailed description will be omitted.

[0038] As mentioned above, in the related art, since the influence of factors such as weather and date on the power generation load of the power grid is not considered, it is difficult to obtain accurate power load prediction results.

[0039] Specifically, a variety of methods and techniques have been applied in the field of electricity consumption forecasting and power load forecasting. For example, in statistical methods, regression analysis and time series analysis can be used for modeling; in machine learning methods, support vector machines, decision trees, neural networks and other methods can be used for modeling; hybrid model methods can also be used, such as combining statistical methods and machine learning methods to complement each other with their respective advantages.

[0040] However, none of these methods take into account the relationship between power load and weather and date. For example, the meteorological data such as temperature, rainfall, irradiance, etc. are different in different seasons, which will lead to different power loads. For example, the power consumption levels at different time nodes such as holidays and working days are also different. The prediction methods in related technologies do not make targeted predictions for power loads in different situations, and thus cannot provide guidance for the power generation plan arrangement of the power grid. In addition, when predicting, the use of rolling prediction methods may lead to prediction results with accumulated errors, resulting in inaccurate predictions.

[0041] Therefore, there is an urgent need for a complete power load forecasting method that takes into account different scenarios.

[0042] In view of the above problems, the present disclosure provides a method for predicting power load of a power grid, a computer-readable storage medium, and a computer program product to solve or at least alleviate the above problems.

[0043] According to a first aspect of an exemplary embodiment of the present disclosure, a method for predicting the power load of a power grid is provided. The prediction method can be performed by an electronic device having computing capabilities. The electronic device can be, for example, a terminal device or a server, wherein the terminal device can be, for example, a tablet computer, a laptop computer, a digital assistant, etc.; the server can be an independent server, a server cluster, a cloud computing platform, or a virtualization center.

[0044] In an example application scenario, the electronic device can obtain the period power load data of the power grid to be predicted, the period meteorological data of the area where the power grid to be predicted is located, and the time information corresponding to the period power load data and the period meteorological data; based on the period power load data, the period meteorological data and the time information, data features can be obtained, wherein the data features include power load coding features, meteorological features and time features; based on the data features, multiple power load prediction models can be constructed, wherein the multiple power load prediction models correspond to different prediction time intervals; the power load prediction model corresponding to the target time in the multiple power load prediction models can be used to predict the power load at the target time to obtain the predicted power load at the target time.

[0045] Here, the electronic device can, for example, be communicatively connected to a measuring device, a control system or a data center of a wind turbine or wind farm, so as to obtain data required to execute the above method, such as period power load data, period meteorological data and time information.

[0046] The method for predicting the power load of the power grid according to the present disclosure can take into account the meteorological characteristics and time characteristics of the time to be predicted when predicting the power load of the power grid, thereby achieving more accurate power load prediction, providing guidance for the power generation prediction of the power grid, improving the effectiveness of the power grid planning and arrangement, and ensuring the safe and stable operation of the power grid.

[0047] The method for predicting the power load of a power grid according to the present disclosure may include the following steps:

[0048] like Figure 1 As shown, in step S110, the period power load data of the power grid to be predicted, the period meteorological data of the area where the power grid to be predicted is located, and the time information corresponding to the period power load data and the period meteorological data can be obtained.

[0049] Here, the period may refer to data of any period, for example, real data or predicted data of a certain period of time. The period power load data and period meteorological data of any period may be obtained, and the corresponding time information may be determined.

[0050] As an example, the power load may be the sum of the electric power taken from the power system by the electric devices in the power grid at a predetermined time, which may have a preset conversion relationship with the power consumption. The power load data may be determined based on the power data in the power grid. In the embodiment of the present disclosure, the power consumption at the corresponding time may be further predicted based on the prediction result of the power load. Meteorological data may include, for example, temperature data, precipitation data, wind speed data, and irradiance data. Time information may include, for example, time, holidays, and scheduled rest days.

[0051] Figure 2 An example process of implementing a method for predicting the power load of a power grid according to the present disclosure by a prediction system is shown. Figure 2As shown, the prediction method may include six parts: data collection, data reading, data preprocessing, feature generation, model training, and result output. Among them, data collection may include data collection and warehousing procedures. Specifically, power load data and meteorological data during a predetermined period may be collected. For example, power load data and meteorological data during 96 points of each day in a predetermined historical time period may be collected, but are not limited to, wherein 96 points refers to data collected at 96 time points every day, i.e., collected every 15 minutes. These data may be stored in a database such as a MySQL database. When executing the prediction method, data may be read from a MySQL database, for example, by a SQL statement.

[0052] Here, considering that data quality may affect the accuracy of the final prediction, as an example, in this step, a data preprocessing process can also be performed on the read period power load data and period meteorological data. Since power load is affected by many factors, such as weather, season, population growth, economic changes, etc., there may be various types of abnormal conditions in the load data, such as acquisition equipment failure. Through the data preprocessing process, abnormal conditions can be eliminated or weakened, which can improve the accuracy and reliability of power and electricity forecasts.

[0053] Figure 3 FIG. 1 is an example process for detecting data anomalies in a method for predicting power load of a power grid according to an exemplary embodiment of the present disclosure. Figure 3 Describe the process of data anomaly detection.

[0054] As an example, data preprocessing may include data anomaly detection. Data anomaly detection may include at least one of sudden anomaly removal, mutation anomaly detection, and offset anomaly detection.

[0055] Sudden anomalies may refer to sudden and unexpected anomalies, which may be caused by major events, such as large-scale power outages. Such anomalies are usually difficult to detect and have an irregular duration, resulting in high detection costs. Therefore, for such anomalies, manual labeling can be used to remove them. For example, the data corresponding to the time when the sudden abnormal event occurred can be manually labeled.

[0056] Mutation anomaly detection can refer to the detection of data mutation points. For mutation anomalies, since such anomalies are easier to identify, for some industries or industry loads, due to their small size, there may be some large users with sudden increase in electricity demand, resulting in a mutation in power load. In a single-mode detection, such a real mutation may be confused with the mutation caused by data measurement error. Therefore, multiple detection methods can be used to detect mutation anomalies. Only when multiple detection methods detect mutation anomalies is it determined that data mutation anomalies are detected to improve the accuracy of detection. For example, the mutation anomaly data in the data can be the intersection of the detection results of multiple detection methods.

[0057] As an example, Figure 3 As shown, the windowed interquartile range (IQR) anomaly detection algorithm, the wavelet decomposition anomaly detection algorithm, and the anomaly detection based on PCA (ADPCA) algorithm can be used to detect mutation anomalies.

[0058] Specifically, in the windowed IQR anomaly detection algorithm, assuming that the data to be detected (such as power load data or meteorological data) is l, the number of days to be detected is N, the window width is w, and the sequence after detection is The outliers after detection can be marked as empty Nan, and the set P1 = {} is constructed.

[0059] In this case, we can calculate l iw,(i+1)w The lower quartile Q1 and the upper quartile Q3 of the quartile can be obtained, so that the interquartile range IQR=Q3-Q1 can be obtained, where i=0,...,96N / / w, / / indicates rounding down.

[0060] Based on the interquartile range, the data upper limit and data lower limit of the data to be detected can be calculated. Specifically, the data lower limit is D=Q1-1.5×IQR, and the data upper limit is U=Q3+1.5×IQR. Based on the data upper limit and data lower limit, the sequence after detection can be obtained, for example, it can be expressed by the following formula (1):

[0061]

[0062] Among them, j=iw,…,(i+1)w, j can be added to the set P1 to obtain the detection result, j→P1.

[0063] In the wavelet decomposition anomaly detection algorithm, it can be assumed that the time series data before detection is l, and the time series data after detection is Daubechies wavelet No. 3 is selected as the wavelet basis function. Here, the discrete wavelet transform can be denoted as DWT, and the inverse transform can be denoted as iDWT. The wavelet decomposition anomaly detection algorithm can be expressed by the following equations (2) to (5):

[0064] g i ,h i =DWT(l), P2={} (2)

[0065]

[0066] Where i represents the number of decomposition levels, for example, i can be 1; g l Represents the low-frequency part of the low-frequency signal in the frequency domain; hi represents the high-frequency part in the frequency domain; It represents the time domain performance of the low-frequency part of the frequency domain; represents the time domain performance of the high frequency part of the frequency domain; W represents the windowed anomaly detection operator. In the above formula, the superscript i of each part can be regarded as the i-th layer of decomposition.

[0067] In the anomaly detection algorithm based on the PCA algorithm, it is assumed that the time series data before detection is l and the time series data after detection is The principal component analysis (PCA) decomposition operator is P, the sorting operator is S, and the set P3={} is constructed. For example, the data can be normalized by the following equations (6) to (8):

[0068] l max =max(l)(6)

[0069] l min =min(l) (7)

[0070] l′=(ll min ) / (l max -l min ) (8)

[0071] Among them, l′ represents the normalized data, l max Indicates the maximum value of the original data sequence, l min Represents the minimum value of the original data sequence.

[0072] Based on the normalized data, the PCA operator can be used to reduce the dimension of the data space, for example, it can be expressed by the following formula (9):

[0073] l″=PCA(l′)(9)

[0074] Among them, l″ represents the data after dimensionality reduction.

[0075] Based on the reduced-dimensional data, the projection error δ = S(ll″) can be calculated and sorted, where S represents the sorting operator. The first 95% of the points with the smallest projection error can be taken, which is recorded as set A. The complement of set A can be defined as Among them, the set The points are outliers, and these outliers can be added to the set P3 to get the detection results, that is,

[0076] Based on the above three methods, the mutation anomaly data in the data can be the intersection of the detection results of each detection method, such as the mutation anomaly set P T It can be expressed as P T =P1∩P2∩P3.

[0077] Data offset anomaly can refer to data anomalies at multiple data points or within a period of time. Offset anomalies usually occur in industrial and industry data, and are generally caused by line collection anomalies or transfer anomalies. Such abnormal data mutations usually only occur at the beginning and end of data anomalies, and their start and end times are not fixed. Therefore, it is often difficult to detect offset anomalies, and their impact on data prediction is more serious than that of mutation anomalies. Therefore, their detection should be more stringent. For example, multiple detection methods can be used to detect offset anomalies. When any of the multiple detection methods detects an offset anomaly, it is determined that the data segment is detected as an offset anomaly to improve the accuracy of the detection. For example, the offset anomaly data in the data can be the union of the detection results of multiple detection methods.

[0078] As an example, Figure 3 As shown, the persistence-based anomaly detection (PersistAD) algorithm and the level shift anomaly detection algorithm can be used to detect the shift anomaly.

[0079] In the persistent anomaly detection algorithm, it can be assumed that the detection sequence is l and the sequence length is N l , l [i，j] is the data of the i-th to,th points, l i For the ith data, the detection window width is w, and the quantile interval is c, the median can be calculated as median=median(l [i，i+w] ), calculate the interquartile range IQR = Q3-Q1, where i = 1, ..., N l -w, construct set P4={}.

[0080] Here, the reasonable range of data can be U = [median-c*IQR, median+c*IQR], in response to Then determine li+w is an abnormal data point. In this way, the i+wth data can be marked as an abnormal point and added to the set P4 to obtain the detection result, that is, i+w→P4, l i =Nan, t∈P.

[0081] In the level shift anomaly detection algorithm, it can be assumed that the detection sequence is l and the sequence length is N l , l [i，j] is the data from the i-th to the j-th point, the detection window width is w, and the detection degree is c.

[0082] In this case, the first median m1=median(l [(i-2)w+1，(i-1)w] ) and the second median m2 = median(l [(i-1)w+1,iw] ), where i = 2, ..., N l / / w.

[0083] Thus, in response to Then ((i-1)w+1, ..., iw) is determined to be an abnormal data segment, ((i-1)w+1, ..., iw) → P5, l i =Nan, i∈P5.

[0084] Based on the above two methods, the mutation anomaly data in the data can be the intersection of the detection results of each detection method, such as the offset anomaly set P s It can be expressed as P s =P4∩P5.

[0085] As another example, data preprocessing may also include data repair. Specifically, before executing the step of obtaining data features based on the period power load data, the period meteorological data and time information, the prediction method further includes: performing data repair on the period power load data or the period meteorological data.

[0086] Here, when repairing data, since power load data and meteorological data have periodicity and trend, power load data is affected by temperature data in meteorological data, so when repairing data, meteorological data can be repaired first. The underlying algorithm for repairing the two types of data can adopt the missing value filling method based on time series data modeling.

[0087] For example, Figure 4 As shown, data repair can be performed in the following ways:

[0088] In step S410, the time series of the data to be repaired may be decomposed to obtain multiple subsequences.

[0089] Here, the multiple subsequences may include a basic trend portion subsequence and a period portion subsequence.

[0090] For example, Figure 5 As shown, the data to be repaired can be obtained, and the time series of the data can be decomposed using a time series decomposition method such as time series robust decomposition to obtain a basic trend part subsequence and a period part subsequence. However, the decomposition result is not limited to including the above subsequences, for example, it can also include a residual part.

[0091] For example, assuming that the original data sequence is l, the original data sequence l can be decomposed by the STL algorithm (Seasonal and Trend decomposition using Loess, STL), and STL (l) = l S +l T +l R , where l S represents the basic trend subsequence, l T represents a periodic partial subsequence, l R Represents the residual part.

[0092] In subsequent steps, each subsequence obtained in step S410 may be repaired by time series modeling to obtain a repaired time series.

[0093] Specifically, in step S420, in response to a target subsequence having missing data among the multiple subsequences, a derived feature may be generated based on the target subsequence.

[0094] Here, the derived features may be extracted from the subsequence based on the time feature.

[0095] As an example, for the trend part subsequence of meteorological data, the derived features may include, for example, one or more of the year, month, day of the year, and lunar solar term information. For the periodic part subsequence of meteorological data, the derived features may include, for example, one or more of the meteorological periodic part at the same time yesterday and the time point information. For the trend part subsequence of power load data, the derived features may include, for example, one or more of the year, month, day of the year, holiday information, scheduled rest day information, meteorological trend part value at the same time point, and power load trend part at the same time point yesterday. For the periodic part subsequence of power load data, the derived features may include, for example, one or more of the week, whether it is a weekend, time point information, meteorological periodic part at the same time point, and power load periodic part at the same time point yesterday.

[0096] In step S430, the missing data in the target subsequence may be predicted based on the target subsequence and the derived features, and the predicted data values ​​may be filled into the target subsequence.

[0097] As an example, in this step, multiple missing data prediction methods can be used, and different prediction methods can correspond to different time features. For example, the data in the target subsequence and the derived features can be distinguished according to the time feature to obtain multiple data types, each data type corresponds to a predetermined time feature, and different data types correspond to different time features; the missing data can be predicted using the prediction method corresponding to the time feature of the data type to obtain the predicted data value.

[0098] Here, the time feature may be, for example, a preset month, and the data belonging to the preset month is classified into the same data type, and the data not belonging to the preset month is classified into another data type. The preset month may be, for example, June to August, and / or January to February. The missing data prediction method may include, for example, a multiple regression model algorithm, a random forest algorithm, and the like.

[0099] Specifically, since extreme values ​​of power load may occur in special months, it is difficult for machine learning-based algorithms to predict extreme points that exceed historical sample data. Therefore, when repairing data, it is possible to determine whether the repaired data points are special months (such as June to August, January to February). For example, for data in special months, a multivariate regression model algorithm can be used for modeling to predict and fill in missing values; for non-special months, a random forest algorithm can be used for modeling to predict and fill in missing values.

[0100] In this way, by distinguishing different time features and using different prediction methods to repair missing values, the filled data values ​​can be made more consistent with the characteristics of the original data, making the repair results more accurate.

[0101] Refer to the following Figure 6 An example flow of the timing modeling repair steps for the above subsequence is given. Figure 6 As shown, for any subsequence, for example, the basic trend partial subsequence or the period partial subsequence of the power load data during the period, or the basic trend partial subsequence or the period partial subsequence of the meteorological data during the period, in step S610, it can be determined whether the subsequence contains missing values. In response to containing missing values, the subsequence is a target subsequence. In step S620, derived features can be generated based on the target subsequence to obtain a set of subsequences and derived features.

[0102] The data in the target subsequence and the derived features can be distinguished according to the time feature to obtain multiple data types. Here, the time feature can be, for example, a preset month. In response to the data belonging to the preset month, in step S630, the missing data can be predicted by the multivariate regression model modeling algorithm to obtain the predicted data value, and in step S640, the missing value is filled. Here, other external influencing factors such as rainfall, irradiance, wind speed, etc. can also be considered when adopting the multivariate regression model modeling; in response to the data not belonging to the preset month, in step S650, the missing data can be predicted by the random forest modeling algorithm to obtain the predicted data value, and in step S660, the missing value is filled.

[0103] As an example, after step S640 and step S660, it is possible to detect again whether the filled target subsequence contains missing values. In response to the filled target subsequence containing missing values, it is possible to return to execute step S620 until the filled target subsequence does not contain missing values, and the final subsequence repair result can be obtained.

[0104] Return to reference Figure 4 In step S440, the filled target subsequence may be merged with a subsequence without missing data among the multiple subsequences to obtain repaired data.

[0105] In this step, after the target subsequence is repaired, the target subsequence may be combined with other subsequences in the plurality of subsequences, for example, added together, to obtain repaired data.

[0106] For example, the repaired trend part, the cycle part, and the unrepaired residual part may be added together to obtain a final repair sequence.

[0107] Although some examples of data preprocessing are described above, the embodiments of the present disclosure are not limited thereto, and part or all of the above preprocessing process may be omitted, or part of the preprocessing process may be replaced or added.

[0108] Return to reference Figure 1 In step S120, data characteristics may be obtained based on the power load data during the period, the meteorological data during the period, and the time information.

[0109] Here, the data characteristics may include power load coding characteristics, meteorological characteristics, and time characteristics.

[0110] As an example, the power load coding feature can be obtained based on the power load data during the period and the meteorological data during the period. The power load coding feature can be a coding feature based on sensitivity analysis, which can characterize the power load level only under consideration of a single meteorological factor, such as Figure 7As shown, the power load encoding feature may be a daily electricity quantity-temperature encoding feature.

[0111] For example, the period meteorological data may include the period temperature data, and the power load coding feature may be expressed as the level of the power load (eg, daily power consumption) under the corresponding temperature condition, excluding the influence of other factors.

[0112] In one example, the power load coding feature can be obtained in the following manner: based on the period power load data and the period temperature data, a first coding fitting relationship between the power load and the temperature is obtained; and the period temperature data is substituted into the first coding fitting relationship to obtain the power load coding feature.

[0113] In this example, the power load coding feature can be a power load-temperature coding feature based on the time-sharing load, which can be calculated according to the time-sharing (for example, 96 o'clock) power load, so as to obtain richer power load information, making the calculated power load coding feature more accurate.

[0114] Here, the period power load data and the period temperature data can be fitted by a method such as the least square method to obtain a first coding fitting relationship, in which the temperature can be an independent variable and the power load can be a dependent variable. Thus, the period temperature data is substituted back into the first coding fitting relationship to obtain a power load coding feature that excludes other factors except the temperature factor and only considers the temperature factor.

[0115] For example, we can assume that the power load data is L, is the load at time t on day n, and the temperature data is T. is the corresponding temperature at time t on day n, and the day calculation window is w t , the calculation window at the time point is w p , the calculated coding load value is L′, the coding power value is E, and the fitting order is 4, where t=1,…,96,n=1,…,N / / w t .

[0116] In this case, we can get a set of days that meet the window range. Get the set of time points that meet the windowing range

[0117] Based on the set of days and time points that meet the window range, the load set can be obtained and temperature collection

[0118] Based on the least square method, the data in the day number set and the temperature set can be fitted to obtain a first coded fitting relationship f(X): Y=Xβ+μ.

[0119] here, Among them, β and μ represent the set of fitting coefficients,

[0120] Based on the first coding fitting relationship, the power load coding feature set can be obtained: Right now Among them, you can group the collection by day and time point Sort the coded load values ​​L in .

[0121] In another example, the power load coding characteristics can also be obtained in the following way: according to the preset unit time, the period temperature data and the period power load data are respectively aggregated to obtain the temperature aggregation characteristics and the load aggregation characteristics; based on the temperature aggregation characteristics and the load aggregation characteristics, a second coding fitting relationship between the power load and temperature is obtained; the temperature aggregation characteristics are substituted into the second coding fitting relationship to obtain the power load coding characteristics.

[0122] In this example, the power load coding feature can be a statistical value-based power load-temperature coding feature, which can be encoded based on the statistical values ​​of the deadline power load data and the statistical values ​​of the deadline temperature data, so that the power load encoding can be completed more quickly, and the data format requirements for the acquired deadline power load data and deadline temperature data are not high, which makes it easier to implement coding calculations.

[0123] As an example, there may be multiple statistical values ​​of the data, for example, including but not limited to a daily maximum value, a daily minimum value, and a daily average value, so as to characterize the overall distribution state of the data.

[0124] For example, we can assume that the daily power load series is E, the temperature series is T, and the fitting order is 4.

[0125] By aggregating the temperature series according to the daily maximum, minimum, and average values, we can get the daily maximum value set T within N days. max , minimum value set T min And the mean value set T mean :

[0126] T max ={max(t [96i，96*(i+1)] |i=1…N)}

[0127] T min ={min(t [96i，96*(i+1)] |i=1…N)}

[0128] T mean ={mean(t [96i，96*(i+1)] |i=1…N)}

[0129] Accordingly, the power load series is aggregated according to the daily maximum, minimum, and average values, and the maximum value set E of each day within N days can be obtained. max , minimum value set E min And the mean value set E mean .

[0130] E max ={max(e [96i，96*(i+1)] |i=1…N)}

[0131] E min ={min(e [96i，96*(i+1)] |i=1…N)}

[0132] E mean ={mean(e [96i，96*(i+1)] |i=1…N)}

[0133] The data in the corresponding set can be fitted based on the polynomial regression algorithm to obtain the second coded fitting relationship. Specifically, the data in the temperature minimum value set and the power load minimum value set are fitted to obtain the second coded fitting relationship f between the daily temperature minimum value and the daily power load minimum value. min ; Fit the data in the temperature maximum value set and the power load maximum value set to obtain the second coded fitting relationship f between the daily temperature maximum value and the daily power load maximum value max ; Fit the data in the temperature average value set and the power load average value set to obtain the second coded fitting relationship f between the daily temperature average value and the daily power load average value mean :

[0134]

[0135] Wherein, e1, e2 and e3 represent the maximum value, minimum value and average value of daily power load, respectively, μ, β1, β2, β3 and β4 are fitting coefficients. Although the same symbol is used to represent the fitting coefficients, they are different in different fitting relationships (such as the above f min 、f max and f mean ), the fitting coefficients may be different. Thus, the coded daily power load value can be obtained based on the average value of each statistical value of the daily power load, that is, F E =(e1+e2+e3) / 3.

[0136] Here, the period power load data and the period temperature data can be fitted by a method such as the least square method to obtain a first coding fitting relationship, in which the temperature can be an independent variable and the power load can be a dependent variable. Thus, the period temperature data is substituted back into the first coding fitting relationship to obtain a power load coding feature that excludes other factors except the temperature factor and only considers the temperature factor.

[0137] Two exemplary methods for obtaining power load coding features are described above, and one method can be selected according to actual needs.

[0138] Meteorological characteristics can be obtained based on the period meteorological data and time information. Figure 7 As shown, the meteorological characteristics may include, for example, at least one of temperature characteristics, precipitation characteristics, wind speed characteristics, and irradiance characteristics.

[0139] As an example, the temperature characteristic F T May include at least one of the following: Today's maximum temperature Today's minimum temperature Today's average temperature Yesterday's maximum temperature Yesterday's lowest temperature Yesterday's average temperature Mark consecutive high temperature days and consecutive low temperature days.

[0140] As an example, the precipitation feature F P May include at least one of the following: Average precipitation for the day Cumulative precipitation in the previous 5 days Wherein, d represents the current day, di represents the i-th day before the current day, and i ranges from 1 to 4.

[0141] As an example, the wind speed characteristic F W The average wind speed W for the day can be included mean , irradiance characteristic F R It may include at least one of the following items: average irradiance R from 6:00 to 20:00 on the day mean , aggregate meteorological features weather ={F P , F T , F W , F R}.

[0142] The time feature may include a time mark feature F for indicating a preset time. timeMark and the time encoding feature F of the time period represented by trigonometric functions timeCode , forming the time feature F time ={F timeMark , FtimeCode}.

[0143] As an example, the time-stamping feature F timeMark May include at least one of the following: day of the week (F weekday ), whether it is the weekend (F isWekkend ), whether it is a holiday (F isHoliday ), whether it is a scheduled holiday (F isAdjustedDay ), Holiday Type (F typeofHoliday ), distance from the base date (F distanceToBaseDay ), Month(F month ), the day of the month (F dayOfMonth ), the first day of the year (F dayIfYear ).

[0144] As an example, the time encoding feature F timeCode May include at least one of the following:

[0145] In addition to the above power load coding characteristics F E 、Meteorological characteristics weather and time feature F time In addition, data features may also include other features such as economic features, etc. other , for example, data feature Feature = {F E , F weather , F time , F other}. For example, Figure 7 As shown, economic indicators such as economic indicators, employment indicators, and price indicators can be optionally added according to the difficulty of data acquisition and the characteristics of the prediction object.

[0146] Return to reference Figure 1 In step S130, multiple power load prediction models can be constructed based on data characteristics.

[0147] Here, multiple power load forecasting models may correspond to different forecasting time intervals. As an example, different forecasting time intervals may be distinguished by month, for example, targeted modeling may be performed according to different months in which data is located to obtain multiple power load forecasting models.

[0148] As an example, the power load levels of the prediction time intervals corresponding to the multiple power load prediction models may be different. For example, the prediction time interval may include a first time interval, a second time interval, and other time intervals except the first time interval and the second time interval. The first time interval is, for example, a time interval corresponding to a preset temperature range or a time interval corresponding to a preset season, and the second time interval is a time interval including a preset time node.

[0149] The preset temperature range may be, for example, a temperature range higher than the first preset temperature or lower than the second preset temperature, and the preset season may be, for example, summer or winter. Specifically, since the proportion of cooling / heating load affected by temperature will increase in the power load components in summer or winter (or in a temperature range higher than the first preset temperature or lower than the second preset temperature), the power load will be easily affected by extreme weather, and the electricity consumption may easily reach or exceed the historical peak, causing serious power supply problems.

[0150] The preset time node can be, for example, the month in which a festival such as the Spring Festival occurs. Specifically, there will be a Spring Festival effect in the Spring Festival month, that is, the power load or power consumption will gradually decrease as the Spring Festival approaches; and the power load or power consumption will gradually increase as the Spring Festival moves away.

[0151] In other time intervals except the first time interval and the second time interval, the power consumption is often relatively stable. Therefore, in such time intervals, the power load is in a normalized state.

[0152] In the above example, multiple power load prediction models may include a normal load model, a temperature load model and a node load model, wherein the temperature load model is used to predict the power load in a first time interval corresponding to a preset temperature range, the node load model is used to predict the power load in a second time interval including a preset time node, and the normal load model is used to predict the power load in other time intervals except the first time interval and the second time interval.

[0153] By dividing different prediction time intervals in the above manner and constructing corresponding models, the particularity of the preset temperature and preset time nodes can be taken into account, and the power load can be predicted separately according to the influence of such temperature and time nodes on the power load, so that the division of time intervals and the construction of multiple models are more reasonable, and the accuracy of the prediction results is improved. Specific examples of constructing a normal load model, a temperature load model, and a node load model according to an embodiment of the present disclosure will be described in detail below.

[0154] In step S140, the power load prediction model corresponding to the target time among the multiple power load prediction models may be used to predict the power load at the target time to obtain the predicted power load at the target time.

[0155] In this step, the prediction time interval corresponding to the target time can be determined, so as to determine the corresponding power load prediction model to predict the power load at the target time and obtain the predicted power load.

[0156] As an example, based on the predicted power load and the relationship between power load and power consumption, the power consumption at the target time can be predicted. For example, the daily power consumption can be predicted, and the daily power consumption can be summarized and summed by month to obtain monthly power consumption information.

[0157] In this way, the power generation required by the power grid can be estimated, and the power system can be guided to make reasonable expected power transmission arrangements, avoiding problems such as insufficient power supply or temporary increase in power generation that increase the operating burden of the power system, and maintaining the stable operation of the power grid.

[0158] The following will describe in detail the exemplary process of constructing the normal load model, the temperature load model and the node load model. Here, for different power load forecasting models, different forecasting methods can be used to build the models.

[0159] As an example, a normal load model can be constructed in the following manner: Step S810, based on the model parameters, configure the current machine learning model to be trained; Step S820, use the data features to train the machine learning model to obtain a trained machine learning model; Step S830, use the trained machine learning model as the normal load model.

[0160] Here, the model parameters can be obtained by performing hyperparameter optimization on the training results obtained in the previous training of the machine learning model.

[0161] As an example, the machine learning model to be trained may be one or more, for example, may include one or more of a CatBoost model, an XGBoost model, an SVR model, and a DNN model. However, the embodiments of the present disclosure are not limited thereto, and other models may also be used. For example, a model library may be pre-built, and a model may be selected from the model library as the machine learning model to be trained.

[0162] In step S810, Figure 8 As shown, the database can be read to determine whether there are optimal parameters of each model trained within a predetermined time period (e.g., 7 days) in the database. Here, the optimal parameters can be obtained by performing hyperparameter optimization on the training results obtained in the previous training of the machine learning model. In response to the existence of available optimal parameters, the optimal parameters can be substituted into the model to perform a subsequent training process; in response to the absence of available optimal parameters, a model parameter search method such as a Bayes parameter search method can be used to search for the optimal parameters of each model, and the optimal parameters can be added to the database to re-execute the above step of determining whether the optimal parameters exist in the database.

[0163] In step S820, the above-mentioned data features and the power load corresponding to the data features can be used as training data, for example, D=(E, Feature). Based on the training data, the machine learning model to be trained can be trained using existing training methods such as unsupervised learning, for example, multiple trained models can be obtained.

[0164] In step S830, the trained machine learning model can be used as a normal load model. In an example where there are multiple models, the prediction results of multiple trained models can be fused as a normal load model to obtain a final fused prediction result. As an example, the prediction results can be fused using an averaging method.

[0165] Since the normal load model is used to predict the time period when the power load is relatively stable, the normalized power load characteristics can be learned by training the machine learning model, and the model can be used for prediction. In this way, more comprehensive and comprehensive influencing factors related to the power load can be taken into account, and a comprehensive prediction result can be obtained, making the prediction more accurate.

[0166] As an example, Fig. 9 As shown, the temperature load model can be constructed in the following ways:

[0167] In step S910, a plurality of first sample features may be screened out based on data features.

[0168] Here, the first sample feature can be obtained by screening based on the first time interval. The multiple first sample features can include data features corresponding to the first time interval and data features corresponding to time intervals adjacent to the first time interval.

[0169] For example, the first time interval may include a specific month corresponding to summer or winter, and the data features corresponding to the first time interval may include, for example, data features D1 of the same month in previous years of the specific month, and the data features corresponding to adjacent time intervals of the first time interval may include, for example, data features D2 of the past 30 days of the specific month and data features D3 of the next month of the same period in previous years of the specific month.

[0170] As an example, Fig.10 As shown, the time feature in the first sample feature may include, for example, daily statistical temperature, which may include, for example, daily average temperature. The accumulated daily statistical temperature may be obtained by weighted calculation based on the daily average temperature: T′ D =0.5*T D +0.3*T D-1 +0.2*T D-2 , where T D Indicates the average daily temperature of the current day, T D-1 Indicates the average daily temperature of the previous day, TD-2 represents the daily average temperature of the day before yesterday. Based on the accumulated daily statistical temperature and the corresponding accumulated daily power load, a first sample feature can be obtained.

[0171] In step S920, based on the temperature feature and the power load coding feature in each first sample feature, a first modeling fitting relationship between the temperature and the power load corresponding to each first sample feature can be obtained.

[0172] As an example, a polynomial regression model may be used to fit the temperature feature and the power load coding feature to obtain a first modeling fitting relationship, which may be expressed as: Wherein, E1 represents the power load coding feature, T1 represents the temperature feature, μ, β1, β2, β3 and β4 represent the fitting coefficients. Accordingly, the derivative equation f′(T1) of the first modeling fitting relationship can be obtained.

[0173] In step S930, based on the first modeling fitting relationship corresponding to each first sample feature, a temperature basic load portion corresponding to each first sample feature and a temperature influence load portion corresponding to any first sample feature may be obtained.

[0174] Here, the temperature base load part can be the part of the first modeling fitting relationship whose derivative is the first specific value, and the temperature influence load part can be the difference between the derivative of the first modeling fitting relationship and the temperature base load part. Here, any of the above-mentioned first sample features can be, for example, data features of the past 30 days in a specific month. As an example, the first specific value v1 can be, for example, 0.

[0175] Specifically, each first modeling fitting relationship can be decomposed to obtain the corresponding temperature base load part. For example, assuming The temperature base load part can be expressed as The temperature-affected load part can be expressed as

[0176] In this case, taking the above data features D1, D2 and D3 as an example, the corresponding temperature basic load part and temperature influence load part can be calculated for the data features D1, D2 and D3 respectively, for example Fig.10 The first basic load 1, the second basic load 2 and the third basic load 3 in can be expressed as: D1: D2: D3:

[0177] In step S940, on the basis of the normal load model, the temperature influence load part and the temperature basic load part of each first sample characteristic may be introduced to construct a temperature load model.

[0178] In the example where the first time interval includes summer and winter, due to the existence of cooling and heating loads in summer and winter, and the frequent occurrence of extreme weather phenomena, extreme loads / power consumption occur, posing a major hidden danger to the operation of the power grid. Therefore, when predicting the power load / power consumption in summer and winter, adjustments can be made based on the normalized forecast results of normal months.

[0179] As an example, the growth ratio of the temperature basic load part of each first sample feature can be calculated to obtain the total temperature basic load part, and the total temperature basic load part is merged with the temperature influence load part to obtain the first part of the prediction result, which can be expressed by the following formula (10):

[0180]

[0181] Among them, E1 represents the first part of the prediction result, represents the total temperature base load portion, Indicates the part of the load that is affected by temperature.

[0182] Here, the first part represents the influence of temperature on power load based on the prediction result of the normal load model. Therefore, the second part of the prediction result can be the prediction result of the normal load model, and the first part and the second part can be combined to obtain the final prediction result, which can be expressed by the following formula (11):

[0183] E=(E1+E2) / 2(11)

[0184] Among them, E represents the final prediction result, and E2 represents the second part of the prediction result.

[0185] By predicting the power load in the time period significantly affected by temperature in the above-mentioned way, the influence of temperature can be partially superimposed on the normalized power load forecast result, so that the final forecast result not only takes into account the influence of temperature, but also does not deviate from the general trend of normal power load, making the forecast result more accurate and more in line with the actual situation.

[0186] As an example, Fig.11 As shown, the node load model can be constructed in the following ways:

[0187] In step S1110, second sample features corresponding to the second time interval may be screened out based on the data features.

[0188] As an example, the second sample feature may include daily electricity data and daily statistical temperature data within the second time interval. The daily electricity data may be obtained based on the power load coding feature, such as Fig.12As shown, the daily statistical temperature data may include, for example, the daily maximum temperature, the daily minimum temperature and the daily average temperature.

[0189] Specifically, the power load coding feature L in the second time interval can be converted into daily power data E, and the temperature coding feature T in the second time interval can be converted into daily statistical temperature data (T max , T min , T mean Here, we can use the daily statistical temperature data (T max , T min , T mean ), and calculate the corresponding basic daily electricity consumption part and meteorological influence part respectively. In the following, the maximum daily temperature T max Take the following as an example to illustrate: the daily minimum temperature T min , daily average temperature T mean The calculation process of is similar.

[0190] In step S1120, a second modeling fitting relationship between temperature and power load may be obtained based on the temperature feature and the power load coding feature in the second sample feature.

[0191] Taking the above example, the second sample feature can be grouped by year, for example, y=1,…,Y, where y represents the yth year and Y represents the maximum year in the data. For the daily maximum temperature set for the yth year, it can be written as Here, if Fig.13 As shown, polynomial modeling can be performed to obtain a polynomial model and its derivative function, wherein the fitting order can be 4th order, and the second modeling fitting relationship of the following formula (12) is obtained:

[0192]

[0193] Among them, f y represents the second modeling fitting relationship, β1, β2, β3 and β4 are polynomial fitting coefficients. Substituting back to the second modeling fitting relationship above, we can get the daily power load set for year y:

[0194] In step S1130 , the node base load portion may be determined based on the second modeling fitting relationship at the target temperature.

[0195] Here, the target temperature may be a temperature that makes the derivative of the second modeling fitting relationship a second specific value. As an example, the second specific value v2 may be 0, for example.

[0196] In this step, the basic daily electricity quantity set of each year may be used as the node basic load part, wherein the basic daily electricity quantity of each year may be the power load corresponding to the target temperature.

[0197] Specifically, the second modeling fitting relationship f y Take the derivative and calculate the optimal temperature f corresponding to the second specific value v2 y′ (T max ) = target temperature of v2 Here, the target temperature can be calculated

[0198] The target temperature can be substituted back into the second modeling fitting relationship to obtain the basic daily electricity consumption in year y Therefore, based on the basic daily electricity consumption of each year, the node basic load part can be obtained. For example, the basic daily electricity consumption of each year can be added to the basic daily electricity consumption set In, that is,

[0199]

[0200] The basic daily electricity consumption can be collected Modeling for the year, such as the basic daily electricity E basic Fitting relationship with year Y: E basic =μ+βY. Based on this fitting relationship, substituting the predicted year, we can get the node foundation load part of the predicted year from the perspective of maximum temperature analysis.

[0201] As mentioned above, the daily statistical temperature data can include the daily maximum temperature, daily minimum temperature and daily average temperature. Therefore, based on a similar process, the node basic load part from the perspective of minimum temperature analysis can also be obtained. and the node base load part from the perspective of average temperature analysis

[0202] Thus, the final node basic load part can be obtained by integrating the above three node basic load parts, for example, it can be the average value of the three node basic load parts, that is, the final node basic load part

[0203] In step S1140 , the temperature load portion may be determined based on a derivative of the second modeling fitting relationship.

[0204] In this step, the temperature load part can be obtained by deriving the second modeling fitting relationship.

[0205] Using the above second modeling fitting relationship f y As an example, the second modeling fitting relationship f can be determined yThe derivative function f within a preset time period (e.g. the latest year) Y′ . You can get the maximum daily temperature corresponding to the preset time Substitute it into the derivative function to obtain the temperature load part from the perspective of maximum temperature analysis

[0206] As mentioned above, the daily statistical temperature data can include the daily maximum temperature, daily minimum temperature and daily average temperature. Therefore, based on a similar process, the temperature load part from the perspective of minimum temperature analysis can also be obtained. and the temperature load part from the perspective of average temperature analysis

[0207] Thus, the final temperature load portion can be obtained by combining the above three temperature load portions, for example, it can be the average value of the three temperature load portions, that is, the final temperature load portion For example, Fig.13 As shown, when the month of the Spring Festival is the above preset time node, the predicted daily maximum temperature, daily minimum temperature and daily average temperature of the Spring Festival month can be substituted into the formula to obtain the temperature load part

[0208] In step S1150, the node-influencing load portion may be determined based on the time feature corresponding to the preset time node in the second sample feature.

[0209] In this step, the power load portion affected by the time node may also be determined. As an example, the time feature corresponding to the preset time node may include a node code portion and a week code portion.

[0210] The node code part may include, for example, how many days are left before the node and whether it is the same day of the node, etc. Taking the Spring Festival as an example, the node code part may include how many days are left before the Spring Festival and whether it is the same day of the Spring Festival, etc. The week code part may indicate coding of each week.

[0211] In step S1160, a node load model may be constructed based on the node basic load part, the node influence load part, and the temperature load part.

[0212] In this step, the node basic load part, the node influence load part and the temperature load part can be integrated, for example, they can be added or weighted and added to obtain the node load model. For example, it can be expressed by the following formula (12):

[0213]

[0214] Among them, E represents, Indicates the temperature load part, represents the node basic load part, C1 represents the number of days from the node, C2 represents whether it is the node day, C3 represents the week code, μ, β1, β2, β3, β4 and β5 represent the fitting coefficients. It should be noted that although the fitting coefficients are represented by μ and β in the above formulas, the fitting coefficients can be different for different polynomials.

[0215] By predicting the power load in a time interval including a special time node in the above manner, the power load can be divided into multiple parts for consideration, taking into account both the influence of the time node and the influence of temperature, so that the prediction result is more accurate and closer to the actual situation.

[0216] According to the power load prediction method described above, a power load prediction method based on the background of power grid power supply can be provided, and a power forecasting model under different scenarios can be constructed to obtain accurate power load forecast values. At the power grid level, it can effectively improve the effectiveness of power grid planning and ensure the safe and stable operation of the power grid.

[0217] In addition, according to the prediction method of the embodiment of the present invention, it can be oriented towards the background of power grid supply security, with the goal of more accurate power load forecasting results. Compared with the previous monthly granularity power load forecasting method, it uses richer daily dimension information. Compared with the previous daily dimension prediction, and then aggregates it into a monthly dimension power load prediction method, it avoids the deviation accumulation effect caused by rolling prediction through power load-temperature encoding, and according to different prediction situations, calls a prediction model that is more suitable for the situation, so as to achieve a more accurate monthly power load forecasting effect, and provide a more accurate boundary information basis for power grid supply security.

[0218] According to the embodiment of the present disclosure, the above prediction method is verified by taking the monthly power consumption prediction of a certain region from May 2022 to April 2023 as the implementation scenario. Here, the power load can be predicted based on the above prediction method, and then based on the relationship between the power load and the power consumption, the monthly power consumption in the time interval is predicted to obtain the monthly power consumption information.

[0219] Specifically, the data from January 1, 2019 to April 30, 2023 in the region are used for testing, and the above prediction method is used to make monthly predictions for the region from May 1, 2022 to April 30, 2023, that is, the electricity consumption of the next month is predicted each time, and the prediction results are as follows: Fig.14 and Fig.15 shown.

[0220] Fig.14 The daily electricity consumption forecast results from May 1, 2022 to April 30, 2023 are shown. Fig.15The monthly electricity consumption results from May 2022 to April 2023 are shown, where the horizontal axis is time (in months) and the vertical axis is electricity consumption (in 100 million Mwh). Fig.14 and Fig.15 It can be seen that the prediction results are similar to the actual electricity consumption results, and the prediction accuracy is high. The accuracy of the monthly prediction is shown in Table 1 below:

[0221] Table 1

[0222]

[0223] According to a second aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. Specifically, the method for predicting the power load of a power grid according to an embodiment of the present disclosure can be written as a computer program and stored on a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute the method for predicting the power load of the power grid according to the exemplary embodiment of the present disclosure. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer program and any associated data, data files and data structures in a non-transitory manner and provide the computer program and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. In one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0224] According to a third aspect of an embodiment of the present disclosure, a computer program product is provided. The computer program product includes computer instructions. When the computer instructions are executed by a processor, a method for predicting the power load of a power grid according to an embodiment of the present disclosure can be implemented.

[0225] In addition, the method for predicting the power load of the power grid according to the embodiment of the present disclosure can be implemented on an electronic device. Specifically, the electronic device includes: at least one processor; at least one memory storing processor-executable instructions, wherein the processor-executable instructions, when executed by the at least one processor, prompt the at least one processor to execute the method for predicting the power load of the power grid according to the embodiment of the present disclosure.

[0226] As an example, the electronic device may be a PC, a tablet device, a personal digital assistant, a smart phone, or other device capable of executing the above instruction set. Here, the electronic device is not necessarily a single electronic device, but may also be any device or circuit capable of executing the above instruction (or instruction set) individually or in combination. The electronic device may also be part of an integrated control system or system manager, or may be configured as a portable electronic device interconnected with a local or remote (e.g., via wireless transmission) interface.

[0227] In electronic devices, the processor may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0228] The processor can execute instructions or codes stored in the memory, wherein the memory can also store data. Instructions and data can also be sent and received through the network via the network interface device, wherein the network interface device can adopt any known transmission protocol.

[0229] The memory may be integrated with the processor, for example, RAM or flash memory is arranged within an integrated circuit microprocessor or the like. In addition, the memory may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory and the processor may be operatively coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor can read files stored in the memory.

[0230] In addition, the electronic device may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.) All components of the electronic device may be connected to each other via a bus and / or a network.

[0231] The specific implementation methods of the present disclosure have been described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments may be modified and varied without departing from the principles and spirit of the present disclosure, the scope of which is defined by the claims and their equivalents. These modifications and variations should also be within the scope of protection of the claims of the present disclosure.

Claims

1. A method for predicting power load of a power grid, characterized in that: The prediction method comprises: Acquire period power load data of the power grid to be predicted, period meteorological data of the area where the power grid to be predicted is located, and time information corresponding to the period power load data and the period meteorological data; Based on the power load data of the period, the meteorological data of the period and the time information, data features are obtained, wherein the data features include power load coding features, meteorological features and time features; Based on the data characteristics, construct multiple power load forecasting models, wherein the multiple power load forecasting models correspond to different forecasting time intervals; The power load prediction model corresponding to the target time among the multiple power load prediction models is used to predict the power load at the target time to obtain the predicted power load at the target time.

2. The prediction method according to claim 1, characterized in that: The period meteorological data includes period temperature data, wherein the power load coding feature is obtained by: Based on the power load data during the period and the temperature data during the period, a first coded fitting relationship between the power load and the temperature is obtained; Substituting the period temperature data into the first coding fitting relationship, the power load coding feature is obtained.

3. The prediction method according to claim 1, characterized in that: The period meteorological data includes period temperature data, wherein the power load coding feature is obtained by: According to a preset unit time, respectively aggregating the temperature data of the period and the power load data of the period to obtain a temperature aggregation feature and a load aggregation feature; Based on the temperature aggregation feature and the load aggregation feature, a second coded fitting relationship between power load and temperature is obtained; Substitute the temperature aggregation feature into the second coding fitting relationship to obtain the power load coding feature.

4. The prediction method according to claim 1, characterized in that: The multiple power load forecasting models correspond to different power load levels in the forecast time intervals, and the multiple power load forecasting models include a normal load model, a temperature load model, and a node load model. Among them, the temperature load model is used to predict the power load in the first time interval corresponding to the preset temperature range, the node load model is used to predict the power load in the second time interval including the preset time node, and the normal load model is used to predict the power load in other time intervals except the first time interval and the second time interval.

5. The prediction method according to claim 4, characterized in that: The normal load model is constructed in the following way: Based on the model parameters, configuring the machine learning model to be trained, wherein the model parameters are obtained by performing hyperparameter optimization on the training results obtained in the previous training of the machine learning model; Based on the data features, training the machine learning model to obtain a trained machine learning model; The trained machine learning model is used as the normal load model.

6. The prediction method according to claim 4 or 5, characterized in that: The temperature load model is constructed in the following way: Based on the data features, a plurality of first sample features are screened out, wherein the plurality of first sample features include data features corresponding to the first time interval and data features corresponding to time intervals adjacent to the first time interval; Based on the temperature feature and the power load coding feature in each first sample feature, a first modeling fitting relationship between the temperature and the power load corresponding to each first sample feature is obtained respectively; Based on the first modeling fitting relationship corresponding to each first sample feature, a temperature basic load part corresponding to each first sample feature and a temperature influence load part corresponding to any first sample feature are obtained, wherein the temperature basic load part is a part of the first modeling fitting relationship whose derivative is a first specific value, and the temperature influence load part is a difference between the derivative of the first modeling fitting relationship and the temperature basic load part; On the basis of the normal load model, the temperature influence load part and the temperature basic load part of each first sample characteristic are introduced to construct the temperature load model.

7. The prediction method according to claim 4, characterized in that: The node load model is constructed in the following way: Based on the data feature, filtering out a second sample feature corresponding to the second time interval; Based on the temperature feature and the power load coding feature in the second sample feature, a second modeling fitting relationship between temperature and power load is obtained; determining a node base load portion based on the second modeling fitting relationship at a target temperature, wherein the target temperature is a temperature at which a derivative of the second modeling fitting relationship is a second specific value; determining a temperature load component based on a derivative of the second modeled fit relationship; Determine a node load-affecting portion based on a time feature in the second sample feature that corresponds to the preset time node; The node load model is constructed based on the node basic load part, the node influence load part and the temperature load part.

8. The prediction method according to claim 1, characterized in that: Before executing the step of obtaining data features based on the power load data during the period, the meteorological data during the period and the time information, the prediction method further includes: performing data repair on the power load data during the period or the meteorological data during the period, wherein the data repair is performed in the following manner: Decomposing the time series of the data to be repaired to obtain multiple subsequences, wherein the multiple subsequences include a basic trend part subsequence and a period part subsequence; In response to a target subsequence having missing data among the multiple subsequences, generating a derived feature based on the target subsequence, wherein the derived feature is data extracted from the subsequence based on a time feature; Based on the target subsequence and the derived features, predict missing data in the target subsequence, and fill the target subsequence with the predicted data values; The filled target subsequence is merged with a subsequence without missing data among the multiple subsequences to obtain repaired data.

9. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to perform the method for predicting the power load of a power grid as described in any one of claims 1-8.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, a method for predicting power load of a power grid according to any one of claims 1 to 8 is implemented.

Images