Power consumption demand prediction method and system based on big data

Abstract

The invention relates to the technical field of electric quantity prediction, in particular to an electricity consumption demand prediction method and system based on big data, and the method comprises the steps: obtaining the electricity consumption data of a target user in historical time, the electricity consumption data comprising the electricity consumption per day; taking any day as a target day, calculating the time dependence degree of the target day, and carrying out smoothing processing on all the electricity consumption of which the time dependence degree is greater than a set value to obtain target data; the target data and all the electricity consumption with the time dependence degree smaller than or equal to a set value form a data set; and inputting the data set into the prediction model to obtain predicted electric quantity of the target user in future time. According to the scheme, the electric quantity in the future time can be predicted, and data support is provided for subsequent power supply.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power prediction. More particularly, the present application relates to a power demand prediction method and system based on big data. BACKGROUND

[0002] With the application of smart meters and big data technology, the power system can collect massive power consumption data. Big data technology enables real-time updating and dynamic adjustment of power demand prediction. Through real-time data stream processing, power companies or enterprises can timely adjust power generation and transmission plans and optimize resource allocation.

[0003] For example, during peak power consumption periods, power companies can flexibly dispatch generating units based on real-time prediction results to avoid power supply shortages or surpluses.

[0004] When predicting power, an ARIMA model is used to predict power consumption. The autoregressive integrated moving average model (ARIMA model) is a statistical method for analyzing and predicting time series data. It combines three parts: autoregression (AR), difference (I), and moving average (MA). The autoregression part assumes a linear relationship between the current value and the past values; the difference part is used to handle non-stationarity in time series to make the sequence stable; the moving average part corrects the current prediction by using the past prediction errors.

[0005] Because the ARIMA model can capture trends, periodicity, and random fluctuations in time series, it is particularly good at handling non-seasonal, stationary time series. Therefore, the ARIMA model can be used to predict future power demand for several days or weeks, combined with the target user's power consumption data.

[0006] However, the target user's power consumption data often has some uncertainty and instability, i.e., it is affected by seasonal fluctuations and periodic fluctuations. These factors can cause data to be non-stationary, with complex time dependencies, making it difficult to eliminate fluctuations in non-stationary time series using the ARIMA model, and thus leading to inaccurate power demand prediction. SUMMARY

[0007] The present application aims to provide a power demand prediction method and system based on big data to solve the problem of inaccurate power demand prediction using existing prediction models. To this end, the present application provides solutions in the following two aspects.

[0008] In the first aspect, the present application provides a power demand prediction method based on big data, comprising: obtain power consumption data of the target user in historical time, the power consumption data including power consumption in different days; divide each day in the historical time into a working day category and a non-working day category; take any day as a target day, calculate a time dependence degree of the target day, the time dependence degree being positively correlated with a sum of a seasonal fluctuation degree and a periodic fluctuation degree of the corresponding target day; the seasonal fluctuation degree representing seasonal trend change of the target day; the periodic fluctuation degree being positively correlated with a first difference and a first difference degree; the first difference being an absolute value of a difference between power consumption of the target day and a mean value of power consumption of all days belonging to one category of the target day, and the first difference degree representing difference change between power consumption of the target day and a mean value of power consumption of all days of another category; smooth all power consumptions with a time dependence degree greater than a set value to obtain target data; the target data and all power consumptions with a time dependence degree less than or equal to the set value constituting a data set; input the data set into an ARIMA model to obtain predicted power consumption of the target user in future time.

[0009] The above scheme combines regularity characteristics of power consumption data of the target user in historical time, obtains seasonal fluctuation degree and periodic fluctuation degree of power consumption of each day, selects power consumption with greater time dependence according to the seasonal fluctuation degree and the periodic fluctuation degree of power consumption of each day, and smooths the selected power consumption, thereby improving accuracy of subsequent power demand prediction.

[0010] Optionally, the seasonal fluctuation degree is: ; wherein, is power consumption of the target user on the mth day in the ath month, is a fitting slope value of power consumption data of the ath month, and m represents a serial number, is a fitting slope of a fitting curve of power consumption data in a local time sequence segment containing power consumption on the mth day in the ath month, is expected power consumption of the mth day in the ath month in the fitting curve, is a maximum power consumption in power consumption data of the ath month, is a mean value of fitting slope values of the ath month, the bth month and the cth month; the bth month and the cth month belong to a seasonal feature with the ath month, and the fitting slope value is obtained by a least square method.

[0011] ​This approach comprehensively considers changes in overall, short-term, and long-term trends, providing a more comprehensive picture of the fluctuations in electricity consumption. This approach not only considers the impact of a single factor, but also the interactions of multiple factors, thus more comprehensively reflecting the complex temporal structure of electricity consumption.

[0012] Optionally, the degree of periodic fluctuation when the i-th day belongs to the category of weekday for: ; in, 、 are the first mean of the target user’s electricity consumption data on all working days in month a, and the second mean of the electricity consumption data on all non-working days, is the electricity consumption on the i-th day in month a, is the absolute value of the difference between the first mean and the second mean, and norm() is the normalization function.

[0013] The above solution obtains the periodic changes within the working day by analyzing the changes in the target user's electricity consumption on the working day.

[0014] Optionally, the degree of periodic fluctuation when the i-th day belongs to the category of non-working day for: ; in, 、 are the first mean of the target user’s electricity consumption data on all working days in month a, and the second mean of the electricity consumption data on all non-working days, is the absolute value of the difference between the first mean and the second mean, norm() is the normalization function, is the electricity consumption on the i-th day in the a-th month.

[0015] The above solution can obtain the periodic changes of non-working days by analyzing the changes in electricity consumption of target users on non-working days.

[0016] Optionally, it also includes denoising the electricity consumption data within a historical period.

[0017] Optionally, the seasonal characteristics are spring, summer, autumn or winter, wherein spring is from March to May, summer is from June to August, autumn is from September to November, and winter is from December to February of the following year.

[0018] Optionally, the time dependence when the i-th day belongs to the category of weekday for: ;in, a periodic fluctuation degree when the i-th day of the a-th month belongs to the category of weekdays, a seasonal fluctuation degree of the i-th day of the a-th month.

[0019] The above scheme provides a method for accurately calculating the time dependence degree.

[0020] Optionally, the process of smoothing all power consumptions with a time dependence degree greater than a set value to obtain target data is: The key sequence is smoothed by using a moving average method to obtain target data.

[0021] Optionally, it further includes a step of judging the stability of the data set, specifically: performing difference processing on the data set to obtain a difference sequence, performing ADF test on the difference sequence, and when the test value is less than a set threshold, considering that the data set is stationary; otherwise, the data set is not stationary, and second-order difference processing is performed on the difference sequence and retested or the target data is smoothed.

[0022] In a second aspect, the present application provides a power consumption demand prediction system based on big data, comprising: a processor; a memory storing computer instructions for power consumption demand prediction based on big data, when the computer instructions are run by the processor, the system executes the above-mentioned power consumption demand prediction method based on big data.

[0023] The present application has the following advantages: The present application can analyze the power consumption data of the target user in the historical time to obtain the regular change of the power consumption of the target user in the historical time, determine the power consumption with greater fluctuation, process it, obtain a more stable data set, and then predict the power consumption of the target user, and provide corresponding strategies for subsequent power supply. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 The step flowchart of the power consumption demand prediction method based on big data in the embodiment is schematically shown; Figure 2 The structural block diagram of the power consumption demand prediction system based on big data in the embodiment is schematically shown. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. The described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0026] Specifically, if Figure 1 As shown, the electricity demand forecasting method based on big data provided in this embodiment includes the following steps: Step S1: Obtain the daily electricity consumption of the target user in the historical period to obtain electricity consumption data.

[0027] The target users may be home users, enterprises or production plants, etc.

[0028] In this embodiment, the daily electricity consumption of the target user is collected by a smart meter, and the electricity consumption is stored in the form of a time series to obtain electricity consumption data.

[0029] Furthermore, the collected electricity consumption data is subjected to denoising and interpolation. For example, the data is checked for missing values ​​and linear interpolation is used to fill in the missing values. A mean filter algorithm is used to remove noise from the electricity consumption data and reduce the impact of extreme values.

[0030] Among them, the linear interpolation method and the mean filtering algorithm are both existing technologies and will not be described in detail here.

[0031] The above historical time can be a month of collection or part of a month. For example, if the historical time is November, the electricity consumption data will be composed of the electricity consumption from the 1st to the 30th. Of course, the historical time can also be from the 12th to the 30th of November, in which case the electricity consumption data will be composed of the electricity consumption from the 12th to the 30th.

[0032] Step S2: Obtain a data set.

[0033] In this embodiment, it is taken into account that the electricity consumption data of the target user often has a certain degree of uncertainty, making it difficult to predict the subsequent electricity consumption of the target user. Therefore, it is necessary to analyze the electricity consumption data of the target user to obtain information that can characterize the seasonal and periodic characteristics of the target user's electricity consumption, obtain the degree of time dependence, and then filter out the electricity consumption data with a large degree of time dependence, and smooth it to obtain the target data to eliminate the instability caused by time. The target data and the electricity consumption data with a small degree of time dependence are combined into a data set to predict the subsequent electricity consumption of the target user, so that the prediction is as consistent as possible with the electricity consumption of the target user, and provide data support for power generation or power supply.

[0034] In this embodiment, the process of obtaining the time dependence degree includes steps S21-S22, specifically: Step S21 : Taking any day as a target day, calculating the seasonal fluctuation degree and the cyclical fluctuation degree of the target day.

[0035] The process of obtaining the seasonal fluctuation degree is as follows: First, the seasonal characteristics of the target month are determined, and the other two months corresponding to the seasonal characteristics are obtained according to the target month.

[0036] In this embodiment, the seasonal characteristics are spring, summer, autumn or winter. The electricity consumption in summer (June-August) and winter (December-February) is affected by air conditioning and other equipment, and the electricity consumption is relatively high. The climate is suitable in spring (March-May) and autumn (September-November), and the electricity consumption is relatively low.

[0037] Specifically, taking the a-th month as an example, the b-th month and the c-th month belong to the same seasonal characteristics. For example, when the target month is October, the other two months corresponding to the seasonal characteristics are September and November.

[0038] Secondly, the electricity consumption of each day in the other two months is obtained, and the least square method is used to curve fit all the electricity consumption in the three months to obtain the fitting slope value of each month.

[0039] For example, the fitting slope value of the electricity consumption data in the a-th month, the b-th month and the c-th month is obtained by using the least square method, respectively denoted as , , The average value is denoted as to reflect the long-term trend change of the electricity consumption in the season; wherein the fitting slope value of each month reflects the trend change of the overall electricity consumption data of the corresponding month.

[0040] Then, the electricity consumption data in the local time sequence segment containing the target day is obtained, and linear fitting is performed to obtain the fitting slope of the fitting curve.

[0041] The local time sequence segment can be the electricity consumption data within 5 days, or the electricity consumption data within 7 days or more. By obtaining the fitting slope in the local time sequence segment, the short-term trend change of the electricity consumption of the target day is reflected.

[0042] The target day in the local time sequence segment can be located at the center of the local time sequence segment, or at the end.

[0043] Finally, the seasonal fluctuation degree of the target day is obtained.

[0044] Specifically, the seasonal fluctuation degree is: ; Wherein, is the electricity consumption of the user on the m-th day in the a-th month, is the fitting slope value of the electricity consumption data of the a-th month, and m represents the serial number. is the fitting slope of the fitting curve of the electricity consumption data in the partial time sequence segment of the electricity consumption of the i-th day in the a-th month, is the fitting slope of the fitting curve of the electricity consumption data in the partial time sequence segment of the electricity consumption of the i-th day in the a-th month, is the expected electricity consumption of the m-th day in the a-th month in the fitting curve, is the maximum electricity consumption in the electricity consumption data of the a-th month, is the mean of the sum of the fitting slope values of the a-th month, the b-th month and the c-th month.

[0045] wherein, represents the absolute deviation between the actual electricity consumption and the expected electricity consumption of the i-th day. This part reflects the deviation degree of the electricity consumption data from the overall trend of the a-th month, wherein, i.e. is the expected electricity consumption obtained according to the fitting curve, is the intercept of the fitting curve.

[0046] represents the difference between the fitting slope of the m-th day (short-term slope) and the fitting slope value of the a-th month (overall slope), which reflects the deviation between the short-term trend and the overall trend. represents the difference between the fitting slope of the m-th day and the mean of the fitting slope values within the seasonal characteristics (long-term trend slope), which reflects the deviation between the short-term trend and the long-term trend within the seasonal characteristics. represents the normalization processing, which can use the maximum and minimum value normalization, etc. The use of absolute value above can avoid the mutual offset of positive and negative deviations, and more accurately reflect the deviation degree.

[0047] In the embodiment, before calculating the periodic fluctuation degree of the target day, it is also necessary to divide each day in the collected historical time to obtain the category of working days and the category of non-working days, and then determine the category to which the target day belongs.

[0048] For example, when the i-th day belongs to the category of working days, the periodic fluctuation degree is: ; wherein, , are respectively the first mean of the electricity consumption data of all working days in the a-th month and the second mean of the electricity consumption data of all non-working days in the a-th month, is the electricity consumption of the i-th day in the a-th month, is the absolute value of the difference between the first mean and the second mean, and norm() is a normalization function.

[0049] wherein, a difference between the power consumption when the i-th day belongs to the working day category and the first mean value (average power consumption of working days), the smaller the value, the closer the power consumption of the day to the first mean value. a difference between the power consumption when the i-th day belongs to the working day category and the second mean value of non-working days. a change in the difference between the power consumption when the i-th day belongs to the working day category and the second mean value of non-working days, the closer the value to 0, the smaller the deviation of the power consumption of the day from the absolute value of the difference between the first mean value and the second mean value, and the more stable the periodic fluctuation degree; otherwise, the more unstable. a normalization process can be used, such as max-min normalization.

[0050] When the i-th day belongs to the non-working day category, the periodic fluctuation degree is: ; wherein, , the first mean value of the power consumption data of all working days in the a-th month and the second mean value of the power consumption data of all non-working days of the target user, respectively, the absolute value of the difference between the first mean value and the second mean value, norm() is a normalization function, the power consumption of the i-th day in the a-th month.

[0051] wherein, a difference between the power consumption when the i-th day belongs to the non-working day category and the second mean value of non-working days, the smaller the value, the closer the power consumption of the day to the second mean value of non-working days. a difference between the power consumption when the i-th day belongs to the non-working day category and the first mean value of working days. a change in the difference between the power consumption when the i-th day belongs to the non-working day category and the first mean value; the closer the value to 0, the smaller the deviation of the power consumption of the day from the absolute value of the difference between the first mean value and the second mean value, and the more stable the periodic fluctuation degree; otherwise, the more unstable. a normalization process can be used, such as max-min normalization.

[0052] The above purpose of obtaining the seasonal fluctuation degree and the periodic fluctuation degree is to identify and quantify the seasonal fluctuation and the periodic fluctuation in the power consumption data of the user by in-depth analysis, and then accurately understand the daily power consumption change rule of the target user, thereby providing a reliable basis for subsequent prediction and adjustment.

[0053] Step S22, obtaining the time dependence degree of the target day based on the seasonal fluctuation degree and the periodic fluctuation degree.

[0054] In this embodiment, the time-dependent degree corresponding to the daily power consumption is obtained by the seasonal fluctuation degree and the periodic fluctuation degree.

[0055] Specifically, when the i-th day of the a-th month belongs to the working day category, the time-dependent degree of the power consumption of the target user on the i-th day of the a-th month is ; wherein, is the periodic fluctuation degree when the i-th day of the a-th month belongs to the working day category, is the seasonal fluctuation degree of the i-th day of the a-th month.

[0056] When the i-th day of the a-th month belongs to the working day category, the time-dependent degree of the power consumption of the target user on the i-th day of the a-th month is , wherein, is the periodic fluctuation degree when the i-th day of the a-th month belongs to the non-working day category, is the seasonal fluctuation degree of the i-th day of the a-th month.

[0057] Wherein, the greater the seasonal fluctuation degree and the periodic fluctuation degree, the greater the corresponding time-dependent degree, that is, the greater the influence of time, and the corresponding data shows the characteristic of instability.

[0058] After obtaining the time-dependent degree, the power consumption data in the historical time is also filtered, specifically: in response to the time-dependent degree being greater than a set value, the corresponding power consumption is taken as the key power, and then the key sequence is obtained.

[0059] The set value is 0.5, that is, when the time-dependent degree is greater than 0.5, the power consumption is taken as the key power.

[0060] Then, the key sequence is smoothed to obtain the target data. The smoothing method can use the moving average method. Illustratively, the process of smoothing is: the average value of the left and right two power consumptions adjacent to any key power in the key sequence is taken as the value of any key power, and for the first and last two key powers in the key sequence, for the first key power, the average value of the two key powers after the key power is replaced; for the last key power, the average value of the two key powers before the key power is replaced.

[0061] In this embodiment, by screening out the target data, it can be determined which days of power consumption are more dependent on time, and then the overall prediction will be greatly affected, so it needs to be paid attention to and adjusted, which helps to improve the calculation efficiency.

[0062] ​​Further, in the embodiment, the data set is also tested to determine the stationarity of the data set, specifically including: The data set is differentiated to obtain a difference sequence, and ADF test is performed on the difference sequence, when the test value is less than the set threshold value, it is considered that the data set has been stationary, otherwise, the data set is not stationary, and the smoothing processing of the target data needs to be continued or the second-order differentiation is continued and tested again.

[0063] The threshold value is set to 0.05.

[0064] The first-order differentiation is the operation of the difference between each power consumption of the data set and the previous power consumption. The second-order differentiation is the differentiation operation on the difference sequence after the first-order differentiation.

[0065] The ADF test is an effective method for judging the stationarity of a time series. Specifically, by calculating the first-order differentiation, estimating the regression equation, calculating the test statistic and comparing it with the critical value, whether the time series is stationary can be accurately judged, which provides strong support for subsequent time series analysis. Since the ADF test is prior art, it will not be described in detail here.

[0066] In step S3, the data set is input into the prediction model to obtain the predicted power consumption of the target user in the future time.

[0067] In the embodiment, the data set is input into the ARIMA model to obtain the predicted power consumption of the target user in the future time, which provides data support for subsequent power supply or power generation strategy.

[0068] Since the ARIMA model is prior art, the specific implementation process will not be described here.

[0069] The above scheme can obtain more stable data sets, thereby improving the accuracy and stability of the power consumption prediction of the target user.

[0070] In the present application, the meaning of "a plurality of" is at least two, such as two, three or more, unless otherwise explicitly specified.

[0071] The scheme of the present application can analyze the power consumption data collected in the historical time to screen out power consumption data with a large time dependence degree, and perform smoothing processing on the power consumption data to eliminate the influence of complex time, obtain relatively stationary data, and then use the data set as the ARIMA model to predict the power consumption of the target user.

[0072] The present application also provides a power demand prediction system based on big data. Figure 2As shown, the system comprises a processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement the big data based electricity demand prediction method according to the present application.

[0073] The system also comprises a communication bus and a communication interface and other components well known to those skilled in the art, the settings and functions of which are known in the art, and thus will not be described here.

[0074] In the present application, the aforementioned memory can be any tangible medium containing or storing a program which can be used by or in conjunction with an instruction execution system, apparatus or device. For example, the computer readable storage medium can be any appropriate magnetic storage medium or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random-Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC) and the like, or any other medium that can be used to store desired information and can be accessed by an application, module or both. Any such computer storage medium can be part of the device or accessible or connectable to the device. Any application or module described in the present application can be implemented by computer readable / executable instructions stored or otherwise held by such computer readable medium.

[0075] Although the specification has described a number of embodiments of this application, only in an exemplary manner. Various alternatives to the embodiments of the application described herein can be employed in practice.

Claims

1. A method for predicting electricity demand based on big data, characterized in that, The method comprises the following steps: obtaining the power consumption data of the target user in the historical time, wherein the power consumption data comprises power consumption in different days; dividing each day in the historical time into a working day category and a non-working day category; taking any day as a target day, calculating the time dependence degree of the target day, wherein the time dependence degree is positively correlated with the sum of the seasonal fluctuation degree and the periodic fluctuation degree of the corresponding target day; the seasonal fluctuation degree represents the seasonal trend change of the target day; the periodic fluctuation degree is positively correlated with the first difference and the first difference degree; the first difference is the absolute value of the difference between the power consumption of the target day and the average power consumption of all days in one category to which the target day belongs; and the first difference degree represents the difference change between the power consumption of the target day and the average power consumption of all days in another category; performing smoothing processing on all power consumptions with a time dependence degree greater than a set value to obtain target data; and the target data and all power consumptions with a time dependence degree less than or equal to the set value constitute a data set; inputting the data set into an ARIMA model to obtain the predicted power consumption of the target user in the future time.

2. The big data based electricity demand prediction method of claim 1, wherein, The degree of seasonal fluctuation Is: ； wherein, is the power consumption of the target user on the mth day in the ath month, is the fitting slope value of the power consumption data of the ath month, and m represents the serial number, is the fitting slope value of the fitting curve of the linear fitting of the power consumption data in the local time sequence segment containing the power consumption of the mth day in the ath month, is the fitting slope of the fitting curve of the linear fitting of the power consumption data in the local time sequence segment containing the power consumption of the mth day in the ath month, is the expected power consumption of the mth day in the ath month in the fitting curve, is the maximum power consumption in the power consumption data of the ath month, is the average of the sum of the fitting slope values of the ath month, the bth month and the cth month; the bth month and the cth month belong to a seasonal feature with the ath month, and the fitting slope values are obtained by the least square method.

3. The big data based electricity demand prediction method of claim 2, wherein, the degree of periodic fluctuation when the day i belongs to the category of weekdays is: ； wherein, , are respectively the first mean of the electricity consumption data of the target user in all working days in the a-th month, the second mean of the electricity consumption data in all non-working days, is the electricity consumption on the i-th day in the a-th month, is the difference absolute value of the first mean and the second mean, and norm() is a normalization function.

4. The big data based electricity demand prediction method of claim 2, wherein, when the ith day belongs to the category of non-working days, the periodic fluctuation degree is: ； wherein, , are respectively a first mean of the power consumption data of the target user on all working days in the a-th month, a second mean of the power consumption data of the target user on all non-working days in the a-th month, is a difference absolute value of the first mean and the second mean, norm() is a normalization function, is the power consumption on the i-th day in the a-th month.

5. The big data based electricity demand prediction method of claim 1, wherein, The method further comprises denoising the power consumption data in the historical time.

6. The big data based electricity demand prediction method of claim 2, wherein, The seasonal feature is spring, summer, autumn or winter, wherein spring is from March to May, summer is from June to August, autumn is from September to November, and winter is from December to February of the next year.

7. The big data based electricity demand prediction method of claim 3, wherein, Day i is a time-dependent degree when it belongs to the weekday category is: ； wherein, is the degree of periodic fluctuation for the category of the ith day of the ath month when it is a weekday, is the degree of seasonal fluctuation for the ith day of the ath month.

8. The big data based electricity demand prediction method of claim 1, wherein, The process of performing smoothing processing on all power consumptions with a time dependence degree greater than a set value to obtain target data comprises the following steps: taking all power consumptions with a time dependence degree greater than a set value as a key sequence; performing smoothing processing on the key sequence by using a moving average method to obtain target data. 9.The big data based electricity demand forecasting method according to claim 1 or 8, characterized in that, The method further comprises a step of judging the stability of the data set, specifically comprising the following steps: performing difference processing on the data set to obtain a difference sequence, performing ADF test on the difference sequence, and when the test value is less than a set threshold, considering that the data set is stable; otherwise, the data set is unstable, and second-order difference processing needs to be performed on the difference sequence and retested or the target data needs to be smoothed.

10. A power demand forecasting system based on big data, characterized by, The system comprises: a processor; a memory storing computer instructions for big data-based power consumption demand prediction, wherein when the computer instructions are run by the processor, the system performs the big data-based power consumption demand prediction method according to any one of claims 1-9.

Images