Photovoltaic energy storage control method based on intelligent learning

Abstract

The present invention relates to the technical field of data processing, and more specifically, to a photovoltaic energy storage control method based on intelligent learning. The method includes: predicting each photovoltaic energy storage data sequence through the exponential smoothing method, obtaining historical high error patterns based on the prediction errors, calculating the adjustment coefficients of each historical high error pattern, screening similar patterns based on the similarity between the real-time high error pattern and the historical high error pattern in the same energy storage data sequence, using the similarity corresponding to each historical high error pattern in the similar patterns as weights, performing weighted averaging on the adjustment coefficients of the corresponding historical high error patterns, adjusting the predicted value of the latest high error data of the corresponding item based on the weighted adjustment coefficients, and performing photovoltaic energy storage control based on the adjusted predicted value. The present invention can improve the accuracy of the predicted value of each photovoltaic energy storage data obtained, thereby achieving precise control of photovoltaic energy storage.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing. More specifically, the present invention relates to a photovoltaic energy storage control method based on intelligent learning. Background Art

[0002] With the accelerated transformation of the global energy structure towards low-carbonization, photovoltaic power generation, with its clean and renewable characteristics, has become one of the core pillars of the renewable energy system. However, the output characteristics of photovoltaic power generation have significant intermittency and volatility. For example, affected by natural factors such as day-night alternation, cloud occlusion, and seasonal changes, its output power will show non-stationary random fluctuations.

[0003] There is a contradiction between this volatility and the rigid demand of the power grid for stable power output, which may lead to problems such as power grid frequency deviation and voltage instability. Therefore, as an "energy buffer pool", the energy storage system needs to achieve flexible storage and precise scheduling of electric energy through intelligent control strategies to suppress photovoltaic power fluctuations and improve the power grid's accommodation capacity. In this context, how to optimize the response efficiency of the energy storage system through accurate data prediction and adaptive control technology has become a key challenge for the industry's technological upgrade.

[0004] Currently, prediction methods based on time series analysis, such as the exponential smoothing method, are usually used to predict future photovoltaic output and energy storage demand. However, in the process of predicting photovoltaic energy storage data by the exponential smoothing method, due to the coupling influence of multiple factors such as light intensity, temperature, and meteorological mutations on photovoltaic energy storage data, there are natural photovoltaic change patterns in the data that are difficult to identify. If the exponential smoothing method is directly used to predict photovoltaic energy storage data, these data change trends cannot be concerned, resulting in inaccurate results in the process of predicting photovoltaic energy storage data when the data fluctuates, thus affecting the subsequent control effect. Summary of the Invention

[0005] To solve the problem that the conventional exponential smoothing method cannot refer to the historical data change trend in the prediction of photovoltaic energy storage data, resulting in low prediction accuracy and affecting the precise control of photovoltaic energy storage, the present invention provides a photovoltaic energy storage control method based on intelligent learning. The method includes:

[0006] Obtain multiple photovoltaic energy storage data collected at each sampling point to obtain multiple energy storage data sequences;

[0007] Use the exponential smoothing method to predict each data in each energy storage data sequence. Based on the prediction error, screen the historical high-error patterns in each energy storage data sequence, and with the goal of reducing the prediction error, calculate the adjustment coefficient of each historical high-error pattern based on the number of data in each historical high-error pattern and the ratio of the actual value to the predicted value of the data in the corresponding historical high-error pattern;

[0008] Obtain the real-time high-error mode of each energy storage data sequence, and screen the similar modes of the real-time high-error mode based on the similarity between the real-time high-error mode and the historical high-error mode of the same item.

[0009] Take the similarity between each historical high-error mode and the real-time high-error mode in the similar modes as the weight, and perform a weighted average on the adjustment coefficients of all historical high-error modes in the similar modes, so as to adjust the predicted value of the latest high-error data of the corresponding item based on the weighted adjustment coefficient, and perform photovoltaic energy storage control based on the adjusted predicted value.

[0010] By screening the similar modes of the real-time high-error mode, the present invention can find multiple historical high-error modes with high similarity to the real-time high-error mode. Then, using the similarity between each historical high-error mode in the similar modes as the weight, a weighted average is performed on the adjustment coefficients of all historical high-error modes in the similar modes, which can increase the weight of the adjustment coefficient of the historical high-error mode with a large similarity to the real-time high-error mode, so that the determined weighted adjustment coefficient can comprehensively consider the adjustment coefficients of multiple historical high-error modes. Therefore, based on the weighted adjustment coefficient, the latest high-error data can be corrected more accurately, providing a more accurate data basis for the determination of the photovoltaic energy storage control strategy.

[0011] Preferably, the method for obtaining the similarity between the real-time high-error mode and the historical high-error mode of the same item includes:

[0012] For the real-time high-error mode and any historical high-error mode in the same energy storage data sequence, the ratio of the difference between adjacent data values in the real-time high-error mode to the difference between adjacent data values in any historical high-error mode is used as the first index; the difference between the ratio of the average value of the data values and the average value of the data predicted values in the real-time high-error mode and the ratio of the average value of the data values and the average value of the data predicted values in any historical high-error mode is used as the second index.

[0013] Calculate the similarity between the real-time high-error mode and the historical high-error mode in the same energy storage data sequence. The similarity is negatively correlated with the difference between 1 and the first index, and the second index.

[0014] The present invention can measure the similarity of the data change gradient between the corresponding two modes through the first index, and measure the similarity of the mutual relationship between the data values and the predicted values through the second index. Measuring the similarity of the corresponding two modes from different dimensions can ensure the accuracy of the determined similarity.

[0015] Preferably, the similarity satisfies the following relational expression:

[0016] ;

[0017] In the formula, is the similarity between the real-time high error mode and the th historical high error mode in the th energy storage data sequence; , are respectively the values of the th data and the th data in the real-time high error mode; , are respectively the values of the th data and the th data in the historical high error mode; , are respectively the average value and the average predicted value of all data in the real-time high error mode; , are respectively the average value and the average predicted value of all data in the historical high error mode; is the absolute value symbol; is the natural exponential function; is the number of data in the real-time high error mode.

[0018] Preferably, after determining the similarity between the real-time high error mode in the same energy storage data sequence and any historical high error mode, it further includes:

[0019] Taking the similarity between the real-time high error mode in any energy storage data sequence and any historical high error mode in the same energy storage data sequence as the initial similarity, and respectively extracting the data segment corresponding to the real-time high error mode in the same energy storage data sequence and the data segment corresponding to any historical high error mode from the remaining energy storage data sequences to form two data segment sets;

[0020] Calculating the average similarity of all data segments belonging to the same energy storage data sequence in the two data segment sets, and using the average similarity to correct the initial similarity, and the correction value is the average of the average similarity and the initial similarity.

[0021] The present invention corrects the similarity determined based on the first index and the second index, can provide another dimension of measurement standard, reduces the interference of noise, and thus can further ensure the accuracy of the determined similarity.

[0022] Preferably, screening the similar modes of the real-time high error mode based on the similarity between the real-time high error mode and the historical high error mode of the same item includes:

[0023] Arrange the historical high-error patterns of the same item according to the magnitude of the correction value with respect to the similarity to the real-time high-error mode, and use the set composed of several historical high-error patterns selected from the front to the back as the similar pattern of the real-time high-error mode.

[0024] The screening of similar patterns in the present invention can provide reference data for the subsequent determination of the weighted adjustment coefficient.

[0025] Preferably, the adjustment coefficients of each historical high-error pattern satisfy the following relational expression:

[0026] ;

[0027] In the formula, is the adjustment coefficient of the th historical high-error pattern in the th energy storage data sequence item; is the value of the rd data in the th historical high-error pattern in the th energy storage data sequence item; is the predicted value of the th data in the th historical high-error pattern in the th energy storage data sequence item; is the number of data in the th historical high-error pattern in the th energy storage data sequence item; is the product symbol.

[0028] Preferably, the method for obtaining the historical high-error pattern includes:

[0029] Take the data with a prediction error greater than the preset error threshold in each energy storage data sequence as high-error data, and take the continuous high-error data segment with the number of consecutive occurrences of high-error data greater than or equal to the preset value as a historical high-error pattern, so as to obtain all the historical high-error patterns in each energy storage data sequence.

[0030] Preferably, the error threshold is an adaptive error threshold, and the adaptive error threshold satisfies the following relational expression:

[0031] ;

[0032] In the formula, is the adaptive error threshold of the rd data in the th energy storage data sequence item; is the maximum value of the prediction errors of all data in the th energy storage data sequence item; is the The prediction error of the nth data in the energy storage data sequence; is the standard deviation of the prediction errors of all data in the nth energy storage data sequence; is the average prediction error value of all data in the nth energy storage data sequence; is the hyperbolic tangent function.

[0033] The present invention screens high-error data by means of an adaptive error threshold, which can reduce misjudgment situations caused by normal data fluctuations, thereby ensuring the accuracy of the determined high-error data.

[0034] Preferably, based on the weighted adjustment coefficient, the prediction value of the latest high-error data of the corresponding item is adjusted, including:

[0035] Performing a multiplication operation on the weighted adjustment coefficient and the prediction value of the latest high-error data of the corresponding item to complete the adjustment of the prediction value of the latest high-error data.

[0036] Preferably, the multiple photovoltaic energy storage data includes real-time irradiance, temperature, and output power.

[0037] The present invention has the following effects:

[0038] When predicting photovoltaic energy storage data by using the exponential smoothing method, the present invention determines the weighted adjustment coefficient of the latest high-error data by referring to multiple historical high-error patterns with a high similarity to the real-time high-error pattern, and the adjustment coefficient of the historical high-error pattern with a high similarity to the real-time high-error pattern has a greater weight, thereby ensuring the accuracy of the weighted adjustment coefficient. Furthermore, based on this weighted adjustment coefficient, the prediction value of the latest high-error data can be adjusted in real time, reducing the prediction error and improving the accuracy of the prediction value, and thus realizing the precise control of photovoltaic energy storage. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a schematic flow chart of the steps of the photovoltaic energy storage control method based on intelligent learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will describe in detail the specific embodiments of the present invention with reference to the drawings.

[0041] Referring to Figure 1 , the photovoltaic energy storage control method based on intelligent learning includes steps S1 - S4, specifically as follows:

[0042] S1: Obtain multiple photovoltaic energy storage data collected at each sampling point to obtain multiple energy storage data sequences.

[0043] In an exemplary embodiment of the present invention, multiple photovoltaic energy storage data includes real-time irradiance, temperature, and output power.

[0044] Specifically, within a sampling period, such as the most recent month at the current moment, at a fixed sampling frequency, such as once per hour, the real-time irradiance, temperature, and output power of the photovoltaic energy storage device can be collected simultaneously to obtain multiple energy storage data sequences within this month. Currently, appropriate sampling periods and sampling frequencies can also be selected according to specific situations, and this embodiment does not make special limitations on this.

[0045] S2: Use the exponential smoothing method to predict each data in each energy storage data sequence. Based on the prediction error, screen the historical high-error patterns in each energy storage data sequence, and with the goal of reducing the prediction error, calculate the adjustment coefficient of each historical high-error pattern based on the number of data in each historical high-error pattern and the ratio of the actual value to the predicted value of the data in the corresponding historical high-error pattern.

[0046] It should be noted that during the process of predicting photovoltaic energy storage data using the exponential smoothing method, when the data shows a trend change, the prediction effect will deteriorate. Therefore, the present invention records the data with higher errors for subsequent analysis and use.

[0047] It should be further noted that since there may be abnormal data in the photovoltaic energy storage data, the probability of its prediction error being greater than the error threshold is also very high. Therefore, in order to reduce the interference of abnormal data, the present invention analyzes in units of high-error patterns, so that individual high-error data can be eliminated.

[0048] In an exemplary embodiment of the present invention, the determination of the historical high-error patterns in each energy storage data sequence can be achieved through the following steps:

[0049] Data with a prediction error greater than a preset error threshold in each energy storage data sequence is used as high-error data, and a continuous high-error data segment with the number of consecutive high-error data occurrences greater than or equal to a preset value is used as a historical high-error pattern to obtain all historical high-error patterns in each energy storage data sequence.

[0050] Among them, the prediction error is the difference between the actual value and the predicted value of the photovoltaic energy storage data.

[0051] Exemplarily, when the number of occurrences of high-error data in any energy storage data sequence is greater than or equal to a preset value, such as 3 times, then the data segment with consecutive high-error data is used as a historical high-error pattern in this energy storage data sequence, so that all historical high-error patterns in each energy storage data sequence can be screened.

[0052] In an exemplary embodiment of the present invention, the error threshold is an adaptive error threshold. Specifically, the adaptive error threshold satisfies the following relational expression:

[0053] ;

[0054] In the formula, is the adaptive error threshold of the th data in the th energy storage data sequence; is the maximum value of the prediction errors of all data in the th energy storage data sequence; is the prediction error of the th data in the th energy storage data sequence; is the standard deviation of the prediction errors of all data in the th energy storage data sequence; is the average prediction error value of all data in the th energy storage data sequence; is the hyperbolic tangent function.

[0055] Among them, reflects the normalized value. The larger this value is, the greater the prediction error of the th data in the th energy storage data sequence. By setting a larger adaptive error threshold, it is possible to avoid misjudging normal data as high-error data due to normal fluctuations of the data. The hyperbolic tangent function can limit the range of this normalized value to (-1, 1), so that the case where the prediction error is negative can be considered.

[0056] Optionally, by combining the value and the value, the adaptive error threshold can adapt to the overall error level of the data in each energy storage data sequence while adapting to the fluctuation range of the overall prediction error, thereby improving the detection sensitivity and reducing false alarm situations.

[0057] In another embodiment, an appropriate error threshold can also be selected according to empirical values.

[0058] Furthermore, after determining the historical high-error patterns in each energy storage data sequence based on the adaptive error threshold, the adjustment coefficients of each historical high-error pattern can be calculated to reduce the prediction errors of the data in each historical high-error pattern.

[0059] Specifically, the adjustment coefficients of each historical high-error pattern satisfy the following relational expression:

[0060] ;

[0061] In the formula, is the adjustment coefficient of the th historical high error mode in the th energy storage data sequence; is the value of the th data in the th historical high error mode in the th energy storage data sequence; is the predicted value of the th data in the th historical high error mode in the th energy storage data sequence; is the number of data in the th historical high error mode in the th energy storage data sequence; is the product symbol.

[0062] Among them, reflects the difference between the data and the predicted value in the th historical high error mode in the th energy storage data sequence. When this value is greater than 1, it means that the predicted value of the data in this historical high error mode is too small. At this time, the predicted value of the data in this historical high error mode needs to be adjusted upward to reduce the prediction error.

[0063] It should be noted that by using the number of data in the historical high error mode to perform geometric mean (i.e., taking the th root) of the obtained continued product, the influence of the number of data in each historical high error mode can be eliminated, and the sensitivity to extreme values can be reduced.

[0064] Furthermore, the adjustment coefficients of all historical high error modes in each energy storage data sequence can be determined through the calculation method of the adjustment coefficient.

[0065] S3: Obtain the real-time high error mode of each energy storage data sequence, and screen the similar modes of the real-time high error mode based on the similarity between the real-time high error mode and the historical high error mode of the same item.

[0066] It should be noted that during the prediction of the real-time data of each photovoltaic energy storage data, when there is a large deviation between the predicted value and the actual value of the real-time data, it is necessary to timely determine an appropriate adjustment coefficient to accurately adjust the predicted value of the real-time data.

[0067] Optionally, for any photovoltaic energy storage data, if the three recently collected data are all high error data (i.e., the difference between the actual value and the predicted value is greater than the adaptive error threshold), then the data segment composed of these three data is defined as the real-time high error mode of this photovoltaic energy storage data.

[0068] In an exemplary embodiment of the present invention, the determination of the similarity between the real-time high-error mode and the historical high-error mode in the same energy storage data sequence can be achieved through the following steps:

[0069] For the real-time high-error mode and any historical high-error mode in the same energy storage data sequence, the ratio of the difference between adjacent data values in the real-time high-error mode to the difference between adjacent data values in any historical high-error mode is used as the first index; the difference between the ratio of the average value of the data values and the average value of the data prediction values in the real-time high-error mode and the ratio of the average value of the data values and the average value of the data prediction values in any historical high-error mode is used as the second index; calculate the similarity between the real-time high-error mode and the historical high-error mode in the same energy storage data sequence, and the similarity is negatively correlated with the difference between 1 and the first index, as well as the second index.

[0070] It should be noted that when calculating the similarity between the real-time high-error mode and the historical high-error mode in the same energy storage data sequence, in order to keep the number of data consistent, only three data in the historical high-error mode are retained. Among them, the three retained data in the historical high-error mode are the first three data.

[0071] Specifically, the similarity between the real-time high-error mode and the historical high-error mode in the same energy storage data sequence satisfies the following relational expression:

[0072] ;

[0073] In the formula, is the similarity between the real-time high-error mode and the th historical high-error mode in the rd energy storage data sequence; , are the values of the th data and the th data in the real-time high-error mode respectively; , are the values of the th data and the th data in the historical high-error mode respectively; , are the average value and the average predicted value of all data in the real-time high-error mode respectively; , are the average value and the average predicted value of all data in the historical high-error mode respectively; is the absolute value symbol; is the natural exponential function; is the number of data in the real-time high-error mode, and in this embodiment .

[0074] Among them, is the first index between the real-time high error mode and the historical high error mode, reflecting the similarity of the data change gradients of the two high error modes. The closer the value is to 1, the closer the data change gradients in the two high error modes are, and further, the more similar the data corresponding to the two high error modes are, and the greater the corresponding similarity is.

[0075] Optionally, when calculating the first index, the difference in the values of adjacent data can be the difference in values or the absolute value of the difference in values.

[0076] is the second index between the real-time high error mode and the historical high error mode, reflecting the similarity of the mutual relationship between the values and predicted values of each data in the two high error modes. The smaller the value, the smaller the difference in the mutual relationship between the values and predicted values of each data in the two high error modes, and further, the more similar the mutual relationship between the values and predicted values of each data in the two high error modes is, and the greater the corresponding similarity is.

[0077] Optionally, when calculating the second index, the difference between the value of and the value of

[0078] In another embodiment, it is also possible to use the relational expression: ; to calculate the similarity between the real-time high error mode and any historical high error mode in the same energy storage data sequence.

[0079] In an exemplary embodiment of the present invention, after determining the similarity between the real-time high error mode and any historical high error mode in the same energy storage data sequence, the following steps are further included:

[0080] Step 1: Take the similarity between the real-time high error mode in any energy storage data sequence and any historical high error mode in the same energy storage data sequence as the initial similarity, and respectively extract the data segments corresponding to the real-time high error mode in the same energy storage data sequence and the data segments corresponding to any historical high error mode from the remaining energy storage data sequences to form two data segment sets;

[0081] Exemplarily, for the real-time high-error mode and the historical high-error mode in any energy storage data sequence, first determine the positions of the real-time high-error mode and the historical high-error mode, and then find two data segments corresponding to the position of the real-time high-error mode from the remaining two energy storage data sequences (the total number of energy storage data sequences obtained in the present invention is 3) to obtain a data segment set, and find two data segments corresponding to the position of the historical high-error mode from the remaining two energy storage data sequences to obtain another data segment set, thus forming two data segment sets.

[0082] Step 2: Calculate the average similarity of the data segments belonging to the same energy storage data sequence in the two data segment sets, and use the average similarity to correct the initial similarity, where the correction value is the average of the average similarity and the initial similarity.

[0083] It should be noted that in photovoltaic energy storage data, there is a relationship of mutual influence between different items of data. Therefore, when one item of data changes, other items of data will also have similar changes, and the similarity can be adjusted accordingly, so as to accurately measure the similarity between the real-time high-error mode and each historical high-error mode in the same energy storage data sequence.

[0084] Specifically, the correction value of the similarity between the real-time high-error mode and any historical high-error mode in the same energy storage data sequence satisfies the following relational expression:

[0085] ;

[0086] In the formula, is the correction value of the similarity between the real-time high-error mode and the th historical high-error mode in the th energy storage data sequence; is the similarity between the real-time high-error mode and the th historical high-error mode in the th energy storage data sequence; is the similarity between the data segment corresponding to the real-time high-error mode in the th energy storage data sequence and the data segment corresponding to the historical high-error mode in the th energy storage data sequence; is the number of energy storage data sequences, and in this embodiment ; is the summation symbol.

[0087] Furthermore, after obtaining the correction value of the similarity between the real-time high-error mode and any historical high-error mode in the same energy storage data sequence, the similar modes of the real-time high-error mode in each energy storage data sequence can be screened based on the correction value.

[0088] In an exemplary embodiment of the present invention, the determination of the similarity pattern of the real-time high-error pattern in each energy storage data sequence can be achieved through the following steps:

[0089] Arrange the historical high-error patterns of the same item according to the magnitude of the correction value of the similarity with the real-time high-error pattern, and use the set composed of several historical high-error patterns selected from the front to the back as the similarity pattern of the real-time high-error pattern.

[0090] Exemplarily, the number of historical high-error patterns in the th energy storage data sequence can be denoted as , then the first historical high-error patterns can be selected from the ordered historical high-error patterns in the th energy storage data sequence in the order from the front to the back, so as to obtain the similarity pattern of the real-time high-error pattern in the th energy storage data sequence. In this embodiment ; where is the number of historical high-error patterns in the similarity pattern of the real-time high-error pattern in the th energy storage data sequence; is the number of historical high-error patterns in the th energy storage data sequence; is the ceiling symbol.

[0091] S4: Use the similarity between each historical high-error pattern in the similarity pattern and the real-time high-error pattern as the weight to perform weighted averaging on the adjustment coefficients of all historical high-error patterns in the similarity pattern, so as to adjust the predicted value of the latest high-error data of the corresponding item based on the weighted adjustment coefficient, and perform photovoltaic energy storage control based on the adjusted predicted value.

[0092] Specifically, the weighted adjustment coefficient of the latest high-error data in each energy storage data sequence satisfies the following relational expression:

[0093] ;

[0094] Where is the weighted adjustment coefficient of the latest high-error data in the th energy storage data sequence; is the correction value of the similarity between the real-time high-error pattern and the th historical high-error pattern in the th energy storage data sequence; is the number of historical high-error patterns in the similarity pattern of the real-time high-error pattern in the th energy storage data sequence; is the summation symbol.

[0095] Furthermore, the calculation formula of the weighted adjustment coefficient can be used to calculate the weighted adjustment coefficient of the latest high-error data in each energy storage data sequence, so as to adjust the predicted value of the latest high-error data of the corresponding item based on the weighted adjustment coefficient.

[0096] In an exemplary embodiment of the present invention, the adjustment of the predicted value of the latest high-error data in each energy storage data sequence can be achieved through the following steps:

[0097] Perform a multiplication operation on the weighted adjustment coefficient and the predicted value of the latest high-error data of the corresponding item to complete the adjustment of the predicted value of the latest high-error data.

[0098] Optionally, the predicted value of the latest high-error data in each energy storage data sequence can also be adjusted by summing the weighted adjustment coefficient, so as to achieve real-time adjustment of the predicted value of each photovoltaic energy storage data.

[0099] Optionally, the predicted value of each photovoltaic energy storage data obtained through steps S1 - S4 can be provided to relevant staff to make corresponding adjustments to the photovoltaic energy storage after discovering a change trend in the photovoltaic energy storage data, and perform photovoltaic energy storage control.

[0100] It should be understood that various alternative solutions to the embodiments of the present invention described herein can be adopted in the process of practicing the present invention.

Claims

1. A photovoltaic energy storage control method based on intelligent learning, characterized in that Including: Obtain multiple photovoltaic energy storage data collected at each sampling point to obtain multiple energy storage data sequences; Use the exponential smoothing method to predict each data in each energy storage data sequence, and regard the data with a prediction error greater than the preset error threshold in each energy storage data sequence as high-error data, and regard the continuous high-error data segment with the number of consecutive occurrences of high-error data greater than or equal to the preset value as a historical high-error pattern, so as to obtain all historical high-error patterns in each energy storage data sequence; Based on the prediction error, screen the historical high-error patterns in each energy storage data sequence, and with the goal of reducing the prediction error, calculate the adjustment coefficient of each historical high-error pattern based on the number of data in each historical high-error pattern and the ratio of the actual value to the predicted value of the data in the corresponding historical high-error pattern; Obtain the real-time high-error pattern of each energy storage data sequence, arrange the historical high-error patterns of the same item according to the magnitude of the correction value of the similarity with the real-time high-error pattern, and use the set composed of several historical high-error patterns selected from front to back as the similar pattern of the real-time high-error pattern; Use the similarity between each historical high-error pattern in the similar pattern and the real-time high-error pattern as the weight, perform weighted averaging on the adjustment coefficients of all historical high-error patterns in the similar pattern, so as to adjust the predicted value of the latest high-error data of the corresponding item based on the weighted adjustment coefficient, and perform photovoltaic energy storage control based on the adjusted predicted value.

2. The photovoltaic energy storage control method based on intelligent learning according to claim 1, wherein The method for obtaining the similarity between the real-time high-error pattern and the historical high-error pattern of the same item includes: For the real-time high-error pattern and any historical high-error pattern in the same energy storage data sequence, regard the ratio of the difference in adjacent data values in the real-time high-error pattern to the difference in adjacent data values in the any historical high-error pattern as the first index; regard the difference between the ratio of the average value of the data values in the real-time high-error pattern to the average value of the data predicted values and the ratio of the average value of the data values in the any historical high-error pattern to the average value of the data predicted values as the second index; Calculate the similarity between the real-time high-error pattern and the historical high-error pattern in the same energy storage data sequence. The difference between 1 and the first index, and the second index and the similarity are negatively correlated.

3. The photovoltaic energy storage control method based on intelligent learning according to claim 2, wherein The similarity satisfies the following relational expression: ； Wherein, is the similarity between the real-time high error mode and the th historical high error mode in the th energy storage data sequence; , are the values of the th data and the th data in the real-time high error mode respectively; , are the values of the th data and the th data in the historical high error mode respectively; , are the average value and the average predicted value of all data in the real-time high error mode respectively; , are the average value and the average predicted value of all data in the historical high error mode respectively; is the absolute value symbol; is the natural exponential function; is the number of data in the real-time high error mode.

4. The photovoltaic energy storage control method based on intelligent learning according to claim 3, wherein After determining the similarity between the real-time high-error pattern and any historical high-error pattern in the same energy storage data sequence, it further includes: Regard the similarity between the real-time high-error pattern in any energy storage data sequence and any historical high-error pattern in the any energy storage data sequence as the initial similarity, and respectively extract the data segment corresponding to the real-time high-error pattern in the any energy storage data sequence and the data segment corresponding to the any historical high-error pattern from the remaining energy storage data sequences to form two data segment sets; Calculate the average similarity of all data segments belonging to the same energy storage data sequence in the two data segment sets, and use the average similarity to correct the initial similarity. The correction value is the average of the average similarity and the initial similarity.

5. The photovoltaic energy storage control method based on intelligent learning according to claim 1, wherein The adjustment coefficients of the respective historical high-error modes satisfy the following relational expressions: ； In the formula, is the adjustment coefficient of the th historical high error mode in the th energy storage data sequence; is the value of the th data in the th historical high error mode in the th energy storage data sequence; is the predicted value of the th data in the th historical high error mode in the th energy storage data sequence; is the number of data in the th historical high error mode in the th energy storage data sequence; is the product symbol.

6. The photovoltaic energy storage control method based on intelligent learning according to claim 1, wherein, The error threshold is an adaptive error threshold, and the adaptive error threshold satisfies the following relational expressions: ； In the formula, is the adaptive error threshold of the -th data in the -th energy storage data sequence; is the maximum value of the prediction errors of all data in the -th energy storage data sequence; is the prediction error of the -th data in the -th energy storage data sequence; is the standard deviation of the prediction errors of all data in the -th energy storage data sequence; is the average prediction error value of all data in the -th energy storage data sequence; is the hyperbolic tangent function.

7. The photovoltaic energy storage control method based on intelligent learning according to claim 1, wherein Adjusting the predicted value of the latest high-error data for the corresponding item based on the weighted adjustment coefficient includes: Performing a multiplication operation on the weighted adjustment coefficient and the predicted value of the latest high-error data for the corresponding item to complete the adjustment of the predicted value of the latest high-error data.

8. The photovoltaic energy storage control method based on intelligent learning according to claim 1, wherein The multiple photovoltaic energy storage data includes real-time irradiance, temperature, and output power.

Images