Short-term power load forecasting method for smart power system

Abstract

The present application relates to the technical field of power load prediction, in particular to a short-term power load prediction method for smart power system, the difference between the load target change trend before and after interpolation of the power load in the reference time period of each time of the day is obtained, so as to obtain the final change trend of the load of each time; the load burst index of each period of the day is obtained from the difference between the power load of the same period of the day and the historical day, the correlation of the power load of any two periods of the day and the correlation of the final change trend of the load are combined, the similarity of the change trend of the load of any two periods is obtained, thereby the power consumption scene classification of the period of the day is carried out, and the reference value of each category for power load prediction is obtained according to the similarity of the change trend of the load in each category, the reference value of different power consumption scenes is considered in the power load prediction, and the accuracy of short-term power load prediction is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power load prediction, in particular to a short-term power load prediction method for smart power systems. BACKGROUND

[0002] In the process of predicting the short-term power load of a smart power system, the power load at a certain time in the future is predicted according to the power load in a certain period of history. The power load in a certain period of history usually includes multiple different power consumption scenarios, and the power consumption load of different power consumption scenarios is different, so their importance in prediction should be different. However, in the existing short-term power load prediction process, the importance of different power consumption scenarios in prediction is not considered, and if the importance or reference value of different power consumption scenarios in prediction is completely the same, the accuracy of short-term power load prediction will be affected. SUMMARY

[0003] In order to solve the technical problem of low accuracy of the existing short-term power load prediction, the purpose of the present application is to provide a short-term power load prediction method for smart power systems, and the technical solution adopted is as follows:

[0004] The present application provides a short-term power load prediction method for smart power systems, comprising:

[0005] Respectively determine the load target change trend before and after interpolation of the power load in the reference time period of each time of the day;

[0006] According to the difference between the load target change trends before and after interpolation, adjust the load target change trend after interpolation to obtain the final load change trend of each time;

[0007] Obtain the load burst index of each period of the day from the difference between the final load change trends of the same period of the day and the history day;

[0008] According to the correlation of the power load of any two periods of the day and the correlation of the final load change trend, and the abnormal index of any two periods, obtain the load change trend similarity of any two periods;

[0009] Based on the load change trend similarity, classify the power consumption scenarios of the periods of the day, and according to the load change trend similarity in each category, obtain the reference value of each category for power load prediction.

[0010] In an exemplary embodiment, the process of obtaining the load target change trend comprises:

[0011] obtaining a load initial change trend at each time point from the change of the power load in the reference time period at each time point in the current day and the historical day;

[0012] determining an unstable time period of the load initial change trend in the current day and the historical day;

[0013] obtaining a load change regularity at each time point in the current day from the overlap of the unstable time period in the current day and the historical day, and the difference of the load initial change trend in the current day and the historical day;

[0014] adjusting the load initial change trend by the load change regularity to obtain the load target change trend.

[0015] In an exemplary embodiment, the process of obtaining the load initial change trend comprises:

[0016] determining extreme points of the power load in the reference time period, the extreme points including maximum and minimum;

[0017] determining the power load difference between adjacent extreme points, and the change intensity of the power load between adjacent extreme points;

[0018] fusing the power load difference and the change intensity of all adjacent extreme points in the reference time period to obtain a power load change trend performance;

[0019] obtaining the load initial change trend according to the power load change trend performance and the proportion of the number of extreme points; the load initial change trend is positively correlated with the power load change trend performance and the proportion of the number of extreme points.

[0020] In an exemplary embodiment, the process of obtaining the load change regularity comprises:

[0021] determining the number of unstable time periods in which each time point in the current day is located in the current day and the historical day, the number representing the number of days in the current day and the historical day;

[0022] obtaining the load change regularity at each time point in the current day according to the number of each time point in the current day, and the difference of the load initial change trend between each time point in the current day and the same time point in the historical day; the load change regularity is positively correlated with the number, and negatively correlated with the difference of the load initial change trend.

[0023] In an exemplary embodiment, the process of obtaining the load final change trend comprises:

[0024] determining the number of interpolation data points as extreme points after interpolating the power load in the reference time period of the candidate time; the extreme points include maximum and minimum values; the candidate time is any time of the day;

[0025] obtaining an adjustment coefficient of the candidate time according to the number of interpolation data points and the difference between the load target change trend before and after interpolation of the candidate time; the adjustment coefficient is negatively correlated with the number of interpolation data points and the difference between the load target change trend before and after interpolation of the candidate time;

[0026] adjusting the load target change trend after interpolation of the candidate time according to the adjustment coefficient to obtain the final load change trend of the candidate time.

[0027] In an exemplary embodiment, the process of obtaining the load burst indicator includes:

[0028] determining the difference degree of the final load change trend of the candidate period and the same period of each day in the historical days; the candidate period is any period of the day;

[0029] determining the reference day of the candidate period for the current day; the reference day is a historical day with a difference degree greater than a preset difference threshold;

[0030] obtaining the abnormal performance of the candidate period according to the interval days between the current day and each reference day and the difference degree of the current day and each reference day; the abnormal performance is positively correlated with the interval days and the difference degree;

[0031] obtaining the load burst indicator of the candidate period according to the abnormal performance and the proportion of the number of reference days; the load burst indicator is positively correlated with the abnormal performance and the proportion of the number of reference days.

[0032] In an exemplary embodiment, the process of obtaining the similarity of load change trends of any two periods includes:

[0033] determining the average value of the load burst indicators of the any two periods;

[0034] obtaining the similarity of load change trends of the any two periods from the average value of the load burst indicators of the any two periods, the correlation of power loads of the any two periods, and the correlation of final load change trends; the similarity of load change trends is negatively correlated with the average value of the load burst indicators and positively correlated with the correlation of power loads of the any two periods and the correlation of final load change trends.

[0035] In an exemplary embodiment, the process of classifying the power consumption scene of the period of the current day includes:

[0036] a clustering distance of any two time periods is obtained from the similarity of the load change trend of the two time periods, and the clustering distance is negatively correlated with the similarity of the load change trend;

[0037] a plurality of categories are obtained by clustering each time period in the day based on the clustering distance of any two time periods, and different categories represent different power consumption scenarios.

[0038] In an exemplary embodiment, the reference value obtaining process comprises:

[0039] a minimum value of the similarity of the load change trend in each category and the number of time periods included in each category are obtained;

[0040] the reference value of each category is obtained according to the minimum value of the similarity of the load change trend of each category and the number of time periods, and the reference value is positively correlated with both the minimum value of the similarity of the load change trend and the number of time periods.

[0041] In an exemplary embodiment, after obtaining the reference value of each category for power load prediction, the short-term power load prediction method for smart power systems further comprises:

[0042] a weight of the power load at each time in the day is obtained from the reference value of the category in which each time in the day is located, and the power load at each time in the day is weighted;

[0043] short-term power load prediction is performed according to the weighted power load at each time in the day.

[0044] The present application has the following advantages: the present application determines the load burst indicators of each time period in the day in combination with the difference between the power load of the same time period of the day and the historical day, obtains the similarity of the load change trend of any two time periods of the day, and uses the similarity as a reference for power consumption scenario classification, classifies the current time periods to obtain a plurality of categories, and different categories represent different power consumption scenarios, thereby achieving the division of power consumption scenarios of the day, determining the reference value of each category for power load prediction, and different power consumption scenarios have different reference values in power load prediction. By considering the reference values of different power consumption scenarios in power load prediction, the reference values of various power consumption scenarios in power load prediction can be obtained, the greater the reference value, the more important the role in power load prediction, thereby improving the accuracy of short-term power load prediction. In addition, the present application also considers the accurate acquisition of the load change trend of each time when there is power load data loss, thereby ensuring the accuracy of the reference value of each power consumption scenario for power load prediction. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1is a flow chart of a short-term power load prediction method for a smart power system provided by one embodiment of the present application;

[0046] Figure 2 is a flow chart of obtaining a load target change trend provided by one embodiment of the present application;

[0047] Figure 3 is a flow chart of obtaining a load initial change trend provided by one embodiment of the present application;

[0048] Figure 4 is a flow chart of obtaining a load change regularity provided by one embodiment of the present application;

[0049] Figure 5 is a flow chart of obtaining a load final change trend provided by one embodiment of the present application;

[0050] Figure 6 is a flow chart of obtaining a load burst index provided by one embodiment of the present application;

[0051] Figure 7 is a flow chart of obtaining a load change trend similarity provided by one embodiment of the present application;

[0052] Figure 8 is a flow chart of classifying an electricity use scene of a time period of a day provided by one embodiment of the present application;

[0053] Figure 9 is a flow chart of obtaining a reference value provided by one embodiment of the present application;

[0054] Figure 10 is a flow chart of a short-term power load prediction method for a smart power system provided by one embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to further clarify the technical means and effects of the present application taken to achieve the predetermined purpose of the application, the specific embodiments, structures, features and effects of the present application are described in detail below in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The data information collected by the present application is obtained with full authorization.

[0057] The embodiment provides a short-term power load prediction method for a smart power system, which is used for short-term power load prediction of the smart power system, and the prediction object can be each user in the smart power system. Taking any one user as an example, a smart power meter is arranged at the user, and is used for detecting power load of the user at each moment. The embodiment performs power load prediction according to power load in a certain historical time. The power load in the certain historical time usually includes various different power consumption scenarios, for example, when the weather is hot, the air conditioner operates, thereby causing the power load to be at a high level, and when the temperature is appropriate, the power load is at a low level, and the like. The embodiment combines the power load conditions of different power consumption scenarios, determines reference values of each power consumption scenario in power load prediction, and thereby improves the accuracy of power load prediction.

[0058] The power load collected by the embodiment includes power load data of the day and power load data of historical days. The historical days are consecutive preset days before the day, and the number of the historical days is set according to actual prediction needs, for example, each day within one year before the day. In addition, the sampling frequency of the power load is also set according to actual prediction needs, for example, once per minute.

[0059] In the embodiment, the smart power meter of each user in the smart power system communicates with the cloud server wirelessly, the smart power meter uploads the power load data of the user to the cloud server, and the background monitoring platform of the power company can perform power load prediction according to the power load data received by the cloud server, so as to realize real-time monitoring and management of the power load.

[0060] As shown in Figure 1 The short-term power load prediction method for the smart power system provided by the embodiment includes the following steps:

[0061] Step S1: respectively determining load target change trends before and after interpolation of power load in each time period of the day;

[0062] Step S2: adjusting the load target change trend after interpolation according to the difference between the load target change trends before and after interpolation, to obtain a final change trend of the load at each moment;

[0063] Step S3: obtaining a load burst index of each time period of the day from the difference between the final change trends of the load of the day and the historical days in the same time period;

[0064] Step S4: obtaining load change trend similarity of any two time periods according to the correlation of the power load of the day and the correlation of the final change trends of the load of the two time periods, and the abnormal index of the two time periods;

[0065] Step S5: Based on the similarity of load change trends, classify the electricity consumption scenarios for the time period of the day, and obtain the reference value of each category for power load forecasting based on the similarity of load change trends in each category.

[0066] The following detailed explanation of each step, in conjunction with the accompanying drawings, is provided.

[0067] Step S1: Determine the load target change trend before and after interpolating the power load within the reference time period at each moment of the day.

[0068] When smart meters upload power load data to the cloud server via wireless communication, the load may fluctuate frequently, especially during peak electricity consumption periods. A large amount of power load data needs to be transmitted over the network, which may lead to insufficient network bandwidth and network congestion, resulting in the loss of some power load data transmission.

[0069] For ease of explanation, we will use any time of day as an example, defining any time of day as a candidate time. A reference time period is then determined for each candidate time. In an exemplary embodiment, the candidate time is used as the last time of its reference time period, and a preset number of times is used as the duration of the reference time period, thus forming the reference time period for the candidate time. The preset number of times is set according to actual needs; for example, if the preset number is 50, then the reference time period for the candidate time is a time period with 50 times.

[0070] Under normal circumstances, the number of power load data points within the reference time period of a candidate time is the same as the number of moments within that reference time period, and they correspond one-to-one. However, if power load data transmission is lost at some moments, the reference time period for a candidate time may contain fewer than a preset number of power load data points; for example, the reference time period for a candidate time may contain fewer than 50 power load data points. This embodiment uses data interpolation to supplement the missing power load data within the reference time period of the candidate time. Data interpolation is a common existing technology, and its specific method will not be described here. It should be understood that the power load data before interpolation is the data information actually collected by the power company's backend monitoring platform.

[0071] The load target change trend before and after interpolation of the power load within the reference time period of the candidate time is determined separately. Since the processes for obtaining the load target change trend before and after interpolation of the power load within the reference time period of the candidate time are the same, the following example uses either one, such as the load target change trend of the power load within the reference time period of the candidate time before interpolation. The load target change trend characterizes the change trend of the power load within the reference time period of the candidate time. In an exemplary embodiment, such as...Figure 2 As shown in FIG. 6, a specific process for obtaining the load target change trend is as follows:

[0072] Step S11: Obtain the load initial change trend of each time point from the change of the power load in the reference time period of each time point in the current day and the historical days.

[0073] It should be understood that, before interpolation, for the time points with missing power load, these time points are deleted from the reference time period or are not considered in the process of obtaining the load target change trend, so as to ensure that the time points with power load data constitute continuous time points in time sequence. For example, the time points of the reference time period are , if the power load data of the time point and the time point are missing, the time point and the time point are deleted from the reference time period, so that each time point in the reference time period is .

[0074] For each day in the historical days, the same time point as the candidate time point is also obtained, which is also a candidate time point in each day in the historical days, so as to obtain the reference time period of the candidate time point. Obtain the load initial change trend of the candidate time point from the change of the power load in the reference time period of the candidate time point in the current day and the historical days. It should be understood that the power load data of the historical days is also the power load data before interpolation, that is, the power load data of the historical days actually collected by the background monitoring platform of the power company.

[0075] Taking the reference time period of the candidate time point of the current day as an example, as shown in FIG. 7, a specific process for obtaining the load initial change trend is as follows: Figure 3

[0076] Step S111: Determine the extreme points of the power load in the reference time period.

[0077] Curve fitting is performed on the power load of each time point in the reference time period of the candidate time point of the current day to obtain a power load change curve. Then, each extreme point in the power load change curve is obtained, including the maximum value and the minimum value. The number of extreme points in the power load change curve is obtained.

[0078] Step S112: Determine the power load difference between adjacent extreme points and the change degree of the power load between adjacent extreme points.

[0079] ​The power load difference of adjacent extreme points is obtained. Since the maximum value and the minimum value appear alternately, every two adjacent extreme points include one maximum value and one minimum value. Then, the absolute value of the difference between the power load of the maximum point in the adjacent extreme points and the power load of its adjacent minimum point is obtained as the power load difference of the adjacent extreme points. Thus, the power load difference of each adjacent extreme point in the power load change curve is obtained.

[0080] For any two adjacent extreme points, the power load at each time between the two adjacent extreme points is obtained. Then, the change intensity of the power load at each time between the two adjacent extreme points is obtained. In an exemplary embodiment, the standard deviation of the power load at each time between the two adjacent extreme points is obtained, and the standard deviation is taken as the change intensity. Thus, the change intensity of the power load between each adjacent extreme point in the power load change curve is obtained.

[0081] Step S113: The power load difference and the change intensity of all adjacent extreme points in the reference time period are fused to obtain the power load change trend performance.

[0082] For any two adjacent extreme points, the power load change trend index of the two adjacent extreme points is obtained according to the power load difference of the two adjacent extreme points and the change intensity of the power load. The greater the power load difference, the greater the power load fluctuation, and the greater the power load change trend index, and the two are positively correlated. The greater the change intensity of the power load, the greater the power load fluctuation, and the greater the power load change trend index, and the two are positively correlated. In an exemplary embodiment, the product of the power load difference of the two adjacent extreme points and the change intensity of the power load is obtained, and then the product is normalized. The result after normalization is taken as the power load change trend index of the two adjacent extreme points. The normalization can use the tanh function.

[0083] Then, for all adjacent extreme points in the reference time period of the candidate time, that is, all two adjacent extreme points, the average of the power load change trend indexes of all two adjacent extreme points is calculated, and the result is taken as the power load change trend performance of the candidate time of the day. The stronger the power load change trend performance, the stronger the load initial change trend of the candidate time of the day, and the two are positively correlated.

[0084] Step S114: The load initial change trend is obtained according to the power load change trend performance and the proportion of the number of extreme points.

[0085] The total amount of power load in the reference time period of the candidate time of the day is obtained, and the ratio of the number of extreme points in the power load change curve to the total amount of power load is calculated as the number ratio of extreme points in the reference time period of the candidate time of the day. The greater the number ratio of extreme points, the more frequent the power load rises and falls in the reference time period of the candidate time of the day, the more drastic the change, and the more unstable the change trend, that is, the greater the initial load change trend. Therefore, the initial load change trend is positively correlated with the number ratio of extreme points.

[0086] Based on the above logical analysis, in an exemplary embodiment, the product of the power load change trend performance of the candidate time of the day and the number ratio of extreme points is calculated to obtain the initial load change trend of the candidate time of the day. Similarly, the initial load change trends of each time of each day in the history days are obtained.

[0087] Step S12: Determine the unstable time period of the initial load change trend in the day and the history days.

[0088] Because the power consumption scene of the smart power system is complex, there are many reasons for the unstable power load change trend, for example, line sudden failure causes power failure in some areas, at this time, the unstable power load change is an irregular phenomenon, when predicting the future power load, these power load data will reduce the accuracy of the prediction result. For the unstable time period that often appears in the history days, for example, in the afternoon of summer working days, residents return home and start air conditioners and kitchen appliances, which causes the power load to rise sharply in a short time, at this time, the unstable power load change is a regular phenomenon, when predicting the future, the power load data of this time period is more in line with the actual load change, which can effectively improve the accuracy of the power load prediction.

[0089] The initial load change trend of each time of the day is sequenced according to time sequence to obtain a sequence of initial load change trends, and then the unstable time period of the initial load change trend is determined from the sequence. In an exemplary embodiment, a preset change trend instability threshold is preset for determining whether the initial load change trend of each time is unstable, the value of the preset change trend instability threshold ranges from 0 to 1, and the specific value is determined according to actual judgment needs. For example, if a relatively strict judgment mechanism is required, the preset change trend instability threshold can be set to be relatively small, such as 0.5. The initial load change trend of each time of the day is compared with the preset change trend instability threshold, and the times corresponding to the initial load change trends greater than or equal to the preset change trend instability threshold are determined as unstable times, and the unstable times that are continuous in time sequence form an unstable time period, so as to obtain several unstable time periods of the initial load change trend of the day. Similarly, several unstable time periods of the initial load change trend of each day in the history days are obtained.

[0090] Step S13: obtaining the load change regular performance of each time of the day according to the overlap of the unstable time periods in the day and the history days and the difference between the initial load change trends of the day and the history days.

[0091] The overlap of the unstable time periods in the day and the history days reflects the regular phenomenon of load change, and the more the overlap, the more common the load change. The difference between the initial load change trends of the day and the history days can also reflect the regular phenomenon of load change, and the smaller the difference, the more common the load change. Therefore, the load change regular performance of each time of the day is obtained according to the two parts. In an exemplary embodiment, as shown in FIG. 6, a specific acquisition process of the load change regular performance is as follows: Figure 4

[0092] Step S131: determining the number of unstable time periods in which each time of the day is located in the day and the history days.

[0093] For the candidate time of the day, the number of unstable time periods in which the candidate time is located is initially set to 0, and each day in the day and the history days is traversed. If the candidate time is located in the unstable time period of the day, the number of unstable time periods in which the candidate time is located is increased by 1, until all days in the day and the history days are traversed, so as to obtain the number of unstable time periods in which the candidate time is located. The number of unstable time periods in which the candidate time is located represents the number of days in which the candidate time is located in the unstable time period. In this way, the number of unstable time periods in which each time of the day is located in the day and the history days is obtained.

[0094] ​Step S132: Obtain the load change regularity of each time of the day according to the number of each time of the day and the load initial change trend difference between each time of the day and the historical time.

[0095] For the candidate time, calculate the average of the load initial change trend of the candidate time of each day of the historical day as the load initial change trend of the candidate time of the historical day. Then, obtain the load initial change trend difference between the load initial change trend of the candidate time of the day and the load initial change trend of the candidate time of the historical day, and the load initial change trend difference is specifically the absolute value of the difference of the load initial change trend. The smaller the load initial change trend difference is, the closer the load initial change trend of the candidate time in the day and the historical day is, the more common the load change of the candidate time is, and the stronger the load change regularity is. Therefore, the load change regularity is negatively correlated with the load initial change trend difference.

[0096] The more the number of unstable time periods in which the candidate time of the day is located is, the more common the load change of the candidate time in the day and the historical day is, and the stronger the load change regularity is. Therefore, the load change regularity is positively correlated with the number of unstable time periods in which the candidate time of the day is located.

[0097] Based on the above logic, a specific calculation method of the load change regularity is as follows:

[0098] ;

[0099] Wherein, represents the load change regularity of the i-th time of the day, represents the number of unstable time periods in which the i-th time of the day is located, represents the total number of days of the day and the historical day, represents the load initial change trend difference of the i-th time of the day.

[0100] Step S14: Adjust the load initial change trend by the load change regularity to obtain the load target change trend.

[0101] For the time period of frequent load instability, for example, 14:00-16:00 in the summer high temperature period, the concentrated outbreak of air conditioning load leads to a sudden increase in load, and such instability occurs many times in the historical same period, which belongs to "regular instability". Although such instability fluctuates greatly, it has historical data to support its regularity, so for the initial change trend of the load in these unstable time periods, even if the initial change trend of the load is large, it is also a regular phenomenon in line with the scene regularity. Therefore, the initial change trend of the load at each time of the day is adjusted by the regular performance of the load change at each time of the day to obtain the target change trend of the load at each time of the day. In an exemplary embodiment, the adjustment method is as follows:

[0102] ;

[0103] Wherein, represents the target change trend of the load at the i th time of the day, represents the initial change trend of the load at the i th time of the day.

[0104] The above process obtains the target change trend of the power load at each time of the day before interpolation. Using the above process, the target change trend of the power load at each time of the day after interpolation is obtained. It should be understood that in the process of obtaining the target change trend of the power load at each time of the day after interpolation, the power load data of the historical day involved is also the power load data after interpolation.

[0105] Step S2: According to the difference between the target change trend before and after interpolation, adjust the target change trend after interpolation to obtain the final change trend of the load at each time.

[0106] Because the interpolated power load data is not accurate, it will destroy the real time sequence correlation and trend characteristics of the power load data, and the more important the interpolated power load data is, the greater the impact on the accuracy of the short-term power load prediction. Because the extreme point in the power load data is the key feature of the power load change trend, if the interpolated power load is exactly the extreme point, it may distort the real fluctuation rule of the power load, and the more the number of interpolated power load extreme points, the more serious the damage to the power load change trend.

[0107] In an exemplary embodiment, as shown in Figure 5 , a specific process for obtaining the final change trend of the load is as follows:

[0108] Step S21: Determine the number of interpolated data that are extreme points after interpolating the power load in the reference time period of the candidate time.

[0109] After interpolating the power load in the reference time period of the candidate time point of the day, extreme points in the power load in the reference time period are obtained, the extreme points including maximum values and minimum values. Then, it is determined which of the extreme points belong to the interpolated data, so that the number of interpolated data as extreme points in the interpolated data is obtained. The more the number of interpolated data as extreme points, the more extreme points are introduced in the place where the power load is missing, the more serious the influence on the power load change trend, the greater the influence on the credibility of the power load change trend, that is, the more the power load target change trend after interpolation needs to be adjusted, and then the smaller the corresponding adjustment coefficient, the adjustment coefficient being negatively correlated with the number of interpolated data as extreme points.

[0110] Step S22: obtaining the adjustment coefficient of the candidate time point according to the number of interpolated data and the difference of the power load target change trend before and after interpolation of the candidate time point.

[0111] The difference of the power load target change trend before and after interpolation of the candidate time point of the day is obtained, specifically, the absolute value of the difference of the power load target change trend before and after interpolation of the candidate time point of the day is calculated as the difference of the power load target change trend. The smaller the difference of the power load target change trend, the weaker the influence of interpolation on the change trend of the power load, the higher the credibility of the power load change trend, that is, the smaller the adjustment of the power load target change trend after interpolation, and then the larger the corresponding adjustment coefficient, the adjustment coefficient being negatively correlated with the difference of the power load target change trend. Based on the above logic, a specific confirmation method of the adjustment coefficient is given as follows:

[0112] ;

[0113] Wherein, represents the adjustment coefficient of the i th time point of the day, representing the credibility of the power load change trend of the i th time point of the day, represents the power load target change trend after interpolation of the i th time point of the day, represents the number of interpolated data as extreme points in the interpolated data of the i th time point of the day, represents the number of interpolated data of the i th time point of the day.

[0114] Step S23: adjusting the power load target change trend after interpolation of the candidate time point according to the adjustment coefficient, to obtain the final power load change trend of the candidate time point.

[0115] According to the adjustment coefficient of the i th time point of the day, the interpolated load target change trend of the i th time point of the day is adjusted to obtain the load final change trend of the i th time point of the day. In an exemplary embodiment, the product of the adjustment coefficient of the i th time point of the day and the interpolated load target change trend of the i th time point of the day is calculated, and the product is taken as the load final change trend of the i th time point of the day. By using the above process, the load final change trend of each time point of the day is obtained.

[0116] Step S3: Obtain the load burst index of each period of the day from the difference between the load final change trend of the same period of the day and the historical day.

[0117] Since the historical power load change trend of the intelligent power system is constrained by specific power consumption scenarios, for example, the historical power load change trend is the result of the joint action of specific time (such as weekdays / holidays) and external factors (such as weather, events, production plan). Therefore, the load burst index of each period of the day is obtained from the difference between the load final change trend of the same period of the day and the historical day, which is used to ensure the accuracy of the reference value of each power consumption scenario of the day for power load prediction.

[0118] For any day in the day and the historical day, the day is divided into multiple periods, and the number of periods obtained by the division is determined by the length of the period. In this embodiment, one period contains 60 time points. It should be understood that for any period, there is a corresponding relationship in the day and the historical day, such as the period: 10:00-11:00, which has a corresponding relationship in the day and the historical day.

[0119] In the process of analyzing whether different periods in the day belong to the same power consumption scenario, influenced by unexpected and accidental factors, such as extreme weather mutation, major incidents, equipment failure, and other sudden situations, there may be a burst power consumption scenario. At this time, the obtained power load change trend is affected by the burst scenario, and its change has no obvious rule, which does not have reference value in the process of short-term power load prediction. Therefore, the credibility of the period under the burst scenario is reduced to avoid its damage to the power consumption law of other power consumption scenarios in history, so as to improve the accuracy of the short-term load prediction result of the power system. In an exemplary embodiment, as shown in Figure 6 , a specific process for obtaining the load burst index is given as follows:

[0120] Step S31: Determine the difference degree of the load final change trend of the same period of each day in the historical day and the candidate period.

[0121] For ease of illustration, the candidate period is set as any period in the day, and each day in the historical day has the same period as the candidate period.

[0122] Obtain the final load change trend sequence for the candidate time period, which includes the final load change trend at each moment within the candidate time period. Similarly, obtain the final load change trend sequence for each day in the historical time period that is the same as the candidate time period.

[0123] The degree of difference between the final load change trend sequence of the candidate time period and the final load change trend sequence of the same time period for each day in the historical timeframe is obtained. In an exemplary embodiment, the DTW (Dynamic Time Warping) distance of the final load change trend sequence is obtained, and then the DTW distance is normalized (e.g., using...). Normalization is performed using a method where exp represents an exponential function with the natural constant as its base. The normalized result is used to represent the degree of difference in the final load change trend sequence. The greater the degree of difference, the less similar the final load change trend sequences are.

[0124] Step S32: Determine the reference day for the candidate time period.

[0125] Based on the degree of difference between the final load change trend sequence of the candidate period and the final load change trend sequence of the same period on each day in the historical timeframe, a reference day for the candidate period is determined. In an exemplary embodiment, a preset difference threshold is used to determine whether the degree of difference in the final load change trend sequence is large. The preset difference threshold ranges from 0 to 1, and the specific value is determined according to the actual judgment needs. For example, if a more stringent judgment mechanism is required, the preset difference threshold can be set to a smaller value, such as 0.4.

[0126] The difference between the final load change trend sequence of the candidate period and the final load change trend sequence of the same period of each day in the historical days is compared with the magnitude of the preset difference threshold. The historical days corresponding to the difference greater than the preset difference threshold are identified and defined as reference days. Thus, the reference days for the candidate period on the current day are obtained, and the number of reference days for the candidate period on the current day is obtained.

[0127] Step S33: Based on the number of days between the current day and each reference day, and the degree of difference between the current day and each reference day, obtain the abnormal performance of the candidate time period.

[0128] For each candidate period, the number of days between the current day and each reference day is determined. This interval is the difference between the current day's date and the reference day's date. For example, if the current day is October 10th and a reference day is October 1st, then the interval between the current day and that reference day is 10 - 1, which equals 9 days. The longer the interval between the current day and each reference day, the more random and sudden the power load change trend is during the candidate period, and the stronger the abnormal performance of the candidate period. The two are positively correlated.

[0129] The greater the difference between the current day and each reference day, the more random and sudden the change trend of the power load during the candidate period is, and the stronger the abnormal performance of the candidate period is. The two are positively correlated.

[0130] Therefore, based on the number of days between the current day and each reference day, and the degree of difference between the current day and each reference day, the abnormal performance of the candidate time period is obtained. Based on the above logic, the following method for obtaining abnormal performance is given:

[0131] ;

[0132] in, This indicates the abnormal behavior during the u-th time period of the day. This indicates the degree of difference between the current day of the u-th time period and the c-th reference day. This represents the number of days between the current day of the u-th time period and the c-th reference day. This indicates the number of days between the current day and the first day in the historical timeline. Indicates to The normalization is given by C, which represents the number of reference days for the u-th time period.

[0133] Step S34: Based on the abnormal performance and the proportion of reference days, obtain the load burst index for the candidate time period.

[0134] The ratio of the number of reference days for the current day to the total number of historical days is calculated as the percentage of the number of reference days for the current day. The larger the percentage of the number of reference days for the current day, the more historical days there are that differ significantly from the power load change trend of the current day in the candidate period. The more random and sudden the power load change trend of the candidate period is, the larger the load suddenness index of the candidate period will be. The load suddenness index is positively correlated with the percentage of the number of reference days.

[0135] The stronger the abnormal performance of a candidate time period, and the more likely the power load change trend of that period is a sudden and accidental event, the higher the load surge index for that time period will be. The load surge index is positively correlated with the abnormal performance. Based on this logic, the load surge index for the u-th time period of the day is calculated by multiplying the abnormal performance of that time period by the proportion of reference days for that time period. This process is repeated to obtain the load surge index for each time period of the day.

[0136] Step S4: Based on the correlation between the power load of any two time periods on the same day and the correlation between the final load change trend, as well as the abnormal indicators of any two time periods, obtain the similarity of the load change trend of any two time periods.

[0137] As the electricity consumption scenario determines the power load change trend, different electricity consumption scenarios will eventually lead to significant differences in the power load change trend. Conversely, if the power load change trends of two time periods of the day are highly similar, it means that the electricity consumption scenarios behind them are probably consistent. Therefore, taking any two time periods of the day as an example, the similarity of the load change trends of any two time periods is obtained according to the correlation of the power load of any two time periods and the correlation of the final load change trend, as well as the abnormal index of any two time periods. Thus, the similarity of the load change trends of any two time periods of the day is obtained by traversing all the time periods of the day. In an exemplary embodiment, as shown in FIG. 8, a specific process for obtaining the similarity of the load change trend is as follows: Figure 7

[0138] Step S41: Determine the average of the load burst index of any two time periods.

[0139] The average of the load burst index of any two time periods is calculated. The greater the average of the load burst index, the greater the possibility that the two time periods belong to burst electricity consumption, the lower the similarity of the load change trends of the two time periods, and the negative correlation between the similarity of the load change trend and the average of the load burst index.

[0140] Step S42: Obtain the similarity of the load change trend of any two time periods from the average of the load burst index of any two time periods, the correlation of the power load of any two time periods, and the correlation of the final load change trend.

[0141] The correlation of the power load of any two time periods is obtained. Among them, the u-th time period and the v-th time period of the day are taken as any two time periods of the day. The power load at each time of the u-th time period is obtained (here, the interpolated power load is obtained), thereby forming the power load sequence of the u-th time period in chronological order, and the power load sequence of the v-th time period is obtained in the same way. Then, the correlation of the power load sequences of the u-th time period and the v-th time period is obtained. In an exemplary embodiment, the DTW distance of the power load sequences of the u-th time period and the v-th time period is obtained, and then the DTW distance is negatively correlated normalized (such as using to negatively correlate normalize), and the result is the correlation of the power load sequences of the u-th time period and the v-th time period.

[0142] The correlation of the final load change trend sequence of the u-th time period and the final load change trend sequence of the v-th time period is obtained, and the DTW distance of the final load change trend sequences of the u-th time period and the v-th time period is obtained in the same way, and then the DTW distance is negatively correlated normalized, and the result is the correlation of the final load change trend sequences of the u-th time period and the v-th time period.

[0143] ​The stronger the correlation between the power load sequence of the u-th time period and the v-th time period, the stronger the load change trend similarity between the u-th time period and the v-th time period, and the two are positively correlated; the stronger the correlation between the load final change trend sequence of the u-th time period and the v-th time period, the stronger the load change trend similarity between the u-th time period and the v-th time period, and the two are positively correlated. Based on the above logic, one way to obtain the load change trend similarity between the u-th time period and the v-th time period is as follows:

[0144] ;

[0145] wherein, represents the load change trend similarity between the u-th time period and the v-th time period, represents the average of the load burst indicators of the u-th time period and the v-th time period, represents the correlation between the power load sequence of the u-th time period and the v-th time period, represents the correlation between the load final change trend sequence of the u-th time period and the v-th time period. The stronger the load change trend similarity between the u-th time period and the v-th time period, the more likely the u-th time period and the v-th time period belong to the same power consumption scene.

[0146] Step S5: Based on the load change trend similarity, the time periods of the day are classified into power consumption scenes, and according to the load change trend similarity in each category, the reference value of each category for power load prediction is obtained.

[0147] Since the power load change trend of the same power consumption scene often has stronger similarity, according to the load change trend similarity of any two time periods of the day, the power consumption scenes of each time period of the day are classified, which can reduce the influence of power load change trend in different power consumption scenes on the accuracy of the prediction result. In an exemplary embodiment, as shown in Figure 8 , the specific process of classifying the time periods of the day into power consumption scenes is as follows:

[0148] Step S51: Obtain the clustering distance of any two time periods from the load change trend similarity of the two time periods.

[0149] Taking the u-th time period and the v-th time period of the day as an example, the stronger the load change trend similarity between the two, the shorter the clustering distance between the two, therefore, the clustering distance is negatively correlated with the load change trend similarity. In an exemplary embodiment, the clustering distance between the u-th time period and the v-th time period is equal to . Thus, the clustering distance of any two time periods of the day is obtained.

[0150] Step S52: Based on the clustering distance of any two time periods, cluster each time period of the day to obtain several categories.

[0151] Based on the clustering distance between any two time periods, the K-means clustering algorithm is used to cluster the various time periods of the day, resulting in several clusters. Each cluster represents a category, thus creating several categories. This ensures that clusters with similar clustering distances are grouped together, meaning that similar load change trends are grouped into the same cluster. The K value in the K-means clustering algorithm can be set manually or determined using the elbow rule or silhouette coefficient method. Different categories represent different electricity consumption scenarios, and each category includes time periods within the same electricity consumption scenario, i.e., it includes power load data within the same electricity consumption scenario.

[0152] Then, based on the similarity of load change trends in each category, the reference value of each category for power load forecasting is obtained, such as... Figure 9 As shown, the following is a specific process for obtaining reference value:

[0153] Step S53: Obtain the minimum similarity of load change trends in each category, and the number of time periods included in each category.

[0154] With the first Each category represents any given category, i.e., any type of electricity usage scenario. (The remaining text appears to be incomplete and requires further context.) The minimum similarity of load change trends among all load change trend similarities included in each category. The larger the minimum similarity of load change trends, the stronger the similarity. Within each category of electricity consumption scenarios, the greater the similarity in load change trends between time periods, the higher the similarity. The stronger the similarity of load change trends under the corresponding electricity consumption scenarios for each category, the more valuable the prediction of future electricity load will be. In other words, the greater the reference value, the more positively correlated the reference value is with the minimum similarity of load change trends.

[0155] Get the The number of time periods included in each category; the more time periods, the better. The more valuable a category's prediction of future electricity load is, the greater its reference value; and the reference value is positively correlated with the number of time periods.

[0156] Step S54: Based on the minimum similarity of load change trends for each category and the number of time periods, obtain the reference value for each category.

[0157] According to the The minimum similarity of load change trends for each category and the number of time periods are used to obtain the [number of]th [category]. Based on the above logic, the reference value of each category is given as follows: How to obtain the reference value of each category:

[0158] ;

[0159] wherein, represents the reference value of the th category, represents the minimum value of the load change trend similarity of the th category, represents the number of time periods contained in the th category, represents the total number of time periods of the day, represents the normalization of .

[0160] By using the above process, the reference value of each category is obtained, i.e. the reference value of each power consumption scene of the day. The reference value of each time period in the same power consumption scene is equivalent to the reference value of the power consumption scene to which it belongs, and thus the reference value of each time period of the day is obtained. Furthermore, the reference value of each time instant in the same time period is equivalent to the reference value of the time period to which it belongs, and thus the reference value of each time instant of the day is obtained. The greater the reference value, the more important the role it plays in short-term power load forecasting.

[0161] In an exemplary embodiment, after obtaining the reference value of each category for power load forecasting, as shown in Figure 10 , the short-term power load forecasting method for smart power systems provided by the present embodiment further comprises the following steps:

[0162] Step S6: obtaining the weight of the power load of each time instant of the day from the reference value of the category in which each time instant of the day is located, and weighting the power load of each time instant of the day.

[0163] According to the reference value of the category in which each time instant of the day is located, the weight of the power load of each time instant of the day is obtained. In an exemplary embodiment, the sum of the reference values of all time instants of the day is calculated, and then the ratio of the reference value of each time instant of the day to the sum is calculated as the weight of the power load of each time instant of the day, so that the sum of the weights of the power load of each time instant of the day is 1 under the premise of meeting the logic.

[0164] According to the weight of the power load of each time instant of the day, the power load of each time instant of the day is weighted, and thus the power load of each time instant of the day is updated.

[0165] Step S7: performing short-term power load forecasting according to the weighted power load of each time instant of the day.

[0166] The power load prediction is performed according to the weighted power load at each time of the day. Taking the current time of the day as an example, based on the weighted power load at each time of the day, a weighted power load sequence before the current time is obtained, and the existing prediction algorithm is used, such as using an autoregressive moving average model, or using a deep learning algorithm (such as a multilayer perceptron, a recurrent neural network, a time convolution network, etc.), to obtain a power load prediction sequence of the current time, so that the power load of the next time of the current time can be obtained, and the short-term power load prediction is completed.

[0167] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0168] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment mainly describes the difference from other embodiments.

Claims

1. A short-term power load forecasting method for a smart power system, characterized by, The method comprises the following steps: respectively determining load target change trends before and after interpolation of power load in reference time periods of each time of the day; adjusting the load target change trend after interpolation according to the difference between the load target change trends before and after interpolation, to obtain a final load change trend of each time; obtaining a load burst indicator of each time period of the day from the difference between the final load change trends of the same time period of the day and historical days; obtaining a load change trend similarity of any two time periods according to the correlation of power load and the correlation of final load change trends of the two time periods, and the average value of the load burst indicators of the two time periods; classifying time periods of the day according to the load change trend similarity, and obtaining reference values of each category for power load prediction according to the load change trend similarity in each category; wherein, the process of obtaining the load burst indicator comprises: determining the difference degree of the final load change trends of the same time period of each day in the historical days from a candidate time period; the candidate time period is any time period of the day; determining reference days of the day for the candidate time period, the reference days are the historical days whose difference degree is greater than a preset difference threshold; obtaining an abnormal performance of the candidate time period according to the interval days of the day and each reference day, and the difference degree of the day and each reference day; the abnormal performance is positively correlated with the interval days and the difference degree; obtaining the load burst indicator of the candidate time period according to the abnormal performance and the proportion of the number of reference days; the load burst indicator is positively correlated with the abnormal performance and the proportion of the number of reference days; wherein, the process of obtaining the load change trend similarity of any two time periods comprises: determining the average value of the load burst indicators of the two time periods; obtaining the load change trend similarity of the two time periods from the average value of the load burst indicators of the two time periods, the correlation of power load and the correlation of final load change trends of the two time periods; the load change trend similarity is negatively correlated with the average value of the load burst indicators, and positively correlated with the correlation of power load and the correlation of final load change trends of the two time periods.

2. The smart power system oriented short-term power load forecasting method according to claim 1, wherein, The process of obtaining the load target change trend comprises: obtaining a load initial change trend of each time from the change of power load in the reference time period of each time of the day and historical days; determining an unstable time period of the load initial change trend of the day and historical days; obtaining a load change regular performance of each time of the day from the overlap of the unstable time period of the day and historical days, and the difference between the load initial change trends of the day and historical days; adjusting the load initial change trend to obtain the load target change trend according to the load change regular performance.

3. The smart power system oriented short-term power load forecasting method according to claim 2, wherein, The process of obtaining the load initial change trend comprises: determining extreme points of power load in the reference time period, the extreme points including maximum and minimum; determining the power load difference of adjacent extreme points, and the change intensity of power load between adjacent extreme points; fusing the power load difference and the change intensity of all adjacent extreme points in the reference time period to obtain a power load change trend performance; obtaining the load initial change trend according to the power load change trend performance and the proportion of the number of extreme points; the load initial change trend is positively correlated with the power load change trend performance and the proportion of the number of extreme points.

4. The smart power system oriented short-term power load forecasting method according to claim 2, wherein, The obtaining process of the load change regular performance includes: determining the number of unstable time periods in the day and the historical days at each time of the day, which represents the number of days in the day and the historical days; obtaining the load change regular performance of each time of the day according to the number of each time of the day and the load initial change trend difference between each time of the day and the same time of the historical days; the load change regular performance is positively correlated with the number and negatively correlated with the load initial change trend difference.

5. The smart power system oriented short-term power load forecasting method according to claim 1, wherein, The obtaining process of the load final change trend includes: determining the number of interpolation data as extreme points after interpolating the power load in the reference time period of the candidate time; the extreme points include maximum and minimum; the candidate time is any time of the day; obtaining the adjustment coefficient of the candidate time according to the number of interpolation data and the difference between the load target change trends before and after interpolation of the candidate time; the adjustment coefficient is negatively correlated with the number of interpolation data and the difference between the load target change trends before and after interpolation of the candidate time; adjusting the load target change trend after interpolation of the candidate time according to the adjustment coefficient to obtain the load final change trend of the candidate time.

6. The smart power system oriented short-term power load forecasting method according to claim 1, wherein, The power consumption scene classification of the time period of the day includes: obtaining the clustering distance of any two time periods from the load change trend similarity of any two time periods; the clustering distance is negatively correlated with the load change trend similarity; clustering each time period of the day based on the clustering distance of any two time periods to obtain several categories; different categories represent different power consumption scenes.

7. The smart power system oriented short-term power load forecasting method according to claim 1, wherein, The obtaining process of the reference value includes: obtaining the minimum value of the load change trend similarity in each category and the number of time periods contained in each category; obtaining the reference value of each category according to the minimum value of the load change trend similarity and the number of time periods of each category; the reference value is positively correlated with the minimum value of the load change trend similarity and the number of time periods.

8. The smart power system oriented short-term power load forecasting method according to claim 1, wherein, in the step of After obtaining the reference value of each category for power load prediction, the short-term power load prediction method for smart power system further includes: obtaining the weight of the power load of each time of the day from the reference value of the category in which each time of the day is located, and weighting the power load of each time of the day; performing short-term power load prediction according to the weighted power load of each time of the day.

Images