A low-voltage photovoltaic user power generation load model training and prediction method
Through the low-voltage photovoltaic user power generation load model training method based on time series analysis forecasting method, the problem of low prediction accuracy of photovoltaic power generation load in the existing technology is solved, high-precision hour-level prediction is achieved, and the self-regulation ability of the power grid is improved and Run safely.
Patent Information
- Application Number
- CN202310334951.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-31
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2043-03-31
AI Technical Summary
In the prediction of photovoltaic power generation load, the time scale with a large daily or sampling interval is adopted, resulting in low prediction accuracy and inability to effectively deal with the problem of fluctuations in the output of photovoltaic power stations.
The low-voltage photovoltaic user power generation load model training method is adopted based on time series analysis prediction method. By determining the average value of the sample period, dividing the time series, obtaining initialization parameters, adjusting the model to match the characteristics of different time periods, performing load prediction, and improving prediction accuracy .
It realizes high-precision prediction of photovoltaic power generation load, reduces prediction deviations, and can predict power generation by hour, enhances the self-regulation ability of the power grid, and ensures the safe operation of the power grid.
Smart Images
Figure CN116341681B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic power generation control, and in particular to a low-voltage photovoltaic user power generation load model training and prediction method based on a time series analysis and prediction method. Background Art
[0002] Photovoltaic power generation is affected by the weather, so there is a problem of large fluctuations in the output of photovoltaic power stations. It is very necessary to predict the power generation load of photovoltaic power stations.
[0003] The invention patent application with application number 201811333001.2 discloses a control method for household photovoltaic, energy storage and energy use, which relates to the control field of photovoltaic, energy storage and energy use. The present invention includes the following steps: determining the daily power sequence of the electrical equipment with fixed load; calculating the daily power sequence of all fixed load equipment in the household; using the similar day algorithm to predict the daily power sequence of household photovoltaic power generation; calculating the daily power sequence of the net load of household energy use; adding the load power sequence of the household adjustable load, charging and storing energy during the net load power valley period; establishing an energy optimization model based on energy storage batteries. The present invention optimizes and coordinates energy use based on energy storage, takes into account the requirements of household energy efficiency and living comfort, reduces the time mismatch between photovoltaic power generation and user load, and takes the balance of the comprehensive benefits of users and power grids as the guide in energy optimization and coordination. By establishing an optimization and coordination model, the goal of reducing household energy costs and fluctuations in purchased electricity power is achieved.
[0004] However, existing forecasting technologies use daily or larger sampling intervals to model load series. The models are relatively simple but the prediction accuracy of power generation load is not high. Summary of the invention
[0005] The technical problem to be solved by the present invention is how to improve the accuracy.
[0006] The present invention solves the above technical problems through the following technical solutions:
[0007] The present invention provides a low-voltage photovoltaic user power generation load model training method based on a time series analysis and forecasting method, the method comprising:
[0008] Determine the average value of the sample period according to the total number of load samples and the time length of the load sample distribution; divide the target period into time series according to the average value of the sample period to obtain the interval distribution sequence within the target period;
[0009] Divide the interval distribution sequence into two continuous interval sequences, and use the continuous interval sequences as current sequences, obtain initialization parameters of the time series model for each current sequence, and use the initialization parameters to initialize the pre-established time series model;
[0010] Obtain the predicted value of the initialized time series model for the current sequence, and calculate the relative error between the predicted value and the actual value corresponding to the current sequence; when the relative error is less than the set error, use the initialized time series model as the target model for predicting the power generation load within the time range corresponding to the current sequence; when the relative error is greater than or equal to the set error, use the current sequence as an interval distribution sequence, and return to execute the step of dividing the interval distribution sequence into two continuous interval sequences until the target model is obtained;
[0011] For the period to be predicted, the target model is used to perform load forecasting.
[0012] Optionally, determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution includes:
[0013] The total number of load samples per unit time length is taken as the average value of the sample period.
[0014] Optionally, dividing the interval distribution sequence into two continuous interval sequences includes:
[0015] Divide the interval distribution sequence into two continuous interval sequences evenly.
[0016] Optionally, the obtaining of initialization parameters of the time series model includes:
[0017] Using the formula, Get the level value, trend value, and seasonal value of the time series model, where
[0018] L(1) is the level value; p is the length of the sample period; ∑ is the summation function; Y(i) is the sample value corresponding to the i-th sample period; T(1) is the trend value; S(1) is the seasonal value; X(1) is the intermediate variable required for the seasonal value calculation, and
[0019] The sample values contained in the current sequence are input into the preset function to obtain the corresponding cubic smoothing coefficient.
[0020] Optionally, before the step of determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution, the method further includes:
[0021] Obtain the season information and weather information corresponding to each load sample in the total number of load samples;
[0022] Clustering the load samples according to weather information and season information respectively, and obtaining the load sample sets with the same weather information and the same season information respectively;
[0023] Obtain future weather information and future season information corresponding to the period to be predicted, divide the period to be predicted according to the future weather information corresponding to each interval and the future season information corresponding to each interval, and obtain several future continuous intervals with the same future weather information and future season information;
[0024] For each future continuous interval, a target load sample set with the same seasonal information and weather information is selected from each load sample set, the target load sample set is used as the load sample, and the step of determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution is performed.
[0025] Optionally, the step of selecting a target load sample set having the same season information and weather information from each load sample set includes:
[0026] Using the formula, k = w1*T + w2*L, calculate the selection index of each continuous interval, where,
[0027] k is the selection index of the load sample set; w1 is the weight corresponding to the normalized value of the time interval between the last interval in the continuous interval and the first interval of the period to be predicted; T is the normalized value of the time interval between the last interval in the continuous interval and the first interval of the period to be predicted; w2 is the weight corresponding to the normalized value of the number of intervals contained in the continuous interval; L is the normalized value of the number of intervals contained in the continuous interval;
[0028] The continuous interval corresponding to the maximum value of the selection index is selected as the target load sample set.
[0029] Optionally, the load sample acquisition process in the step of determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution includes:
[0030] Determine whether the sample period is sunny, and if so, obtain the change curve of the direct sunlight angle within the sample period;
[0031] Taking the load sample points in the sample period as the reference points, the change curve is translated into the sample period to obtain the target curve; then the virtual load sample points are calculated based on the target curve, and then the set of the virtual load sample points and the load sample points in the sample period is used as the load sample of the sample period.
[0032] The present invention also provides a method for predicting low-voltage photovoltaic user power generation load based on a time series analysis forecasting method, the method comprising:
[0033] For the period to be predicted, a target model is used to perform load forecasting, wherein the target model is trained using any of the above methods.
[0034] Compared with the prior art, the present invention has the following advantages:
[0035] The present invention uses data features (power generation samples) and time series algorithms to retrain and re-predict the model based on the influencing factors of special events (such as weather changes), thereby reducing the deviation range and improving accuracy.
[0036] Moreover, the present invention can establish a power generation load prediction model with a day as the cycle according to the power supply unit and the hour as the load point, so as to realize the prediction of photovoltaic power generation load on an hourly basis.
[0037] Finally, the present invention further improves the observability and measurability of low-voltage distributed photovoltaics, enhances the self-regulation ability of the power grid, ensures the safe operation of the power grid, and improves the power grid's ability to absorb photovoltaic power. It can also reduce future uncertainties, reduce the risks that may be encountered in decision-making, and enable the decision-making goals to be successfully achieved, which has important practical significance. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 A schematic diagram of a process for predicting low-voltage photovoltaic user power generation load based on a time series analysis forecasting method provided by an embodiment of the present invention;
[0039] Figure 2 A schematic diagram of the principle of low-voltage photovoltaic user power generation load prediction based on the time series analysis forecasting method provided by an embodiment of the present invention;
[0040] Figure 3 A schematic diagram of a process for acquiring load samples provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0041] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiment.
[0042] Example 1
[0043] Figure 1 A schematic diagram of a process for predicting low-voltage photovoltaic user power generation load based on a time series analysis forecasting method provided by an embodiment of the present invention, Figure 2 The schematic diagram of the principle of low-voltage photovoltaic user power generation load prediction based on the time series analysis forecasting method provided by the embodiment of the present invention is as follows: Figure 1 and Figure 2 As shown, the method includes:
[0044] S101: determining the average value of a sample period according to the total number of load samples and the time length of the load sample distribution; dividing the target period into time series according to the average value of the sample period to obtain an interval distribution sequence within the target period;
[0045] For example, there are 120 groups of load samples for 5 days, that is, the load of a single photovoltaic power station is sampled 120 times within 5 days, which can be sampled once every hour or at irregular intervals: the number of load samples included in different dates of these 5 days can be the same or different, for example, there are 10 load samples on the first day, 30 load samples on the second day, 50 load samples on the third day, 10 load samples on the fourth day, and 10 load samples on the fifth day.
[0046] Due to the uneven distribution of load samples, in order to avoid the negative impact of such unevenness on model training, the embodiment of the present invention takes the unit time length of 1 day as an example, then the average value of the sample period is 120 / 5=24, wherein each group of load samples may include one or more load samples that are adjacent in time.
[0047] Then, taking the target period equal to the unit time length of one day as an example, the average value Y of the sample period is used as the basis for division, and the target period is evenly divided into 24 intervals, resulting in the following interval distribution sequence:
[0048] Interval 1, interval 2, interval 3, interval 4, interval 5, ..., interval 22, interval 23, interval 24.
[0049] S102: Divide the interval distribution sequence into two continuous interval sequences, and use the continuous interval sequences as current sequences, obtain initialization parameters of the time series model for each current sequence, and use the initialization parameters to initialize the pre-established time series model;
[0050] Specifically, the interval distribution sequence can be evenly divided into two continuous interval sequences, and the obtained two continuous intervals are continuous interval sequence 1, [1-12]; continuous interval sequence 2, [13-24].
[0051] Then, for each continuous interval sequence, obtain the load samples distributed in the continuous time interval sequence, take the continuous time interval sequence as the current sequence, establish the time series model respectively, and then, use the formula, Get the level value, trend value, and seasonal value of the time series model, where
[0052] L(1) is the level value; p is the length of the sample period; ∑ is the summation function; Y(i) is the sample value corresponding to the i-th sample period; T(1) is the trend value; S(1) is the seasonal value; X(1) is the intermediate variable required for the seasonal value calculation, and
[0053] By inputting the sample values contained in the current sequence into the fmin_l_bfgs_b function in scipy.optimize, several values can be obtained, and the extreme values among the several values are used as the corresponding cubic smoothing coefficients.
[0054] Initialize the corresponding time series model using level value, trend value, seasonal value and cubic smoothing coefficient.
[0055] S103: Obtain the predicted value of the initialized time series model for the current sequence, and calculate the relative error between the predicted value and the actual value corresponding to the current sequence; when the relative error is less than the set error, use the initialized time series model as the target model for predicting the power generation load within the time range of the current sequence; when the relative error is greater than or equal to the set error, use the current sequence as an interval distribution sequence, and return to execute the step of dividing the interval distribution sequence into two continuous interval sequences until the target model is obtained.
[0056] Taking the current sequence as the continuous interval sequence 1, [1-12] as an example, the load samples of [1-12] are input into the initialized time series model to obtain the predicted value at the moment of the 13th load sample, and then the predicted value is subtracted from the actual value of the 13th load sample to obtain the relative error. It can be understood that the load value of the next load sample adjacent to the continuous time interval sequence in the time dimension is the actual value referred to in this step; accordingly, the predicted value output by the initialized time series model is the predicted value at the moment of the next load sample adjacent to the continuous time interval sequence.
[0057] When the relative error is large, half of the total number of load samples contained in the continuous interval sequence is used as the dividing point to divide the continuous interval sequence 1, [1-12] into two continuous interval sequences, obtaining the continuous interval sequence 3, [1-6]; and the continuous interval sequence 4, [7-12].
[0058] Then, for continuous interval sequence 3 and continuous interval sequence 4, the initialization parameters of the time series model are obtained respectively, and the pre-established time series model is initialized using the initialization parameters, thereby realizing cyclic iteration and obtaining the corresponding target model. It can be understood that when the relative error of time series model A for continuous interval sequence 3 is small, time series model A should be used for prediction in the time range [1-6] corresponding to continuous interval sequence 3 in each cycle to be predicted; time series model B should be used for prediction in the time range [7-12] corresponding to continuous interval sequence 4 in each cycle to be predicted.
[0059] In the embodiment of the present invention, according to the law of exponential decay, the initial value has a greater impact on the time series sequence at the beginning. After several iterations of time steps, the impact of the initial value on the time series prediction becomes small, thereby improving the accuracy of the time series model prediction.
[0060] For example, the sample period is divided into:
[0061] Continuous interval sequence 3, [1-6], continuous interval sequence 4, [7-12], continuous interval sequence 2, [13-24], the target models corresponding to the three interval sequences are target model A, target model B, and target model C respectively.
[0062] Furthermore, during the training of the target model, there is an upper limit to the number of continuous interval sequences that each sample period can be divided into, and the upper limit value can be set manually.
[0063] S104: For the period to be predicted, use the target model to perform load prediction.
[0064] For example, the next day of the sample period is taken as the period to be predicted, and then identified according to the target model in the sample period. For example, the period to be predicted is divided into 24 intervals, and then the target model A is used to predict the 1st to 6th intervals of the 24 intervals; the target model A is used to predict the 7th to 12th intervals, and the target model C is used to predict the 13th to 24th intervals.
[0065] By applying the embodiment of the present invention, by dividing the same prediction period into different continuous interval sequences and then using targeted target models for different continuous interval sequences, the matching degree between the model and the load samples corresponding to the continuous interval sequence is improved, thereby improving the prediction accuracy.
[0066] While making predictions, it is also necessary to monitor the actual values of the next day, and then process the actual values and the corresponding moments into new load samples, and then return to execute step S101 to obtain a new target model, and then use the next day after the next day as the period to be predicted, and use the new target model for prediction, and so on.
[0067] Furthermore, during training, the length of the sample period may be consistent with the length of the period to be predicted, or may be a multiple of the length of the period to be predicted, where the multiple is greater than one.
[0068] By applying the above scheme of the present invention, the measured values corresponding to the period to be predicted are processed into new load samples, and then the new load samples are used to train the target model, thereby realizing iterative training of the target model and further improving the accuracy of the target model.
[0069] Example 2
[0070] Embodiment 1 of the present invention is only applicable to sample cycles with sufficient number of daily samples, for example, the method of embodiment 1 is applicable only when the number of load samples included in the sample cycle is not less than 24. Therefore, before using the method of embodiment 1 of the present invention, it is necessary to count the number of load samples included in the sample cycle, and the method of embodiment 1 of the present invention is applied only when the number of samples is greater than or equal to a set threshold, such as 24.
[0071] In order to solve the above problems, Example 2 of the present invention adds the following steps on the basis of Example 1:
[0072] S301 (not shown in the figure): determine whether the sample period is sunny, and if so, obtain a curve of the change of the direct sunlight angle within the sample period;
[0073] Specifically, Figure 3 A schematic diagram of a process for obtaining a load sample provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown in Figure (a), in the curve of the change in direct sunlight angle, the horizontal axis can be time, and the vertical axis can be the angle between the incident angle of sunlight and the plane of the photovoltaic panel illuminated by sunlight.
[0074] The corresponding characterization function is obtained by fitting the curve of the change of the direct sunlight angle. For example, the characterization function is y=a1t 2 +b1t+c1, where a1 is the coefficient of the quadratic term; t is time; b1 is the coefficient of the linear term; c is the constant term; y is the angle between the incident angle of sunlight and the plane of the photovoltaic panel. Generally speaking, this angle is positively correlated with the power generation.
[0075] S302: Taking the load sample points in the sample period as the reference points, the variation curve is translated into the sample period to obtain the target curve; then the virtual load sample points are calculated according to the target curve, and then the set of the virtual load sample points and the load sample points in the sample period is used as the total load sample amount of the sample period.
[0076] Specifically, Figure 3 As shown in Figure (b), the load sample points in the sample period are point A and point B, and then the target curve is translated to the coordinate system corresponding to the sample period. The horizontal axis of the coordinate system corresponding to the sample period is time, and the vertical axis is power generation. The translated target curve passes through point A and point B.
[0077] Furthermore, when the number of load sample points in the sample period is greater than 2 and less than 24, the target curve can be used as a fitting curve to minimize the sum of the vertical lengths from each load sample point to the target curve, thereby obtaining the target curve.
[0078] It should be emphasized that the start time and end time of the horizontal axis of the coordinate system corresponding to the sample period are the same as the start time and end time of the horizontal axis of the coordinate system corresponding to the period to be predicted. For example, if the start time of the horizontal axis of the coordinate system corresponding to the sample period is 6 o'clock, then the start time of the horizontal axis of the coordinate system of the period to be predicted is also 6 o'clock. If the end time of the horizontal axis of the coordinate system corresponding to the sample period is 18 o'clock, then the end time of the horizontal axis of the coordinate system of the period to be predicted is also 18 o'clock.
[0079] By applying the embodiment of the present invention, load samples are predicted according to the direct angle of the sun, which to a certain extent solves the technical problem in Example 1 that the sample cycle must meet 24 samples in 1 day for prediction, and expands the application scope of the solution of Example 1 of the present invention.
[0080] Example 3
[0081] In order to determine the impact of different weather and seasons on the fluctuation of the sampling values corresponding to the load samples, Embodiment 3 of the present invention is based on Embodiment 1 and further includes the following steps before Step S101.
[0082] S405 (not shown in the figure): obtain the season information and weather information corresponding to each load sample in the total number of load samples, wherein the season information includes: spring, summer, autumn, and winter; the weather includes: cloudy, light rain, heavy rain, sunny, and overcast.
[0083] S406 (not shown in the figure): The load samples are then clustered according to the weather information and the season information to obtain load sample sets with the same weather information and the same season information, for example, cloudy in spring, light rain in spring, heavy rain in spring, sunny in spring, and cloudy in spring; cloudy in summer, light rain in summer, heavy rain in summer, sunny in summer, and cloudy in summer; cloudy in autumn, light rain in autumn, heavy rain in autumn, sunny in autumn, and cloudy in autumn; cloudy in winter, light rain in winter, heavy rain in winter, sunny in winter, and cloudy in winter, etc. 20 load sample sets.
[0084] S407 (not shown in the figure): obtaining future weather information and future season information corresponding to the period to be predicted, dividing the period to be predicted according to the future weather information corresponding to each interval and the future season information corresponding to each interval, to obtain a number of future continuous intervals having the same future weather information and future season information;
[0085] For example, the prediction period is divided into 200 intervals, namely, interval 1, interval 2, ..., interval 198, interval 199, interval 100, etc.; the divided future continuous intervals are: interval 1-interval 30 is sunny in spring; interval 50-interval 80 is cloudy in spring; interval 160-interval 190 is sunny in summer;
[0086] S408 (not shown in the figure): for each future continuous interval, a target load sample set with the same season information and weather information is selected from each load sample set.
[0087] For example, if interval 1 to interval 30 is sunny in spring, the load sample set with season information of spring and weather information of sunny is taken as the target load sample set. If interval 50 to interval 80 is cloudy in spring, the load sample set with season information of spring and weather information of cloudy is taken as the target load sample set. If interval 160 to interval 190 is sunny in summer, the load sample set with season information of summer and weather information of sunny is taken as the target load sample set.
[0088] Finally, the target load sample set is used as the input of step S101 to execute embodiment 1 of the present invention.
[0089] By applying the embodiment of the present invention and using a target load sample set with the same season information and weather information to train the model, the negative impact of the season information and weather information on the model training can be further eliminated, thereby further improving the accuracy of the model training.
[0090] Furthermore, the same target load sample set may contain two or more continuous intervals with the same weather information and season information and continuous in time. Therefore, how to select a more appropriate maximum continuous interval from the target load sample set is a technical problem that must be solved.
[0091] In order to solve the above problem, when there is more than one continuous interval in the same target load sample set, the following formula, k=w1*T+w2*L, is used to calculate the selection index of each continuous interval, wherein k is the selection index of the load sample set; w1 is the weight corresponding to the normalized value of the time interval between the last interval in the continuous interval and the first interval of the period to be predicted; T is the normalized value of the time interval between the last interval in the continuous interval and the first interval of the period to be predicted; w2 is the weight corresponding to the normalized value of the number of intervals contained in the continuous interval; L is the normalized value of the number of intervals contained in the continuous interval; in this step, the interval can be a preset period or a preset time step.
[0092] The continuous interval corresponding to the maximum value of the selection index is taken as the target load sample set, and then the target load sample set is taken as the load sample in step S101.
[0093] It should be emphasized that when normalizing the time intervals, the maximum value used is the maximum value of the time interval between the last interval in each continuous interval in the target load sample set and the first interval of the period to be predicted. Similarly, the minimum value used is the minimum value of the time interval between the last interval in each continuous interval in the target load sample set and the first interval of the period to be predicted.
[0094] By applying the above-mentioned embodiment of the present invention, the load samples corresponding to the appropriate continuous interval can be selected as the input of step S101 according to the weight preference set by the user, thereby further improving the training accuracy of the model while reducing the influence of seasonal weather.
[0095] Example 4
[0096] The present invention also provides a method for predicting low-voltage photovoltaic user power generation load based on a time series analysis forecasting method, the method comprising:
[0097] For the period to be predicted, a target model is used to perform load forecasting, wherein the target model is trained using the method shown in any one of Examples 1-3.
[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A low-voltage photovoltaic user power generation load model training method, characterized in that: The method comprises: Determine the average value of the sample period according to the total number of load samples and the time length of the load sample distribution; divide the target period into time series according to the average value of the sample period to obtain the interval distribution sequence within the target period; Divide the interval distribution sequence into two continuous interval sequences, and use the continuous interval sequences as current sequences, obtain initialization parameters of the time series model for each current sequence, and use the initialization parameters to initialize the pre-established time series model; Obtain the predicted value of the initialized time series model for the current sequence, and calculate the relative error between the predicted value and the actual value corresponding to the current sequence; when the relative error is less than the set error, use the initialized time series model as the target model for predicting the power generation load within the time range corresponding to the current sequence; when the relative error is greater than or equal to the set error, use the current sequence as an interval distribution sequence, and return to execute the step of dividing the interval distribution sequence into two continuous interval sequences until the target model is obtained; Before the step of determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution, the method further includes: Obtain the season information and weather information corresponding to each load sample in the total number of load samples; Clustering the load samples according to weather information and season information respectively, and obtaining the load sample sets with the same weather information and the same season information respectively; Obtain future weather information and future season information corresponding to the period to be predicted, divide the period to be predicted according to the future weather information corresponding to each interval and the future season information corresponding to each interval, and obtain several future continuous intervals with the same future weather information and future season information; For each future continuous interval, a target load sample set having the same season information and weather information is selected from each load sample set, the target load sample set is used as a load sample, and the step of determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution is performed; The step of selecting a target load sample set having the same season information and weather information from each load sample set includes: Using the formula, k=w1*T+w2*L, calculate the selection index of each continuous interval, where, k is the selection index of the load sample set; w1 is the weight corresponding to the normalized value of the time interval between the last interval in the continuous interval and the first interval of the period to be predicted; T is the normalized value of the time interval between the last interval in the continuous interval and the first interval of the period to be predicted; w2 is the weight corresponding to the normalized value of the number of intervals contained in the continuous interval; L is the normalized value of the number of intervals contained in the continuous interval; The continuous interval corresponding to the maximum value of the selection index is selected as the target load sample set.
2. A low-voltage photovoltaic user power generation load model training method according to claim 1, characterized in that: Determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution includes: The total number of load samples within a unit time length is taken as the average value of the sample period.
3. A low-voltage photovoltaic user power generation load model training method according to claim 1, characterized in that: The step of dividing the interval distribution sequence into two continuous interval sequences includes: Divide the interval distribution sequence into two continuous interval sequences evenly.
4. A low-voltage photovoltaic user power generation load model training method according to claim 1, characterized in that: The obtaining of the initialization parameters of the time series model includes: Using the formula, , get the level value, trend value and seasonal value of the time series model, where, L (1) is the horizontal value; p is the length of the sample period; Σ is the summation function; Y ( i ) is the sample value corresponding to the i-th sample period; T (l) is the trend value; S (1) is the seasonal value; X (1) is the intermediate variable required for calculating seasonal values, and ; The sample values contained in the current sequence are input into the preset function to obtain the corresponding cubic smoothing coefficient.
5. A low-voltage photovoltaic user power generation load model training method according to claim 1, characterized in that: The load sample acquisition process in the step of determining the average value of the sample period according to the total number of load samples and the time length of the load sample distribution includes: Determine whether the sample period is sunny, and if so, obtain the change curve of the direct sunlight angle within the sample period; Taking the load sample points in the sample period as the reference points, the change curve is translated into the sample period to obtain the target curve; then the virtual load sample points are calculated based on the target curve, and then the set of the virtual load sample points and the load sample points in the sample period is used as the load sample of the sample period.
6. A method for predicting power generation load of low-voltage photovoltaic users, characterized in that: The method comprises: For a period to be predicted, a target model is used to perform load forecasting, wherein the target model is trained using a method as described in any one of claims 1 to 5.