Distributed photovoltaic bearing capacity prediction method and system based on multi-source data analysis

Abstract

The invention discloses a distributed photovoltaic bearing capacity prediction method and system based on multi-source data analysis, and the method comprises the steps: collecting a plurality of distributed photovoltaic sensing data sets, and calling historical photovoltaic power generation data for fusion, and obtaining a multi-source data set; photovoltaic bearing capacity analysis is carried out after the multi-source data set is preprocessed, and a plurality of characteristic values are obtained; performing photovoltaic bearing capacity prediction on the distributed photovoltaic according to the plurality of feature values to obtain an initial prediction result, and determining a prediction error interval according to the initial prediction result; and performing dynamic deviation correction on the initial prediction result according to the error interval to obtain a photovoltaic bearing capacity evaluation result. According to the method, the problem that the accuracy and timeliness of a photovoltaic bearing capacity prediction result are insufficient due to the fact that the prior art depends on a single data source and cannot fully fuse environment change and real-time data is solved, and the accuracy and real-time adaptability of distributed photovoltaic bearing capacity prediction are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of photovoltaic power generation, and particularly relates to a distributed photovoltaic carrying capacity prediction method and system based on multi-source data analysis. BACKGROUND

[0002] A distributed photovoltaic system refers to a solar power generation system installed near the user's location (such as a residential roof, a factory roof, etc.), which mainly adopts an operation mode of "self-generation for self-use, and surplus electricity on the grid". The system uses solar cell modules to convert solar energy into electrical energy, which is preferentially used for local loads, and the excess electricity can be connected to the grid. The distributed photovoltaic system has advantages such as green environmental protection, energy saving and consumption reduction, and flexible use of space, and is an important way to realize clean energy utilization and build a low-carbon society.

[0003] The carrying capacity prediction of a distributed photovoltaic system refers to the process of evaluating how much distributed photovoltaic power generation capacity a region or power grid can access under certain conditions without negatively affecting the safe and stable operation of the power grid. In simple terms, it is to predict how many photovoltaic systems can be installed in a place under the condition of the power grid's bearing capacity. Scientific and reasonable carrying capacity prediction can ensure the efficient and stable operation of distributed photovoltaic systems, and can provide decision support for optimizing resource allocation and reasonably scheduling power output.

[0004] There are existing technologies related to distributed photovoltaic systems. For example, the patent application with publication number CN116799780A constructs topology analysis, power flow analysis and carrying capacity calculation modules in a cloud platform, combines a main and distribution network integrated model with real-time power flow data, evaluates the accessible photovoltaic capacity of a distribution network to realize the evaluation of distributed photovoltaic carrying capacity; the patent application with publication number CN118971139A uses a neural network algorithm and distributed photovoltaic regulation technology to predict photovoltaic processing based on historical data and analyze the carrying capacity of the power grid; the patent application with publication number CN118263859A divides the power supply area by grid, evaluates the carrying capacity of new energy power sources at each voltage level, and predicts future grid load based on load balancing.

[0005] Although the existing technologies optimize the carrying capacity prediction effect of distributed photovoltaic systems to some extent, they mostly rely on historical data for modeling analysis. However, in the actual work process of distributed photovoltaic, environmental and climate conditions and equipment operating states change significantly over time and location, and these changes will affect the evaluation of the actual carrying capacity of distributed photovoltaic. The existing methods do not consider the changes of these environmental factors when predicting the carrying capacity, so they often cannot provide timely and accurate prediction information, affecting the optimized management and decision-making of photovoltaic systems. SUMMARY

[0006] The purpose of this part is to provide a distributed photovoltaic carrying capacity prediction method and system based on multi-source data analysis. By collecting multiple sensing data sets of distributed photovoltaics and calling historical photovoltaic power generation data for fusion, a multi-source data set is obtained. After preprocessing the multi-source data set, photovoltaic carrying capacity analysis is performed to obtain multiple characteristic values. According to the multiple characteristic values, the distributed photovoltaic carrying capacity is predicted to obtain an initial prediction result. According to the initial prediction result, the prediction error interval is determined. According to the error interval, the initial prediction result is dynamically corrected to obtain the photovoltaic carrying capacity evaluation result.

[0007] The first part of the application proposes a distributed photovoltaic carrying capacity prediction method based on multi-source data analysis, which adopts the following technical scheme:

[0008] Step 1, sensing and collecting distributed photovoltaics through multiple different data sources to obtain multiple sensing data sets;

[0009] Step 2, according to the data source, historical photovoltaic power generation data matching the real-time sensing data is called and fused with the multiple sensing data sets to obtain a multi-source standard data set after preprocessing;

[0010] Step 3, traversing the multi-source standard data set for photovoltaic carrying capacity analysis to obtain multiple characteristic values;

[0011] Step 4, based on the multiple characteristic values, a distributed photovoltaic carrying capacity dynamic prediction model is constructed, the distributed photovoltaic carrying capacity is initially predicted through the distributed photovoltaic carrying capacity dynamic prediction model, error determination is performed according to the initial prediction result, and the prediction error interval is determined;

[0012] Step 5, according to the prediction error interval, the initial prediction result is dynamically corrected to obtain a real-time carrying capacity prediction result, feedback is performed according to the real-time carrying capacity prediction result, and a photovoltaic carrying capacity evaluation report is generated.

[0013] Further, the process of fusing the historical photovoltaic power generation data with the multiple sensing data sets is:

[0014] Extracting the feature vector of the real-time sensing data as a matching condition;

[0015] Extracting historical photovoltaic power generation data in the historical database that meets the preset distance measurement criteria with the matching condition;

[0016] According to the time sequence, the historical photovoltaic power generation data is paired with the real-time sensing data to form multiple matching data pairs;

[0017] The matching data pairs are fused based on a weighting rule to obtain a multi-source standard data set.

[0018] Further, the preprocessing step employs wavelet decomposition technology to denoise the multi-source data set.

[0019] Further, the step of photovoltaic carrying capacity analysis includes:

[0020] S301, traversing the multi-source standard data set for random combination to generate a plurality of standard data pairs;

[0021] S302, jointly analyzing the plurality of standard data pairs to obtain a joint correlation coefficient;

[0022] S303, analyzing the data boundary value of the plurality of standard data pairs to obtain a boundary correlation coefficient;

[0023] S304, jointly integrating the joint correlation coefficient and the boundary correlation coefficient to generate a comprehensive correlation coefficient of photovoltaic carrying capacity through an information sharing mechanism;

[0024] S305, using the comprehensive correlation coefficient as an index to search the multi-source standard data set, determine a plurality of initial features and integrate and construct a feature array; dimensionally reduce the feature array to obtain the plurality of feature values.

[0025] Further, a nonlinear dimension reduction method based on similarity probability distribution modeling and KL divergence optimization is employed to dimensionally reduce the feature array.

[0026] Further, a distributed photovoltaic carrying capacity dynamic prediction model is constructed based on the plurality of feature values, and an initial prediction of photovoltaic carrying capacity of distributed photovoltaic is made through the distributed photovoltaic carrying capacity dynamic prediction model, including:

[0027] A long short-term memory network is employed to model distributed photovoltaic according to the plurality of feature values according to time series to construct a distributed photovoltaic carrying capacity dynamic prediction model;

[0028] The multi-source data set is synchronized to the dynamic prediction model for short-term and long-term dependence analysis, respectively, to obtain an initial prediction result, which includes a short-term carrying capacity prediction value and a long-term carrying capacity prediction value.

[0029] Further, the determination method of the error prediction interval is:

[0030] A first error extreme value is determined according to the short-term carrying capacity prediction value;

[0031] A second error extreme value is determined according to the long-term carrying capacity prediction value;

[0032] The upper limit value of the first error extreme value is compared with the upper limit value of the second error extreme value to determine the upper limit value of the error prediction interval.

[0033] comparing the lower limit value of the first error extreme value with the lower limit value of the second error extreme value to determine the lower limit value of the error prediction interval.

[0034] Further, the step of dynamically correcting deviation comprises:

[0035] comparing the initial prediction result with the prediction error interval to determine whether the initial prediction result is in the prediction error interval;

[0036] If the initial prediction result is not in the prediction error interval, the value of the initial prediction result is adjusted to generate the real-time bearing capacity prediction result.

[0037] Further, the value adjustment process of the initial prediction result comprises:

[0038] When the initial prediction result is greater than the upper limit value of the error prediction interval, a first instruction is generated; the first instruction is used for numerical down-regulation of the initial prediction result;

[0039] When the initial prediction result is less than the lower limit value of the error prediction interval, a second instruction is generated; the second instruction is used for numerical up-regulation of the initial prediction result.

[0040] The second part of the application proposes a distributed photovoltaic bearing capacity prediction system based on multi-source data analysis, comprising:

[0041] A multi-source data acquisition module is used for sensing and collecting distributed photovoltaics through multiple different data sources to obtain multiple sensing data sets;

[0042] A multi-source data processing module is used for retrieving historical photovoltaic power generation data matched with real-time sensing data according to the data source, and fusing the historical photovoltaic power generation data with the multiple sensing data sets to obtain a multi-source standard data set after preprocessing;

[0043] A photovoltaic bearing capacity analysis module is used for performing photovoltaic bearing capacity analysis on the multi-source standard data set to obtain multiple characteristic values;

[0044] A photovoltaic bearing capacity prediction module is used for constructing a distributed photovoltaic bearing capacity dynamic prediction model based on the multiple characteristic values, performing initial prediction of photovoltaic bearing capacity of distributed photovoltaics through the distributed photovoltaic bearing capacity dynamic prediction model, performing error determination according to the initial prediction result, and determining a prediction error interval;

[0045] A dynamic deviation correction module is used for dynamically correcting the initial prediction result according to the prediction error interval to obtain a real-time bearing capacity prediction result, performing feedback according to the real-time bearing capacity prediction result, and generating a photovoltaic bearing capacity evaluation report.

[0046] The above description is only a summary of the technical solutions of the present application, in order to enable the technical means of the present application to be implemented according to the content of the description, and in order to enable the above and other purposes, characteristics and advantages of the present application to be more apparent and easy to understand, the following specific embodiments of the present application are described.

[0047] The beneficial effects of the present application are:

[0048] 1. The present application constructs a high-dimensional data input system through multi-source data fusion. Compared with the photovoltaic bearing capacity prediction method of a single data source, the present application introduces environmental parameters, equipment operating state information and historical power generation data to form a multi-source data set, improves the perception ability of the prediction model to the complex state of the distributed photovoltaic system, and enables the prediction model to more comprehensively and accurately reflect the actual bearing capacity of the distributed photovoltaic system under different environmental conditions.

[0049] 2. The present application enhances the standardization and modeling efficiency of data processing through standardization preprocessing and feature extraction operations. The preprocessing method is used to unify various data formats and dimensions, reduce data redundancy and reduce noise interference of the model; at the same time, the feature analysis method is used to extract key parameters affecting photovoltaic bearing capacity, optimize the input dimension of the model, and improve the accuracy and running efficiency of the bearing capacity prediction model.

[0050] 3. The present application introduces a prediction error judgment and dynamic correction mechanism to correct the prediction error in real time on the basis of the traditional prediction process. The dynamic correction mechanism constructs the error extreme interval through short-term and long-term prediction models to realize dynamic adjustment of the initial prediction result, thereby improving the stability and timeliness of the prediction result, and enhancing the adaptability and reliability of the prediction model in actual application scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The flowchart of the distributed photovoltaic bearing capacity prediction method based on multi-source data analysis provided by the embodiments of the present application.

[0052] Figure 2 The structure diagram of the distributed photovoltaic bearing capacity prediction system based on multi-source data analysis provided by the embodiments of the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present application more clear, the technical solutions of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. The embodiments described in the present application are only a part of the embodiments of the present application, not all embodiments. Based on the spirit of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0054] The embodiment of the application provides a distributed photovoltaic carrying capacity prediction method and system based on multi-source data analysis, which solves the technical problem that the existing technology cannot fully integrate environmental changes and real-time data due to the dependence on a single data source, resulting in insufficient accuracy and timeliness of the distributed photovoltaic carrying capacity prediction result, and achieves the technical effect of improving the prediction accuracy and real-time adaptability of the distributed photovoltaic carrying capacity.

[0055] Embodiment one

[0056] As shown in the figure, the embodiment of the application provides a distributed photovoltaic carrying capacity prediction method based on multi-source data analysis, and the following is a specific implementation of the method. Figure 1

[0057] Step S1: Collecting distributed photovoltaic through multiple different data sources to obtain multiple sensing data sets.

[0058] Specifically, real-time sensing collection of distributed photovoltaic is performed by using multiple data sources such as sensors at different positions, different types of monitoring equipment, etc., to obtain multiple real-time sensing data sets. These sensing data sets include real-time meteorological data such as light intensity, temperature, humidity, wind speed, air pressure, etc.; real-time device characteristic data including photovoltaic panel type, installation angle, area, power generation efficiency, etc.; and real-time power grid state data including power grid load, power supply demand, voltage fluctuation, etc.

[0059] Each source of data forms a sensing data set, thereby generating multiple sensing data sets.

[0060] Step S2: Fusing historical photovoltaic power generation data with the multiple sensing data sets according to the data sources to obtain a multi-source standard data set after preprocessing;

[0061] The historical photovoltaic power generation data of the distributed photovoltaic is retrieved from the database (which can be a local database or a cloud database). The historical photovoltaic power generation data is fused with the multiple sensing data sets using data fusion technology to obtain a multi-source data set containing multiple types of information.

[0062] Further, the process of fusing the historical photovoltaic power generation data with the multiple sensing data sets includes:

[0063] Extracting a feature vector of the real-time sensing data as a matching condition;

[0064] Extracting historical photovoltaic power generation data in the historical database that meets a preset distance measurement criterion with the matching condition;

[0065] According to the time sequence, the historical photovoltaic power generation data is paired with the real-time sensing data to form multiple matching data pairs;​

[0066] The matching data pairs are fused based on a weighted rule to obtain a multi-source standard data set.

[0067] Specifically, the embodiment first extracts a feature vector of real-time sensing data, denoted as a matching condition; then traverses all feature vectors of sensing data corresponding to photovoltaic power generation moments from the database, and calculates the Euclidean distance between the feature vectors of sensing data and the matching condition;

[0068] The matching threshold is set as the Euclidean distance being less than ξ (for example, less than or equal to 0.25 in the embodiment), and a plurality of groups of historical photovoltaic power generation data satisfying the matching threshold are screened out;

[0069] Each group of historical photovoltaic power generation data forms a matching data pair with the real-time sensing data, and the matching data pairs are fused by weighting;

[0070] The plurality of fused matching data pairs are sorted according to the time sequence of the historical photovoltaic power generation data to obtain a multi-source data set.

[0071] Further, the preprocessing operation adopts a wavelet decomposition technique to denoise the multi-source data set.

[0072] Specifically, the PyWavelets library in Python is called to remove noise and redundant signals in the multi-source data set by wavelet decomposition. The obtained multi-source standard data set is free of noise interference and has higher data quality, which can provide more accurate data basis for subsequent photovoltaic bearing capacity analysis, and help to improve the accuracy and reliability of photovoltaic bearing capacity prediction.

[0073] Step S3, traversing the multi-source standard data set to perform photovoltaic bearing capacity analysis to obtain a plurality of characteristic values;

[0074] Further, the photovoltaic bearing capacity analysis includes the following steps.

[0075] S301, traversing the multi-source standard data set to perform random combination to generate a plurality of standard data pairs;

[0076] S302, jointly analyzing the plurality of standard data pairs to obtain a joint correlation coefficient;

[0077] S303, analyzing boundary value information of the plurality of standard data pairs to obtain a boundary correlation coefficient;

[0078] S304, jointly evaluating the joint correlation coefficient and the boundary correlation coefficient, sharing information according to an evaluation result, and generating a comprehensive correlation coefficient of photovoltaic bearing capacity;

[0079] S305, retrieve the multi-source standard data set by taking the comprehensive correlation coefficient as an index, determine a plurality of initial features, integrate and construct a feature array; perform dimension reduction processing on the feature array to obtain a plurality of feature values.

[0080] Specifically, step S301: random combination operation is performed in the multi-source standard data set to obtain a plurality of standard data pairs. For example, there are illumination intensity data and photovoltaic panel power generation efficiency data in the multi-source standard data set, and the (illumination intensity value, power generation efficiency value) formed after random combination is a standard data pair. These data pairs reflect the relationship between different data sources.

[0081] Step S302: use correlation analysis method (such as Pearson correlation coefficient calculation method) to analyze the correlation of photovoltaic bearing capacity of these standard data, calculate the joint correlation coefficient between the data in each data pair, and use it to measure the correlation degree between the standard data photovoltaic bearing capacity. The value range of joint correlation coefficient is [-1, 1], the larger the value, the higher the correlation degree. By randomly combining standard data pairs and calculating the joint correlation coefficient, the correlation of photovoltaic bearing capacity can be explored from the perspective of multiple data combinations, and the potential correlation between data can be mined, providing more information for comprehensive evaluation of photovoltaic bearing capacity.

[0082] Step S303: for each standard data pair, obtain its data boundary value, that is, the limit value of each data in the multi-source standard data set. For example, for a standard data pair containing illumination intensity and temperature, find the minimum value, maximum value of illumination intensity and the minimum value, maximum value of temperature, etc. Then, use the correlation analysis method (such as Spearman rank correlation coefficient calculation method) to analyze the correlation of photovoltaic bearing capacity of these data boundary values, and get the boundary correlation coefficient. The value range of boundary correlation coefficient is [-1, 1], the larger the value, the higher the correlation. The boundary correlation coefficient reflects the correlation degree between the data boundary condition and the photovoltaic bearing capacity. Considering the correlation between data boundary value and photovoltaic bearing capacity, the relationship between data in extreme condition and photovoltaic bearing capacity can be found, which helps to better understand the relationship between data and photovoltaic bearing capacity.

[0083] Step S304: jointly evaluate the joint correlation coefficient and the boundary correlation coefficient, that is, combine the joint correlation coefficient and the boundary correlation coefficient for comprehensive evaluation. In the process of joint evaluation, according to the situation of joint correlation coefficient and boundary correlation coefficient, the information is supplemented to each other to determine the comprehensive correlation coefficient of photovoltaic bearing capacity. This embodiment adopts weighted average method to jointly evaluate the joint correlation coefficient and the boundary correlation coefficient, and the process is as follows:

[0084] C int =ω1×C J +ω2×C B ;

[0085] wherein C int is a comprehensive correlation coefficient, C J is a joint correlation coefficient, C B is a boundary correlation coefficient; ω1 and ω2 are weights of the joint correlation coefficient and the boundary correlation coefficient, respectively, satisfying ω1 + ω2 = 1.

[0086] The closer the value of the comprehensive correlation coefficient is to 1, the stronger the positive correlation between the two is. For example, the Pearson correlation coefficient between the light intensity and the photovoltaic panel power generation efficiency is 0.8, indicating that the two have a strong positive correlation. Assuming that the Spearman rank correlation coefficient between the boundary value (maximum value and minimum value) of the light intensity and the photovoltaic carrying capacity is 0.6, indicating that the boundary value and the photovoltaic carrying capacity have a moderate degree of positive correlation. The comprehensive correlation coefficient calculated by the weighted average method (joint correlation coefficient weight 0.7, boundary correlation coefficient weight 0.3) is 0.74, indicating the correlation strength after considering the joint correlation coefficient and the boundary correlation coefficient. The comprehensive correlation coefficient obtained by simultaneous evaluation and information sharing can comprehensively analyze the correlation from two different angles of overall random combination and data boundary, more accurately reflect the relationship between the data and the photovoltaic carrying capacity, and provide a more reliable basis for the extraction of feature values.

[0087] Step S305: Search the multi-source standard data set using the comprehensive correlation coefficient as an index, and find the corresponding initial features according to the value or range of the comprehensive correlation coefficient. The value range of the comprehensive correlation coefficient is set according to actual needs combined with expert experience, for example, comprehensive correlation coefficient > 0.5. In the multi-source standard data set, find the initial features that meet the search conditions. For example, use the database query statement or the conditional filtering function in the programming language to filter out the initial features with a comprehensive correlation coefficient greater than 0.5. Then, arrange the searched initial features in time sequence or spatial distribution to construct a feature array. For example, in the data set, find the initial features with a comprehensive correlation coefficient greater than 0.5, and obtain the following results: light intensity comprehensive correlation coefficient 0.8; temperature comprehensive correlation coefficient 0.6; current comprehensive correlation coefficient 0.4 (not meeting the condition, excluded).

[0088] Then, integrate these initial features to construct a feature array. A nonlinear dimensionality reduction method based on similarity probability distribution modeling and KL divergence optimization is used to process the feature array for dimensionality reduction, and the main feature vectors are selected to construct a low-dimensional data structure, thereby obtaining a plurality of feature values. The steps include:

[0089] (1) Calculate the distribution probability of the feature array to obtain a distribution probability array;

[0090] (2) Construct a point array similarity probability distribution function, and perform dimension reduction processing on the point array distribution of the distribution probability array to generate a dimension reduction similarity divergence of the distribution probability array.

[0091] (3) Label the feature array based on the dimension reduction similarity divergence, and determine a plurality of feature values.

[0092] Specifically, the distribution probability of each element in the feature array is calculated. Different methods can be used according to the type of the feature. If the feature is discrete, a frequency calculation method can be used, that is, the number of occurrences of a certain feature value is divided by the total number of elements. If the feature is continuous, it can be assumed to conform to a certain probability distribution (such as a normal distribution), and then the distribution parameters are estimated by maximum likelihood estimation or other methods to obtain the distribution probability. By calculating the distribution probability of each element in the feature array, a new distribution probability array is obtained, and each element in the distribution probability array represents the distribution probability of the corresponding feature in the original feature array.

[0093] A point array similarity probability distribution function is constructed to measure the similarity between point array distributions. This function is based on the concept of probability distribution and provides a basis for dimension reduction processing by comparing the similarity between different point arrays (here, the point arrays in the distribution probability array). For example, the point array similarity probability distribution function can be constructed based on a probability distribution distance measurement method, such as KL divergence (Kullback-Leibler Divergence) or JS divergence (Jensen-Shannon Divergence). This dimension reduction similarity divergence is a measurement value obtained by dimension reduction processing of the distribution probability array using the point array similarity probability distribution function, and reflects the dispersion degree of the dimension-reduced distribution probability array in terms of similarity.

[0094] The feature array is labeled according to the obtained dimension reduction similarity divergence. A certain threshold value can be set, and when the dimension reduction similarity divergence of a certain feature in the feature array meets certain conditions (such as being less than a certain threshold value), it is determined as a feature value. For example, if the dimension reduction similarity divergence is less than 0.1, the feature is considered to be a representative feature, and these features are labeled as feature values. The plurality of feature values obtained based on the probability distribution and the dimension reduction similarity divergence can more accurately reflect the key information in the feature array, and provide more effective data features for subsequent photovoltaic carrying capacity prediction.

[0095] Step S4, constructing a distributed photovoltaic carrying capacity dynamic prediction model based on the plurality of feature values, performing initial prediction of photovoltaic carrying capacity for distributed photovoltaic through the distributed photovoltaic carrying capacity dynamic prediction model, determining the prediction error interval according to the initial prediction result.

[0096] Further, the long short-term memory network is used to model the distributed photovoltaic according to the time sequence according to the plurality of characteristic values, and a distributed photovoltaic carrying capacity dynamic prediction model is constructed.

[0097] The multi-source data set is synchronized to the dynamic prediction model for short-term and long-term dependence analysis respectively, and an initial prediction result is obtained, the initial prediction result including a short-term carrying capacity prediction value and a long-term carrying capacity prediction value.

[0098] Error analysis is performed according to the short-term carrying capacity prediction value to determine a first error extreme value, and error analysis is performed according to the long-term carrying capacity prediction value to determine a second error extreme value.

[0099] The upper limit value of the first error extreme value is compared with the upper limit value of the second error extreme value to determine the upper limit value of the error prediction interval.

[0100] The lower limit value of the first error extreme value is compared with the lower limit value of the second error extreme value to determine the lower limit value of the error prediction interval.

[0101] Specifically, step S401: first, the structure parameters of the long short-term memory network are determined, such as the number of hidden layers, the number of neurons in each hidden layer, etc. Then, based on the plurality of characteristic values extracted as described above, a characteristic matrix with time sequence attribute is constructed, wherein each row represents multi-dimensional feature data at a time point, and the whole forms a structure of "time step x feature dimension". Table 1 gives an example of the characteristic matrix.

[0102] Table 1: Characteristic matrix example

[0103] Time point Photovoltaic power Voltage fluctuation Solar radiation Temperature t1 100 kW 2％ 500 W / m2 32℃ t2 105 kW 1.8％ 520 W / m 2 ]] 33℃

[0104] Based on the time sequence characteristic matrix, a fixed length sliding time window strategy is used to construct model training samples, that is, the feature sequence in a continuous time period is used as the input of the long short-term memory network model, and the training sample set covering the whole historical period is gradually generated by window sliding. The above training sample set is input into the long short-term memory network, and supervised learning is used for training, wherein the supervised output label is the actual carrying capacity value at the next time point or time interval after the sliding window. In the training process, the weights of each layer are continuously adjusted through error back propagation, so that the prediction result gradually approaches the true data, until the model converges, and a constructed dynamic prediction model is obtained.

[0105] Step S402: The multi-source data set is synchronized to the already constructed dynamic prediction model according to time sequence. In the dynamic prediction model, the mutual dependence relationship of the input data (multi-source data set) in a short time range is analyzed to predict the photovoltaic carrying capacity in the short term, and a short-term photovoltaic carrying capacity prediction value is obtained.

[0106] Similarly, the multi-source data set is synchronized to the dynamic prediction model in time sequence, in which the interdependence of the input data in a long time range is analyzed to predict the photovoltaic carrying capacity in the long term, and a long-term photovoltaic carrying capacity prediction value is obtained.

[0107] The short-term and long-term dependence analyses use the same dynamic prediction model, and the long-term and short-term predictions are realized by adjusting the time step of the input data. The short-term dependence handles the minute or hour level historical features, and the long-term dependence handles the day level sequence evolution. Through the time recursion mechanism of the long short-term memory network structure, the short-term prediction value and the long-term trend prediction value are respectively output. The long-term carrying capacity prediction value and the short-term carrying capacity prediction value are both numerical indicators, i.e. the maximum capacity of the distributed photovoltaic that the power grid can accommodate, and the unit is the actual physical quantity. For example, the short-term carrying capacity prediction value output is the carrying capacity value at a specific time point, which is "the maximum accessible capacity in the next three hours is 850kW"; the long-term carrying capacity prediction value output is the carrying capacity interval range of the time sequence, which is "the daily average carrying capacity in the next 7 days is 720kW±5%".

[0108] Step S403: Collect the actual value of the short-term photovoltaic carrying capacity from the historical data or actual measurement, and then calculate the difference between the short-term photovoltaic carrying capacity prediction value and the actual value. For example, the mean square error (MSE) formula or the mean absolute error (MAE) formula can be used. By analyzing these error values, the first error upper limit value and the first error lower limit value are determined. The first error upper limit value and the first error lower limit value describe the range of the short-term prediction error. Similarly, the actual value of the long-term photovoltaic carrying capacity is collected, the difference between the long-term photovoltaic carrying capacity prediction value and the actual value is calculated, and then the second error upper limit value and the second error lower limit value are determined.

[0109] Step S404: Compare the first error upper limit value with the second error upper limit value, and select the larger value as the upper limit value of the error prediction interval. Compare the first error lower limit value with the second error lower limit value, and select the smaller value as the lower limit value of the error prediction interval.

[0110] Step S5, dynamically correct the initial prediction result according to the prediction error interval to obtain a real-time carrying capacity prediction result, and generate a photovoltaic carrying capacity evaluation report according to the real-time carrying capacity prediction result.

[0111] Further, the initial prediction result is compared with the prediction error interval, and if the initial prediction result is not in the prediction error interval, the value of the initial prediction result is adjusted to generate a real-time carrying capacity prediction result.

[0112] The value adjustment process of the initial prediction result includes:

[0113] When the initial prediction result is greater than the upper limit value of the error prediction interval, a first instruction is generated; the first instruction is used to numerically down-regulate the initial prediction result; when the initial prediction result is less than the lower limit value of the error prediction interval, a second instruction is generated; the second instruction is used to numerically up-regulate the initial prediction result.

[0114] Specifically, the initial prediction result and the determined prediction error interval are obtained, and it is judged whether the initial prediction result is within the prediction error interval. If the initial prediction result is not within the prediction error interval, the initial prediction result is further compared with the upper limit value and the lower limit value of the error prediction interval, respectively, to determine whether the initial prediction result is greater than the upper limit value of the error prediction interval or less than the lower limit value of the error prediction interval.

[0115] When it is determined that the initial prediction result is greater than the upper limit value of the error prediction interval, it indicates that the prediction value is too high, and the prediction result needs to be down-regulated to avoid affecting subsequent decision-making. At this time, a first instruction is generated, which is used to numerically down-regulate the initial prediction result. The first instruction can determine the down-regulation amplitude according to the actual situation. For example, the down-regulation can be performed according to a certain proportion (such as 10% of the initial prediction result), or a specific numerical value can be determined according to historical data and the characteristics of the model for down-regulation. Then, the initial prediction result is subtracted by the corresponding down-regulation numerical value to obtain a real-time bearing capacity prediction result.

[0116] When it is determined that the initial prediction result is less than the lower limit value of the error prediction interval, it indicates that the prediction value is too low, and needs to be up-regulated. At this time, a second instruction is generated, which is used to numerically up-regulate the initial prediction result. Similarly, the second instruction can determine the up-regulation amplitude or algorithm according to the actual situation, for example, up-regulation can be performed according to 15% of the initial prediction result or a specific numerical value can be determined according to other related factors for up-regulation. Then, the initial prediction result is added by the corresponding up-regulation numerical value to obtain a real-time bearing capacity prediction result.

[0117] By adjusting the initial prediction result in real time, the reliability of the prediction result is improved, thereby providing more accurate prediction data for the operation and management of the distributed photovoltaic system and avoiding incorrect decisions caused by abnormal prediction.

[0118] According to the real-time bearing capacity prediction result, a report generation tool is called to automatically generate a photovoltaic bearing capacity evaluation report. The report includes the following contents:

[0119] (1) Prediction data summary: including short-term bearing capacity prediction value, long-term bearing capacity prediction interval, and other data.

[0120] (2) Historical comparison analysis chart: the real-time bearing capacity prediction result is compared with historical data through a visual chart.

[0121] The generated report is presented in a combination of graphs and text and is automatically fed back to the dispatch center to adjust the photovoltaic access.

[0122] The embodiment of the present application adopts multiple data sources for fusion, ensuring comprehensive acquisition of real-time information of photovoltaic systems and environment, and improving sensitivity to variable environmental factors. Through the dynamic correction mechanism, on the basis of the preliminary prediction result, real-time adjustment is made according to the actual situation, which significantly improves the accuracy and real-time adaptability of the prediction result. The real-time feedback and report generation mechanism not only provides scientific bearing capacity prediction, but also provides operable decision support for operation and management personnel. Overall, the embodiment of the present application significantly improves the accuracy, reliability and timeliness of distributed photovoltaic bearing capacity prediction, and provides a solid technical guarantee for efficient operation and intelligent scheduling of photovoltaic systems.

[0123] Embodiment two

[0124] As shown in Figure 2 Based on the same inventive concept as the preceding embodiment one, the embodiment of the present application provides a distributed photovoltaic bearing capacity prediction system based on multi-source data analysis, which comprises:

[0125] A multi-source data acquisition module is used to collect and acquire multiple sets of sensing data of distributed photovoltaic through multiple different data sources.

[0126] A multi-source data processing module is used to retrieve historical photovoltaic power generation data matched with real-time sensing data according to the data source, and fuse the multiple sets of sensing data to obtain a multi-source standard data set after preprocessing.

[0127] A photovoltaic bearing capacity analysis module is used to analyze the photovoltaic bearing capacity by traversing the multi-source standard data set to obtain multiple characteristic values.

[0128] A photovoltaic bearing capacity prediction module is used to construct a distributed photovoltaic bearing capacity dynamic prediction model based on the multiple characteristic values, to perform initial prediction of the photovoltaic bearing capacity of the distributed photovoltaic through the distributed photovoltaic bearing capacity dynamic prediction model, to perform error determination according to the initial prediction result to determine a prediction error interval.

[0129] A dynamic correction module is used to perform dynamic correction on the initial prediction result according to the prediction error interval to obtain a real-time bearing capacity prediction result, to perform feedback according to the real-time bearing capacity prediction result, and to generate a photovoltaic bearing capacity evaluation report.

[0130] Through the foregoing detailed description of the method for predicting distributed photovoltaic bearing capacity based on multi-source data analysis, those skilled in the art can clearly understand the system for predicting distributed photovoltaic bearing capacity based on multi-source data analysis in the embodiment. For the system disclosed in Embodiment Two, since it corresponds to the method disclosed in Embodiment One, it has corresponding functional modules and beneficial effects. For the related parts, refer to the method part for explanation.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, not to limit it. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application. Any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered within the protection scope of the claims of the present application.

Claims

1. A distributed photovoltaic carrying capacity prediction method based on multi-source data analysis, characterized in that, The application relates to a distributed photovoltaic power generation capacity prediction method, which comprises the following steps: Step 1, real-time sensing collection of distributed photovoltaic power generation is carried out through multiple different data sources, multiple sensing data sets are obtained, and the multiple sensing data sets comprise real-time meteorological data, real-time equipment characteristic data and real-time power grid state data; Step 2, historical photovoltaic power generation data matched with the real-time sensing data are called according to the data source, and the historical photovoltaic power generation data are fused with the multiple sensing data sets to obtain multiple source standard data sets after preprocessing; Step 3, photovoltaic power generation capacity analysis is carried out on the multiple source standard data sets, and multiple characteristic values are obtained; Step 4, a distributed photovoltaic power generation capacity dynamic prediction model is constructed based on the multiple characteristic values, initial prediction of photovoltaic power generation capacity of distributed photovoltaic power generation is carried out through the distributed photovoltaic power generation capacity dynamic prediction model, error interval is determined according to the initial prediction result, dynamic correction is carried out on the initial prediction result according to the error interval, real-time capacity prediction result is obtained, and feedback is carried out according to the real-time capacity prediction result. The fusion process of the historical photovoltaic power generation data and the multiple sensing data sets is as follows:

2. The method for distributed photovoltaic carrying capacity prediction based on multi-source data analysis according to claim 1, wherein, a feature vector of the real-time sensing data is extracted as a matching condition; historical photovoltaic power generation data meeting a preset distance measurement criterion with the matching condition in a historical database are extracted; the historical photovoltaic power generation data are paired with the real-time sensing data according to a time sequence to form multiple matching data pairs; the matching data pairs are fused based on a weighting rule to obtain a multiple source data set. The preprocessing step adopts a wavelet decomposition technology to carry out noise reduction on the multiple source data set.

3. The distributed photovoltaic load bearing capacity prediction method based on multi-source data analysis of claim 1, wherein, The photovoltaic power generation capacity analysis step comprises the following steps:

4. The distributed photovoltaic load bearing capacity prediction method based on multi-source data analysis of claim 1, wherein, S301, multiple standard data pairs are generated by randomly combining the multiple source standard data sets; S302, joint analysis is carried out on the multiple standard data pairs to obtain a joint correlation coefficient; S303, data boundary values of the multiple standard data pairs are analyzed to obtain a boundary correlation coefficient; S304, the joint correlation coefficient and the boundary correlation coefficient are fused to generate a comprehensive correlation coefficient of photovoltaic power generation capacity through an information sharing mechanism; S305, the comprehensive correlation coefficient is used as an index to search the multiple source standard data set, multiple initial characteristics are determined, a characteristic array is constructed, dimension reduction processing is carried out on the characteristic array, and the multiple characteristic values are obtained. A nonlinear dimension reduction method based on similarity probability distribution modeling and KL divergence optimization is adopted to carry out dimension reduction processing on the characteristic array.

5. The distributed photovoltaic load bearing capacity prediction method based on multi-source data analysis of claim 4, wherein, The distributed photovoltaic power generation capacity dynamic prediction model is constructed based on the multiple characteristic values, and initial prediction of photovoltaic power generation capacity of distributed photovoltaic power generation is carried out through the distributed photovoltaic power generation capacity dynamic prediction model, and the method comprises the following steps:

6. The distributed photovoltaic load bearing capacity prediction method based on multi-source data analysis of claim 1, wherein, a long short-term memory network is adopted to model distributed photovoltaic power generation according to the multiple characteristic values in a time sequence, and a distributed photovoltaic power generation capacity dynamic prediction model is constructed; the multiple source data sets are synchronously input into the dynamic prediction model for short-term dependence analysis and long-term dependence analysis, and an initial prediction result is obtained, wherein the initial prediction result comprises a short-term capacity prediction value and a long-term capacity prediction value. The error prediction interval determination method is as follows:

7. The distributed photovoltaic load bearing capacity prediction method based on multi-source data analysis of claim 1, wherein, ​ According to the short-term bearing capacity prediction value, error analysis is performed to determine a first error extreme value; According to the long-term bearing capacity prediction value, error analysis is performed to determine a second error extreme value; The upper limit value of the first error extreme value is compared with the upper limit value of the second error extreme value to determine the upper limit value of the error prediction interval; The lower limit value of the first error extreme value is compared with the lower limit value of the second error extreme value to determine the lower limit value of the error prediction interval.

8. The distributed photovoltaic load bearing capacity prediction method based on multi-source data analysis of claim 1, wherein, The step of dynamic correction includes: The initial prediction result is compared with the prediction error interval to determine whether the initial prediction result is in the prediction error interval; If the initial prediction result is not in the prediction error interval, the value of the initial prediction result is adjusted to generate the real-time bearing capacity prediction result.

9. The distributed photovoltaic load bearing capacity prediction method based on multi-source data analysis of claim 8, wherein, The value adjustment process of the initial prediction result includes: When the initial prediction result is greater than the upper limit value of the error prediction interval, a first instruction is generated; the first instruction is used to adjust the value of the initial prediction result downward; When the initial prediction result is less than the lower limit value of the error prediction interval, a second instruction is generated; the second instruction is used to adjust the value of the initial prediction result upward.

10. A distributed photovoltaic carrying capacity prediction system based on multi-source data analysis, characterized in that, The system is used to execute the distributed photovoltaic bearing capacity prediction method based on multi-source data analysis according to any one of claims 1-9, including: A multi-source data acquisition module is used to collect and acquire a plurality of sensing data sets through a plurality of different data sources; A multi-source data processing module is used to retrieve historical photovoltaic power generation data matched with real-time sensing data according to the data source, and fuse the historical photovoltaic power generation data with the plurality of sensing data sets to obtain a multi-source standard data set after preprocessing; A photovoltaic bearing capacity analysis module is used to analyze photovoltaic bearing capacity by traversing the multi-source standard data set to obtain a plurality of characteristic values; A photovoltaic bearing capacity prediction module is used to construct a distributed photovoltaic bearing capacity dynamic prediction model based on the plurality of characteristic values, perform initial prediction of photovoltaic bearing capacity of the distributed photovoltaic through the distributed photovoltaic bearing capacity dynamic prediction model, and determine a prediction error interval according to the initial prediction result; A dynamic correction module is used to perform dynamic correction on the initial prediction result according to the prediction error interval to obtain a real-time bearing capacity prediction result, and generate a photovoltaic bearing capacity evaluation report according to the real-time bearing capacity prediction result.

Images