A new online load forecasting method for power systems

Abstract

The present invention provides a novel online load prediction method for a power system, and relates to the technical field of power load prediction. The method comprises: obtaining output power data, as well as temperature data, humidity data, wind data and light data; determining a first relationship function; obtaining predicted load data through a load prediction model; determining whether the load prediction model needs to be corrected according to the predicted load data and the output power data; if correction is required, obtaining a corrected load prediction model according to the start time of the current detection cycle, output power data, predicted load data, temperature data, humidity data, wind data and light data; obtaining predicted light data according to the first relationship function and weather forecast data; inputting the weather forecast data and the predicted light data into the load prediction model to obtain predicted load data. According to the present invention, the adaptability of the load prediction model to the weather environment can be improved, and the accuracy of the predicted load data can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power load forecasting, and in particular to a novel online forecasting method for electric power system load. Background Art

[0002] In the related art, CN111428926A discloses a regional power load forecasting method considering meteorological factors. Considering the influence of multiple meteorological factors, a new method for power load forecasting is proposed to solve the shortcomings of existing regional power load forecasting technology, add the influence of actual regional meteorological factors, and give full play to the advantages of the forecasting model. The Elman neural network forecast that meets the accuracy requirements is selected, and three meteorological factor processing methods are considered for comparative research. Finally, through the comparison of prediction accuracy, an appropriate method is selected for the prediction of regional residential power load and regional total load. Compared with the single Elman neural network model, the model after considering meteorological factors has different degrees of improvement in prediction accuracy in different scenarios.

[0003] CN115719106A discloses a method for predicting electric load based on multiple meteorological elements, comprising the following steps: step one, obtaining electric load data and meteorological data; step two, decomposing and processing the load data; and step three, calculating the correlation between each meteorological data and the meteorological load by time period through the maximum information coefficient. The advantages of the present invention are as follows: since step two decomposes and processes the load data, and calculates the correlation between each meteorological data and the meteorological load by time period through the maximum information coefficient, the correlation characteristics between the meteorological data and the electric load can be grasped more intuitively, and appropriate influencing factors and a better prediction model can be automatically selected according to the power grid and time period where the prediction target is located, the regularity of historical load data is fully utilized, and the influence of different factors such as weather and user characteristics is combined to improve the efficiency and accuracy of the prediction, and improve the economic and social benefits, which can be increased by 3 to 5 percentage points compared with the traditional time series prediction model.

[0004] Based on the above related technologies, the accuracy of power load forecasting can be improved through meteorological data. However, the meteorological data in the related technologies are long-term and seasonal meteorological data, so the long-term and seasonal changes in power load can be predicted based on meteorological data. However, in the short-term weather changes, for example, rain, snow, wind and other weather within a day may also have an impact on the power load. For example, when it rains, it is often accompanied by cloudy days and dimming of light, which increases the lighting demand and increases the power load. However, it is difficult for the related technologies to accurately consider the impact of real-time weather factors on the power load, and thus it is difficult to ensure the accuracy of the power load forecast results.

[0005] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0006] The present invention provides a novel online load prediction method for a power system, which can solve the technical problems of difficulty in accurately considering the impact of real-time weather factors on power load and difficulty in ensuring the accuracy of power load prediction results.

[0007] According to a first aspect of the present invention, a novel method for online prediction of power system load is provided, comprising:

[0008] At multiple moments in the current detection cycle, the output power data of the transformer in the substation area is obtained;

[0009] At multiple moments in the current detection cycle, temperature data, humidity data, wind data, and light data of sampling locations within the power supply area corresponding to the substation transformer are obtained;

[0010] Determine a first relationship function between temperature data, humidity data, wind data and light data in the power supply area;

[0011] The start time of the current detection cycle, the temperature data, the humidity data, the wind data and the light data are processed by the trained load prediction model to obtain the predicted load data at multiple times of the current detection cycle;

[0012] Determining whether the trained load prediction model needs to be corrected according to the predicted load data at multiple moments in the current detection cycle and the output power data;

[0013] If the trained load forecasting model needs to be corrected, the trained load forecasting model is corrected and trained according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data to obtain a corrected load forecasting model;

[0014] According to the first relationship function and the weather forecast data of the next detection cycle in the power supply area, the predicted light data at multiple moments of the next detection cycle are obtained, wherein the weather forecast data includes the predicted temperature data, the predicted humidity data and the predicted wind speed data at multiple moments of the next detection cycle;

[0015] The end time of the current detection cycle, the weather forecast data of the next detection cycle and the predicted light data at multiple times of the next detection cycle are input into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle.

[0016] According to a second aspect of the present invention, a novel online load prediction system for a power system is provided, comprising:

[0017] The output power module is used to obtain the output power data of the transformer in the substation at multiple times in the current detection cycle;

[0018] A sampling data module, used to obtain temperature data, humidity data, wind data and light data of sampling locations in the power supply area corresponding to the transformer in the area at multiple moments in the current detection cycle;

[0019] A relationship function module, used to determine a first relationship function between the temperature data, humidity data, wind data and light data in the power supply area;

[0020] A prediction model module, used to process the start time of the current detection cycle, the temperature data, the humidity data, the wind data and the light data through a trained load prediction model to obtain predicted load data at multiple times of the current detection cycle;

[0021] A correction determination module is used to determine whether the trained load prediction model needs correction based on the predicted load data at multiple moments in the current detection cycle and the output power data;

[0022] A correction training module, for, if the trained load forecasting model needs correction, performing correction training on the trained load forecasting model according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data, to obtain a corrected load forecasting model;

[0023] a light prediction module, configured to obtain predicted light data at multiple moments of the next detection cycle according to the first relationship function and weather forecast data of the next detection cycle in the power supply area, wherein the weather forecast data includes predicted temperature data, predicted humidity data and predicted wind speed data at multiple moments of the next detection cycle;

[0024] The load prediction module is used to input the end time of the current detection cycle, the weather forecast data of the next detection cycle and the predicted light data at multiple times of the next detection cycle into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle.

[0025] According to a third aspect of the present invention, a new type of online load prediction device for an electric power system is provided, comprising: a processor; a memory for storing processor executable instructions; wherein the processor is configured to call the instructions stored in the memory to execute the new type of online load prediction method for an electric power system.

[0026] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the novel method for online prediction of loads in a power system.

[0027] Technical effect: According to the present invention, the relationship between temperature data, humidity data, wind data and light data can be accurately analyzed during short-term weather changes, and then the accuracy of short-term power load forecasting can be improved based on the relationship. When predicting load data, the load forecasting model can be corrected based on the impact of temperature data, humidity data, wind data and light data on power supply demand, thereby improving the adaptability of the load forecasting model to the current weather environment and the objectivity and accuracy of the load forecasting results. When determining the first relationship function, the first relationship function is obtained based on the first benchmark light data, temperature data, humidity data, wind data and light data, which accurately describes the relationship between light data and temperature data, graduation data and wind data. The periodicity and overall trend of light data changes are taken into account during the calculation process, thereby improving the accuracy, scientificity and objectivity of the first relationship function. When training the load prediction model, the influence of the temperature data, wind data, humidity data and light data on the power demand can be used to determine the influence of the above data on the error of the predicted load data, so as to set the training loss function based on the influence, the relative error of the predicted load data and the first weight representing the total power consumption data of each detection cycle, so that the load prediction model reduces the training loss function during the training process, and makes the trained load prediction model more suitable for the prediction of the power load under the weather environment of the current date, and improves the accuracy of the load prediction model more targeted. When correcting the load prediction model, the influence of the temperature data, wind data, humidity data and light data on the power demand can be used to determine the influence of the above data on the error of the predicted load data, so as to set the loss function based on the influence, the relative error of the predicted load data and the first weight, so that the load prediction model reduces the loss function during the correction training process, and makes the corrected load prediction model more suitable for the prediction of the power load at each moment of the current detection cycle, and improves the accuracy of the load prediction model more targeted, so that the corrected load prediction model can adapt to the rapidly changing weather environment and maintain a high prediction accuracy in the rapidly changing weather environment. When predicting the lighting data of the next detection cycle, the predicted lighting data at multiple moments of the next detection cycle can be obtained based on the second benchmark lighting data, predicted temperature data, predicted humidity data, predicted wind data and the first relationship function. The lighting data can be predicted based on the relationship between lighting and temperature, humidity, wind speed and other data, rather than predicting the lighting data only based on time as in the related art. This can improve the accuracy, scientificity and objectivity of the predicted lighting data and provide a data basis for subsequent load forecasting model predictions.

[0028] It should be understood that the above general description and the following detailed description are exemplary and explanatory only and do not limit the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other embodiments can be obtained based on these drawings without creative work.

[0030] Figure 1 A schematic diagram of a flow chart of a novel online load prediction method for a power system according to an embodiment of the present invention is exemplarily shown;

[0031] Figure 2 A block diagram of a novel power system load online prediction system according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0034] Figure 1 A schematic flow chart of a novel method for online load prediction of a power system according to an embodiment of the present invention is exemplarily shown, wherein the method comprises:

[0035] Step S101, obtaining output power data of the transformer in the substation at multiple moments in the current detection cycle;

[0036] Step S102, obtaining temperature data, humidity data, wind power data and light data of sampling locations in the power supply area corresponding to the transformer in the substation at multiple times in the current detection cycle;

[0037] Step S103, determining a first relationship function among temperature data, humidity data, wind data and light data in the power supply area;

[0038] Step S104, using the trained load prediction model, the start time of the current detection cycle, the temperature data, the humidity data, the wind data and the light data are processed to obtain predicted load data at multiple times of the current detection cycle;

[0039] Step S105, determining whether the trained load prediction model needs to be corrected according to the predicted load data at multiple moments in the current detection cycle and the output power data;

[0040] Step S106, if the trained load forecasting model needs to be corrected, the trained load forecasting model is corrected and trained according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data to obtain a corrected load forecasting model;

[0041] Step S107, obtaining predicted light data at multiple moments in the next detection cycle according to the first relationship function and the weather forecast data of the next detection cycle in the power supply area, wherein the weather forecast data includes predicted temperature data, predicted humidity data and predicted wind speed data at multiple moments in the next detection cycle;

[0042] Step S108, input the end time of the current detection cycle, the weather forecast data of the next detection cycle and the predicted light data at multiple times of the next detection cycle into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle.

[0043] According to the novel online load prediction method for a power system according to an embodiment of the present invention, the relationship between temperature data, humidity data, wind data and light data can be accurately analyzed during short-term weather changes, and the accuracy of short-term power load prediction can be improved based on the relationship. When predicting load data, the load prediction model can be corrected based on the impact of temperature data, humidity data, wind data and light data on power supply demand, thereby improving the adaptability of the load prediction model to the current weather environment and improving the objectivity and accuracy of the load prediction results.

[0044] According to an embodiment of the present invention, in step S101, output power data of the transformer in the substation is obtained at multiple moments in the current detection cycle.

[0045] For example, a detection cycle is 1 hour, and the output power data of the substation transformer is obtained every half an hour. In one detection cycle, 120 output power data can be obtained.

[0046] According to an embodiment of the present invention, in step S102, at multiple moments in the current detection cycle, temperature data, humidity data, wind data and light data of sampling locations in the power supply area corresponding to the substation transformer are obtained.

[0047] For example, the sampling location is not affected by indoor temperature, humidity, wind and light, and can accurately reflect outdoor weather conditions. For example, the sampling location is the location with the lowest housing density in the power supply area. The temperature data, humidity data, wind data and light data of the sampling location are obtained, and the outdoor temperature, humidity, wind and light data can be accurately collected.

[0048] According to an embodiment of the present invention, in step S103, a first relationship function among temperature data, humidity data, wind data and light data in the power supply area is determined.

[0049] For example, temperature data, humidity data, wind data, and light data are all related to weather conditions to a certain extent. For example, in rainy and snowy weather, temperature data and humidity data will be higher, and light data will be lower. Based on the correlation of the above data, a first relationship function between temperature data, humidity data, wind data, and light data in the power supply area can be determined.

[0050] According to one embodiment of the present invention, a first relationship function among the temperature data, humidity data, wind data and light data in the power supply area is determined based on temperature data and light data at multiple moments, including: obtaining first historical light data at the same moment as the start moment of a current detection cycle among multiple historical dates; determining first benchmark light data at the start moment of the current detection cycle based on the first historical light data; and obtaining the first relationship function based on the first benchmark light data, the temperature data, the humidity data, the wind data and the light data.

[0051] For example, a date within a week before the date of the current detection cycle is obtained as a historical date, and the first historical illumination data of the same time as the start time of the current detection cycle is obtained from multiple historical dates. For example, if the start time of the current detection cycle is 12 noon, the illumination data of 12 noon every day of the previous week is taken as the first historical illumination data, and the first historical illumination data is averaged to obtain the first reference illumination data at the start time of the current detection cycle. Then, the first relationship function is solved by the first reference illumination data.

[0052] According to an embodiment of the present invention, obtaining the first relationship function according to the first reference illumination data, the temperature data, the humidity data, the wind data and the illumination data includes: obtaining a first undetermined coefficient equation of the first relationship function according to formula (1),

[0053] (1),

[0054] in, is the illumination data at the i-th moment of the current detection cycle, is the first reference illumination data, is the temperature data at the i-th moment of the current detection cycle, is the wind force data at the i-th moment of the current detection cycle, is the humidity data at the i-th moment of the current detection cycle, , , , , , , , and The first undetermined coefficient of the first undetermined coefficient equation is solved according to the first benchmark lighting data, the temperature data, the humidity data, the wind data and the lighting data to obtain the solution value of the first undetermined coefficient; the first relationship function is obtained according to the solution value of the first undetermined coefficient and the first undetermined coefficient equation.

[0055] According to one embodiment of the present invention, is the first reference illumination data of the current detection cycle, is the variation of light intensity over time, It is the linear variation law of light intensity over time at multiple moments in the current detection cycle, thereby expressing the overall trend of light data changing over time. is a function to be fitted in the form of a sine function, representing a fitting term with a certain periodicity. For example, the current weather condition is cloudy and rainy, and half an hour later, the weather condition becomes sunny. In a change cycle, the temperature data, humidity data, and wind data may have a temporary impact on the light data. In general, temperature data and illumination data have a positive correlation, and humidity data and wind data have a negative correlation with illumination data. For example, when temperature data is high, the weather condition may be sunny. The higher the temperature data, the greater the illumination data. When humidity data and wind data are large, the weather condition may be rainy. The greater the humidity data and wind data, the smaller the illumination data. Therefore, the temperature data item is used as the numerator item, and the humidity data and wind data item are used as the denominator item. With the periodic changes of temperature data, humidity data and wind data in a short period of time, the illumination data in a short period of time may have a periodic impact. Therefore, the temperature data, humidity data and wind data at the i-th moment of the current detection cycle can be periodically fitted with the first undetermined coefficient, thereby expressing the part of the illumination data change in the current detection cycle that conforms to the short-term periodic change. According to the overall trend of illumination data changing over time and the part of illumination data change in the current detection cycle that conforms to the short-term periodic change, the undetermined coefficient equation is obtained, thereby expressing the relationship between the first reference illumination data, temperature data, humidity data, wind data and illumination data.

[0056] According to one embodiment of the present invention, fitting can be performed based on multiple parameters involved in the above undetermined coefficient equation, that is, fitting can be performed based on the first reference illumination data at the start of the current detection cycle, the temperature data, humidity data, wind data and illumination data at multiple moments of the current detection cycle, to solve the above multiple first undetermined coefficients. There are 9 first undetermined coefficients, namely, , , , , , , , and , according to the first benchmark illumination data and the temperature data, humidity data, wind data and illumination data at multiple times of the sampling location, the above 9 first undetermined coefficients are solved to obtain the solution values ​​of the above 9 first undetermined coefficients, and the solution values ​​of the above 9 first undetermined coefficients are substituted into the first relationship function to determine the first undetermined function.

[0057] In this way, based on the first benchmark lighting data, temperature data, humidity data, wind data and lighting data, a first relationship function is obtained, which accurately describes the relationship between the lighting data and the temperature data, graduation data and wind data. The periodicity and overall trend of the lighting data changes are taken into account in the calculation process, which improves the accuracy, scientificity and objectivity of the first relationship function.

[0058] According to one embodiment of the present invention, in step S104, the start time of the current detection cycle, the temperature data, the humidity data, the wind data and the light data are processed by a trained load prediction model to obtain predicted load data at multiple moments of the current detection cycle.

[0059] For example, when the start time of the current detection cycle is different, the relationship between the temperature data, humidity data, wind data and light data is different. For example, when the start time of the detection cycle is noon, the light data is strong, and the temperature data, humidity data, and wind data have a greater impact on the light data. In addition, the temperature data, humidity data, wind data and light data also have a greater impact on the power load. For example, when the light becomes dim, the lighting demand increases, which leads to an increase in the power load, or when the temperature drops, the heating demand increases, which leads to an increase in the power load. When the start time of the detection cycle is midnight, the light data is almost 0, and the temperature data, humidity data, and wind data have little impact on the light data, and the impact on the power load is also small. The load prediction model is a deep learning neural network model, which can be processed based on the start time of the current detection cycle, temperature data, humidity data, wind data and light data to obtain the predicted load data at multiple times of the current detection cycle.

[0060] According to one embodiment of the present invention, the training step of the load prediction model includes: processing the start time, temperature data, humidity data, wind data and light data of multiple detection cycles in the previous historical date of the current date through the load prediction model to obtain the predicted load data at multiple times in the multiple detection cycles in the previous historical date of the current date; obtaining the output power data at multiple times in the multiple detection cycles in the previous historical date of the current date; obtaining the total power consumption data in the detection cycles with the same start time in the multiple historical dates; determining the first weight of each detection cycle based on the total power consumption data in the detection cycles with the same start time; and determining the training loss function of the load prediction model according to formula (2): ,

[0061] (2),

[0062] in, is the temperature data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the temperature data at the first moment of the kth detection cycle in the previous historical date of the current date, is the wind force data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the wind force data at the first moment of the kth detection cycle in the previous historical date of the current date, is the humidity data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the humidity data at the first moment of the kth detection cycle in the previous historical date of the current date, is the illumination data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the illumination data at the first moment of the kth detection cycle in the previous historical date of the current date, is the output power data at the jth moment of the kth detection cycle in the previous historical date of the current date, The load forecasting model outputs the predicted load data at the jth moment of the kth detection cycle in the previous historical date of the current date. is the first weight of the kth detection cycle, is the number of detection cycles in each historical date, n is the number of moments in the previous detection cycle, j≤n, k≤ , and k, j, n and are all positive integers; according to the training loss function, the load forecasting model is trained to obtain the trained load forecasting model.

[0063] According to one embodiment of the present invention, according to one embodiment of the present invention, the error in the prediction at the first moment of any detection cycle is small, the prediction result is more accurate, and the size of the error is positively correlated with the moment in the detection cycle. It is the ratio of the temperature data at the jth moment in the kth detection cycle in the previous historical date to the temperature data at the 1st moment. The closer the ratio is to 1, the smaller the impact of temperature data changes on power demand, and thus, the smaller the impact of temperature data changes on the error of predicted load data. It is the ratio of the wind data at the jth moment in the kth detection cycle in the previous historical date to the wind data at the 1st moment. The closer the ratio is to 1, the smaller the impact of wind data changes on power demand, and thus, the smaller the impact of wind data changes on the error of predicted load data. It is the ratio of the humidity data at the jth moment in the kth detection cycle in the previous historical date to the humidity data at the 1st moment. The closer the ratio is to 1, the smaller the impact of humidity data changes on power demand, and thus, the smaller the impact of humidity data changes on the error of predicted load data. It is the ratio of the illumination data at the jth moment in the kth detection cycle in the previous historical date to the illumination data at the first moment. The closer the ratio is to 1, the smaller the impact of the change in illumination data on the power supply demand, and thus the smaller the impact of the change in illumination data on the error of the predicted load data. It means that temperature data, wind data and humidity data are positively correlated with power demand, and light data is negatively correlated with power demand. For example, when the temperature data is higher, the air conditioner needs to be turned on for cooling, and the power demand is greater. When the wind data and humidity data are larger, it may be raining outside and lighting is needed, and the power demand is greater. The larger the light data, the less lighting demand and the smaller the power load. Therefore, the items related to temperature data, wind data and humidity data are placed in the numerator, which means that the larger the temperature data, humidity data and wind data are relative to the data at the first moment of the kth detection cycle in the previous historical date, that is, , and The larger the value of , the greater the power load, and the greater the impact on the error of the predicted load data. The items related to the light data are placed in the denominator, which means that the smaller the light data is relative to the data at the first moment of the kth detection cycle in the previous historical date, the greater the power load, and the greater the impact on the error of the predicted load data. It is the output power data at the jth moment of the kth detection cycle in the historical date before the current date. The difference between the output power data and the predicted load data is the ratio of the output power data, that is, the relative error of the predicted load data. is the first weight of the kth detection cycle in the previous historical date of the current date. The total power consumption data of the same detection cycle in multiple historical dates can be averaged to obtain the average total power consumption data of each detection cycle, and the first weight of each detection cycle is determined based on the average total power consumption data of each detection cycle. For example, the first weight of the detection cycle with a larger average total power consumption data is higher.

[0064] According to one embodiment of the present invention, using The relative errors of the predicted load data at each moment of each detection cycle are weighted averaged with the first weight to obtain a training loss function. During the training process, the training loss function is reduced, thereby reducing the error between the output power data and the predicted load data, improving the accuracy of the load prediction model for the predicted load data, and thus improving the accuracy of the load prediction model.

[0065] In this way, the influence of temperature data, wind data, humidity data and light data on the error of predicted load data can be determined through the influence of the above data on electricity demand, and then the training loss function is set based on the influence, the relative error of the predicted load data and the first weight representing the total electricity consumption data of each detection cycle, so that the training loss function of the load prediction model is reduced during the training process, and the trained load prediction model is more suitable for the prediction of power load under the weather environment of the current date, so as to improve the accuracy of the load prediction model in a more targeted manner.

[0066] According to an embodiment of the present invention, the above training process can be performed once a day or every few days, so that the trained load forecasting model can adapt to the daily weather environment. However, the daily weather environment may change rapidly, and the load forecasting model trained once a day or every few days may be difficult to adapt to the weather environment of a certain detection period during the period of rapid weather changes, thereby failing to obtain accurate forecasting results. Therefore, in this case, the prediction model can also be corrected.

[0067] According to an embodiment of the present invention, in step S105, it can be first determined whether the trained load forecasting model can adapt to the weather environment of the current detection period, that is, it can be determined whether the trained load forecasting model needs to be corrected. According to the predicted load data at multiple moments of the current detection period and the output power data, it is determined whether the trained load forecasting model needs to be corrected.

[0068] For example, based on the predicted load data and output power data at multiple moments in the current detection cycle, it is determined whether the trained load prediction model is applicable to the prediction of the current time period, thereby determining whether the trained load prediction model needs to be corrected. For example, the error between the predicted load data in the current time period output by the load prediction model and the measured output power data can be determined. When the error is large, it can be determined that the trained load prediction model needs to be corrected.

[0069] According to one embodiment of the present invention, in step S106, if the trained load prediction model needs to be corrected, the trained load prediction model is corrected and trained according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data to obtain a corrected load prediction model.

[0070] According to one embodiment of the present invention, if a trained load prediction model needs to be corrected, the trained load prediction model is corrected and trained according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data to obtain a corrected load prediction model, including: obtaining total power consumption data in detection cycles with the same start time in multiple historical dates; determining a first weight for each detection cycle according to the total power consumption data in detection cycles with the same start time; determining a first weight for the current detection cycle according to the start time of the current detection cycle and the first weight of each detection cycle; determining a loss function of the load prediction model according to the first weight of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data; and correcting and training the trained load prediction model according to the loss function of the load prediction model to obtain a corrected load prediction model.

[0071] According to one embodiment of the present invention, the method for obtaining the first weight of each detection cycle is similar to that of the above-mentioned training, and will not be repeated here. Further, the first weight of the current detection cycle can be selected based on the start time of the current detection cycle. For example, the first weights of all detection cycles in a day can be obtained, and the first weight of the current detection cycle can be selected from the first weights of all detection cycles in a day based on the start time of the current cycle.

[0072] According to an embodiment of the present invention, the loss function of the load prediction model is determined according to the first weight of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data, including: determining the loss function of the load prediction model according to formula (3): ,

[0073] (3),

[0074] in, is the output power data at the i-th moment of the current detection cycle, is the forecast load data at the i-th moment of the current detection cycle, is the temperature data at the i-th moment of the current detection cycle, is the temperature data at the first moment of the current detection cycle, is the wind force data at the i-th moment of the current detection cycle, is the wind force data at the first moment of the current detection cycle, is the humidity data at the i-th moment of the current detection cycle, is the humidity data at the first moment of the current detection cycle, is the illumination data at the i-th moment of the current detection cycle, is the illumination data at the first moment of the current detection cycle, is the first weight of the current detection cycle, n is the number of moments in the previous detection cycle, i≤n, and both i and n are positive integers.

[0075] According to one embodiment of the present invention, the loss function determined by formula (3) is similar to the training loss function determined by formula (2), and the meaning of each parameter of formula (3) is not repeated here. However, the loss function determined by formula (3) only uses the first weight, output power data, predicted load data, temperature data, humidity data, wind data and light data of the current detection cycle for training, and does not involve data of multiple detection cycles. Therefore, there is no need to use the first weight to weight the relative errors of the predicted load data in multiple detection cycles.

[0076] According to one embodiment of the present invention, the load forecasting model can be trained using the above-mentioned loss function to minimize the training loss function, thereby improving the accuracy of the load forecasting model and adapting it to the weather environment of the current detection period. After the training is completed, the corrected load forecasting model is obtained.

[0077] In this way, the influence of temperature data, wind data, humidity data and light data on the error of predicted load data can be determined through the influence of the above data on electricity demand, so as to set the loss function based on the influence, the relative error of the predicted load data and the first weight, so that the loss function of the load prediction model can be reduced during the correction training process, and the corrected load prediction model can be more suitable for the prediction of power load at various moments in the current detection cycle, and the accuracy of the load prediction model can be improved in a more targeted manner, so that the corrected load prediction model can adapt to rapidly changing weather environments and maintain a high prediction accuracy in rapidly changing weather environments.

[0078] According to one embodiment of the present invention, in step S107, based on the first relationship function and the weather forecast data of the next detection cycle in the power supply area, the predicted light data at multiple moments in the next detection cycle is obtained, wherein the weather forecast data includes predicted temperature data, predicted humidity data and predicted wind data at multiple moments in the next detection cycle.

[0079] For example, predicted illumination data at multiple moments in the next detection cycle are obtained based on the first relationship function and predicted temperature data, predicted humidity data, and predicted wind data at multiple moments in the next detection cycle.

[0080] According to one embodiment of the present invention, based on the first relationship function and the weather forecast data of the next detection cycle in the power supply area, the predicted light data at multiple moments in the next detection cycle is obtained, including: obtaining second historical light data at the same moment as the end moment of the current detection cycle among multiple historical dates; determining second benchmark light data at the end moment of the current detection cycle based on the second historical light data; and obtaining the predicted light data at multiple moments in the next detection cycle based on the second benchmark light data, the predicted temperature data, the predicted humidity data, the predicted wind speed data and the first relationship function.

[0081] For example, a date within a week before the date of the current detection cycle is taken as the historical date, and the second historical lighting data at the same time as the end time of the current detection cycle, for example, the end time of the current detection cycle is 2 pm, and the lighting data at 2 pm every day of the previous week is taken as the second historical lighting data; an averaging operation is performed based on the second historical lighting data to determine the second benchmark lighting data at the end time of the current detection cycle, and then the predicted lighting data at multiple times of the detection cycle are determined based on the second benchmark lighting data and the first relationship function.

[0082] According to an embodiment of the present invention, the predicted illumination data at multiple moments of the next detection cycle are obtained according to the second reference illumination data, the predicted temperature data, the predicted humidity data, the predicted wind speed data and the first relationship function, including: obtaining the predicted illumination data at the i-th moment of the next detection cycle according to formula (4): ,

[0083] (4),

[0084] in, is the second reference illumination data, is the predicted humidity data at the i-th moment of the next detection cycle, is the predicted wind data at the i-th moment of the next detection cycle, is the predicted temperature data at the i-th moment of the next detection cycle, for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of .

[0085] According to one embodiment of the present invention, the end time of the current detection cycle is used as the start time of the next detection cycle. In this case, the first reference illumination data at the start time of the next detection cycle is the second reference illumination data at the end time of the current detection cycle. Therefore, the second reference illumination data, the solution value of the first undetermined coefficient, and the predicted wind data, predicted humidity data, and predicted temperature data can be substituted into the first relationship function to obtain formula (4), thereby obtaining the predicted illumination data at the i-th moment of the next detection cycle. Then, the predicted illumination data at each moment of the next detection cycle is determined.

[0086] In this way, the predicted light data at multiple moments in the next detection cycle can be obtained based on the second benchmark light data, predicted temperature data, predicted humidity data, predicted wind data and the first relationship function. The light data can be predicted based on the relationship between light and temperature, humidity, wind speed and other data, rather than just predicting light data based on time as in the related art. This can improve the accuracy, scientificity and objectivity of the predicted light data, and provide a data basis for subsequent load forecasting model predictions.

[0087] According to one embodiment of the present invention, in step S108, the end time of the current detection cycle, the weather forecast data of the next detection cycle and the predicted light data at multiple times of the next detection cycle are input into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle.

[0088] For example, the end time of the current detection cycle is the start time of the next detection cycle. The start time of the next detection cycle, the weather forecast data of the next detection cycle, and the predicted light data at multiple times of the next detection cycle are input into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle. The corrected load prediction model can adapt to the current weather conditions, so that the predicted load data with higher accuracy can be obtained based on the above data.

[0089] According to the novel online load prediction method of the power system according to the embodiment of the present invention, the relationship between temperature data, humidity data, wind data and light data can be accurately analyzed during short-term weather changes, and the accuracy of short-term power load prediction can be improved based on the relationship. When predicting load data, the load prediction model is corrected based on the influence of temperature data, humidity data, wind data and light data on power supply demand, so as to improve the adaptability of the load prediction model to the current weather environment and improve the objectivity and accuracy of the load prediction results. When determining the first relationship function, the first relationship function is obtained according to the first benchmark light data, temperature data, humidity data, wind data and light data, which accurately describes the relationship between light data and temperature data, division data and wind data. The periodicity and overall trend of light data changes are considered in the calculation process, which improves the accuracy, scientificity and objectivity of the first relationship function. When training the load prediction model, the influence of the temperature data, wind data, humidity data and light data on the power demand can be used to determine the influence of the above data on the error of the predicted load data, so as to set the training loss function based on the influence, the relative error of the predicted load data and the first weight representing the total power consumption data of each detection cycle, so that the load prediction model reduces the training loss function during the training process, and makes the trained load prediction model more suitable for the prediction of the power load under the weather environment of the current date, and improves the accuracy of the load prediction model more targeted. When correcting the load prediction model, the influence of the temperature data, wind data, humidity data and light data on the power demand can be used to determine the influence of the above data on the error of the predicted load data, so as to set the loss function based on the influence, the relative error of the predicted load data and the first weight, so that the load prediction model reduces the loss function during the correction training process, and makes the corrected load prediction model more suitable for the prediction of the power load at each moment of the current detection cycle, and improves the accuracy of the load prediction model more targeted, so that the corrected load prediction model can adapt to the rapidly changing weather environment and maintain a high prediction accuracy in the rapidly changing weather environment. When predicting the lighting data of the next detection cycle, the predicted lighting data at multiple moments of the next detection cycle can be obtained based on the second benchmark lighting data, predicted temperature data, predicted humidity data, predicted wind data and the first relationship function. The lighting data can be predicted based on the relationship between lighting and temperature, humidity, wind speed and other data, rather than predicting the lighting data only based on time as in the related art. This can improve the accuracy, scientificity and objectivity of the predicted lighting data and provide a data basis for subsequent load forecasting model predictions.

[0090] Figure 2 A block diagram of a novel power system load online prediction system according to an embodiment of the present invention is exemplarily shown, and the system includes:

[0091] The output power module is used to obtain the output power data of the transformer in the substation at multiple times in the current detection cycle;

[0092] A sampling data module, used to obtain temperature data, humidity data, wind data and light data of sampling locations in the power supply area corresponding to the transformer in the area at multiple moments in the current detection cycle;

[0093] A relationship function module, used to determine a first relationship function between the temperature data, humidity data, wind data and light data in the power supply area;

[0094] A prediction model module, used to process the start time of the current detection cycle, the temperature data, the humidity data, the wind data and the light data through a trained load prediction model to obtain predicted load data at multiple times of the current detection cycle;

[0095] A correction determination module is used to determine whether the trained load prediction model needs correction based on the predicted load data at multiple moments in the current detection cycle and the output power data;

[0096] A correction training module, for, if the trained load forecasting model needs correction, performing correction training on the trained load forecasting model according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data, to obtain a corrected load forecasting model;

[0097] a light prediction module, configured to obtain predicted light data at multiple moments of the next detection cycle according to the first relationship function and weather forecast data of the next detection cycle in the power supply area, wherein the weather forecast data includes predicted temperature data, predicted humidity data and predicted wind speed data at multiple moments of the next detection cycle;

[0098] The load prediction module is used to input the end time of the current detection cycle, the weather forecast data of the next detection cycle and the predicted light data at multiple times of the next detection cycle into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle.

[0099] According to one embodiment of the present invention, a new type of online load prediction device for a power system is provided, comprising: a processor; a memory for storing processor executable instructions; wherein the processor is configured to call the instructions stored in the memory to execute the new type of online load prediction method for a power system.

[0100] According to one embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the novel online load prediction method for a power system is implemented.

[0101] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0102] It should be understood by those skilled in the art that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A novel online load forecasting method for power system, characterized in that: include: At multiple moments in the current detection cycle, the output power data of the transformer in the substation area is obtained; At multiple moments in the current detection cycle, temperature data, humidity data, wind data, and light data of sampling locations within the power supply area corresponding to the substation transformer are obtained; Determine a first relationship function between temperature data, humidity data, wind data and light data in the power supply area; The start time of the current detection cycle, the temperature data, the humidity data, the wind data and the light data are processed by the trained load prediction model to obtain the predicted load data at multiple times of the current detection cycle; Determining whether the trained load prediction model needs to be corrected according to the predicted load data at multiple moments in the current detection cycle and the output power data; If the trained load forecasting model needs to be corrected, the trained load forecasting model is corrected and trained according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data to obtain a corrected load forecasting model; According to the first relationship function and the weather forecast data of the next detection cycle in the power supply area, the predicted light data at multiple moments of the next detection cycle are obtained, wherein the weather forecast data includes the predicted temperature data, the predicted humidity data and the predicted wind speed data at multiple moments of the next detection cycle; Input the end time of the current detection cycle, the weather forecast data of the next detection cycle, and the predicted light data at multiple times of the next detection cycle into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle; Determining a first relationship function among the temperature data, humidity data, wind speed data and light data in the power supply area according to the temperature data and light data at multiple moments includes: Acquire the first historical illumination data of the time which is the same as the start time of the current detection cycle among the multiple historical dates; Determine first reference light data at the start time of a current detection cycle according to the first historical light data; Obtaining the first relationship function according to the first reference illumination data, the temperature data, the humidity data, the wind data and the illumination data; Obtaining the first relationship function according to the first reference illumination data, the temperature data, the humidity data, the wind data and the illumination data includes: According to the formula ， The first undetermined coefficient equation of the first relationship function is obtained, where: is the illumination data at the i-th moment of the current detection cycle, is the first reference illumination data, is the temperature data at the i-th moment of the current detection cycle, is the wind force data at the i-th moment of the current detection cycle, is the humidity data at the i-th moment of the current detection cycle, , , , , , , , and is the first undetermined coefficient of the first undetermined coefficient equation; Solving the first undetermined coefficient according to the first reference illumination data, the temperature data, the humidity data, the wind data and the illumination data to obtain a solution value of the first undetermined coefficient; Obtaining the first relationship function according to the solved value of the first undetermined coefficient and the first undetermined coefficient equation; If the trained load forecasting model needs to be corrected, the trained load forecasting model is corrected and trained according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data to obtain the corrected load forecasting model, including: Obtain the total power consumption data within the detection period with the same start time on multiple historical dates; Determine a first weight for each detection cycle according to total power consumption data within the detection cycles with the same start time; Determine the first weight of the current detection cycle according to the start time of the current detection cycle and the first weight of each detection cycle; Determining a loss function of the load prediction model according to the first weight of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data, and the light data; Performing correction training on the trained load forecasting model according to the loss function of the load forecasting model to obtain a corrected load forecasting model; Determining a loss function of the load prediction model according to the first weight of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data, and the light data, including: According to the formula ， Determine the loss function of the load forecasting model ,in, is the output power data at the i-th moment of the current detection cycle, is the forecast load data at the i-th moment of the current detection cycle, is the temperature data at the i-th moment of the current detection cycle, is the temperature data at the first moment of the current detection cycle, is the wind force data at the i-th moment of the current detection cycle, is the wind force data at the first moment of the current detection cycle, is the humidity data at the i-th moment of the current detection cycle, is the humidity data at the first moment of the current detection cycle, is the illumination data at the i-th moment of the current detection cycle, is the illumination data at the first moment of the current detection cycle, is the first weight of the current detection cycle, n is the number of moments in the previous detection cycle, i≤n, and both i and n are positive integers.

2. The novel online load prediction method for power system according to claim 1 is characterized in that: The training steps of the load forecasting model include: The load forecasting model is used to process the start time of multiple detection cycles in the previous historical date, temperature data, humidity data, wind data and light data to obtain the forecast load data at multiple times in the multiple detection cycles in the previous historical date. Obtain output power data at multiple times in multiple detection cycles on a historical date before the current date; Obtain the total power consumption data within the detection period with the same start time on multiple historical dates; Determine a first weight for each detection cycle according to total power consumption data within the detection cycles with the same start time; According to the formula ， Determine the training loss function of the load forecasting model ,in, is the temperature data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the temperature data at the first moment of the kth detection cycle in the previous historical date of the current date, is the wind force data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the wind force data at the first moment of the kth detection cycle in the previous historical date of the current date, is the humidity data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the humidity data at the first moment of the kth detection cycle in the previous historical date of the current date, is the illumination data at the jth moment of the kth detection cycle in the previous historical date of the current date, is the illumination data at the first moment of the kth detection cycle in the previous historical date of the current date, is the output power data at the jth moment of the kth detection cycle in the previous historical date of the current date, The load forecasting model outputs the predicted load data at the jth moment of the kth detection cycle in the previous historical date of the current date. is the first weight of the kth detection cycle, is the number of detection cycles in each historical date, n is the number of moments in the previous detection cycle, j≤n, k≤ , and k, j, n and are all positive integers; The load forecasting model is trained according to the training loss function to obtain the trained load forecasting model.

3. The novel online load prediction method for power system according to claim 1 is characterized in that: Obtaining predicted light data at multiple moments in the next detection cycle according to the first relationship function and the weather forecast data of the next detection cycle in the power supply area, including: Obtaining second historical illumination data at the same time as the end time of the current detection cycle among multiple historical dates; Determine second reference illumination data at the end time of the current detection cycle according to the second historical illumination data; The predicted illumination data at multiple moments in the next detection cycle are obtained according to the second reference illumination data, the predicted temperature data, the predicted humidity data, the predicted wind force data and the first relationship function.

4. The novel online load prediction method for power system according to claim 3 is characterized in that: Obtaining predicted illumination data at multiple moments in the next detection cycle according to the second reference illumination data, the predicted temperature data, the predicted humidity data, the predicted wind speed data, and the first relationship function, comprising: According to the formula ， Get the predicted illumination data for the i-th moment of the next detection cycle , is the second reference illumination data, is the predicted humidity data at the i-th moment of the next detection cycle, is the predicted wind data at the i-th moment of the next detection cycle, is the predicted temperature data at the i-th moment of the next detection cycle, for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of for The solution value of .

5. A new type of online load forecasting system for power system, characterized in that: include: The output power module is used to obtain the output power data of the transformer in the substation at multiple times in the current detection cycle; A sampling data module, used to obtain temperature data, humidity data, wind data and light data of sampling locations in the power supply area corresponding to the transformer in the area at multiple moments in the current detection cycle; A relationship function module, used to determine a first relationship function between the temperature data, humidity data, wind data and light data in the power supply area; A prediction model module, used to process the start time of the current detection cycle, the temperature data, the humidity data, the wind data and the light data through a trained load prediction model to obtain predicted load data at multiple times of the current detection cycle; A correction determination module is used to determine whether the trained load prediction model needs correction based on the predicted load data at multiple moments in the current detection cycle and the output power data; A correction training module, for, if the trained load forecasting model needs correction, performing correction training on the trained load forecasting model according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data, to obtain a corrected load forecasting model; a light prediction module, configured to obtain predicted light data at multiple moments of the next detection cycle according to the first relationship function and weather forecast data of the next detection cycle in the power supply area, wherein the weather forecast data includes predicted temperature data, predicted humidity data and predicted wind speed data at multiple moments of the next detection cycle; A load prediction module is used to input the end time of the current detection cycle, the weather forecast data of the next detection cycle and the predicted light data at multiple times of the next detection cycle into the corrected load prediction model to obtain the predicted load data at multiple times of the next detection cycle; Determining a first relationship function among the temperature data, humidity data, wind speed data and light data in the power supply area according to the temperature data and light data at multiple moments includes: Acquire the first historical illumination data of the time which is the same as the start time of the current detection cycle among the multiple historical dates; Determine first reference light data at the start time of a current detection cycle according to the first historical light data; Obtaining the first relationship function according to the first reference illumination data, the temperature data, the humidity data, the wind data and the illumination data; Obtaining the first relationship function according to the first reference illumination data, the temperature data, the humidity data, the wind data and the illumination data includes: According to the formula ， The first undetermined coefficient equation of the first relationship function is obtained, where: is the illumination data at the i-th moment of the current detection cycle, is the first reference illumination data, is the temperature data at the i-th moment of the current detection cycle, is the wind force data at the i-th moment of the current detection cycle, is the humidity data at the i-th moment of the current detection cycle, , , , , , , , and is the first undetermined coefficient of the first undetermined coefficient equation; Solving the first undetermined coefficient according to the first reference illumination data, the temperature data, the humidity data, the wind data and the illumination data to obtain a solution value of the first undetermined coefficient; Obtaining the first relationship function according to the solved value of the first undetermined coefficient and the first undetermined coefficient equation; If the trained load forecasting model needs to be corrected, the trained load forecasting model is corrected and trained according to the start time of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data and the light data to obtain the corrected load forecasting model, including: Obtain the total power consumption data within the detection period with the same start time on multiple historical dates; Determine a first weight for each detection cycle according to total power consumption data within the detection cycles with the same start time; Determine the first weight of the current detection cycle according to the start time of the current detection cycle and the first weight of each detection cycle; Determining a loss function of the load prediction model according to the first weight of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data, and the light data; Performing correction training on the trained load forecasting model according to the loss function of the load forecasting model to obtain a corrected load forecasting model; Determining a loss function of the load prediction model according to the first weight of the current detection cycle, the output power data, the predicted load data, the temperature data, the humidity data, the wind data, and the light data, including: According to the formula ， Determine the loss function of the load forecasting model ,in, is the output power data at the i-th moment of the current detection cycle, is the forecast load data at the i-th moment of the current detection cycle, is the temperature data at the i-th moment of the current detection cycle, is the temperature data at the first moment of the current detection cycle, is the wind force data at the i-th moment of the current detection cycle, is the wind force data at the first moment of the current detection cycle, is the humidity data at the i-th moment of the current detection cycle, is the humidity data at the first moment of the current detection cycle, is the illumination data at the i-th moment of the current detection cycle, is the illumination data at the first moment of the current detection cycle, is the first weight of the current detection cycle, n is the number of moments in the previous detection cycle, i≤n, and both i and n are positive integers.

6. A new type of online load forecasting equipment for power system, characterized in that: include: processor; A memory for storing processor-executable instructions; wherein the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 4.

Images