Distributed photovoltaic power prediction set fault diagnosis model

Abstract

The invention relates to the technical field of photovoltaic power generation, and discloses a distributed photovoltaic power prediction set fault diagnosis model, which comprises the following steps: calculating the correlation of distributed photovoltaic power output influence factors according to sampling values of the distributed photovoltaic power output influence factors in a historical day and the actual power output of the distributed photovoltaic power output influence factors in the historical day; calculating comprehensive correlation according to the correlation of the intra-day time factor quantized data and the correlation of the inter-day time factor quantized data; obtaining preliminary prediction data according to the comprehensive correlation between the strong correlation moment and the weak correlation moment in the historical day; performing error correction on the preliminary prediction data to obtain final prediction data; preprocessing the final prediction data, and extracting a plurality of features of the final prediction data; the distributed photovoltaic fault type is obtained through the sample similarity; the problem that the prediction data only depends on data fitting and sample training prediction is effectively avoided, the accuracy of the prediction data is improved, and the adaptability of fault diagnosis is enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of photovoltaic power generation, and specifically to a distributed photovoltaic power prediction and fault diagnosis model. Background Art

[0002] With the advancement of the construction of the new power system, distributed photovoltaic power generation has developed rapidly and has become the third largest renewable energy support after hydropower and wind power. The large-scale access of distributed photovoltaic power generation has significantly changed the load characteristics of the distribution network, increasing the difficulty of distribution network regulation and threatening the overall safety of the distribution network. Currently, the distributed photovoltaic power generation does not have a complete and refined power output prediction and reliable fault diagnosis mechanism. The lack of modeling of the influencing factors of power prediction for each distributed photovoltaic power generation unit and the insufficient feature discrimination of fault diagnosis directly affect the system safety and social benefits of the overall distribution network. Therefore, in order to promote the stable and safe grid connection and consumption of distributed photovoltaic power generation, it is urgent to design a distributed photovoltaic power prediction and fault diagnosis model with comprehensive consideration factors, accurate relevant metrics, and fine feature extraction to address the threats to safe and stable operation in the future.

[0003] The existing power prediction and fault diagnosis of distributed photovoltaic power generation usually only rely on the data fitting and sample training of the power output curve. It uses historical power output samples as input to the model and trains through large models such as deep learning and neural networks to achieve prediction + diagnosis of distributed photovoltaics. Although this scheme can to a certain extent characterize the future power output trend of distributed photovoltaics, it does not consider the influencing factors such as output volatility, time interval intermittency, and environmental coercion during the distributed photovoltaic power generation process. It is difficult to comprehensively explain the subsequent power output fluctuations only through the data training of large models, which affects the knowledge deconstruction of the distributed photovoltaic power output curve: on the one hand, this scheme simply takes the distributed photovoltaic power output curve as training data, lacking the refinement of influencing factors and the description of relevance, resulting in data-driven rather than knowledge-driven; on the other hand, this scheme is limited by the inherent defects of most large models and is prone to overlearning of power output data at a certain time section, leading to problems such as overfitting and local optimality. In the future, with the continuous access of distributed photovoltaics, the prediction + diagnosis scheme that only relies on data fitting and sample training cannot adapt to the possible risks in the future. Summary of the Invention

[0004] The purpose of the present application is to provide a distributed photovoltaic power prediction and fault diagnosis model that effectively avoids the problem of prediction data relying only on data fitting and sample training, improves the accuracy of prediction data, and enhances the adaptability of fault diagnosis.

[0005] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A distributed photovoltaic power prediction and fault diagnosis model of the present invention includes the following steps:

[0007] S1. Sample the possible influencing factors of the distributed photovoltaic power output on the historical day to obtain the sampling values of the influencing factors of the distributed photovoltaic power output on the historical day, and obtain the actual power output of the distributed photovoltaic on the historical day;

[0008] S2. Calculate the correlation of the influencing factors of the distributed photovoltaic power output according to the sampling values of the influencing factors of the distributed photovoltaic power output on the historical day and the actual power output of the distributed photovoltaic on the historical day;

[0009] S3. Obtain the power output of the distributed photovoltaic at different times within the historical day and the corresponding light intensity, and obtain the quantization data of the intra-day time factor according to the power output and light intensity at different times within the day; obtain the quantization data of the inter-day time factor according to the different distances between the prediction day and the historical day;

[0010] S4. According to the quantization data of the intra-day time factor and the quantization data of the inter-day time factor, obtain the correlation of the quantization data of the intra-day time factor and the correlation of the quantization data of the inter-day time factor through step S2;

[0011] S5. Calculate the comprehensive correlation at different times of the historical day according to the correlation of the quantization data of the intra-day time factor and the correlation of the quantization data of the inter-day time factor;

[0012] S6. Extract the strongly correlated moments and weakly correlated moments within the historical day according to the comprehensive correlation, and obtain the preliminary prediction data according to the comprehensive correlation of the strongly correlated moments and weakly correlated moments in the historical day and the actual power output of the distributed photovoltaic at the strongly correlated moments and weakly correlated moments in the historical day;

[0013] S7. Perform error correction on the preliminary prediction data to obtain the final prediction data;

[0014] S8. Preprocess the final prediction data, and extract the single-dimensional features, multi-dimensional features and DTW distance of the preprocessed final prediction data;

[0015] S9. According to the extracted single-dimensional features, multi-dimensional features and DTW distance, obtain the fault type of the distributed photovoltaic through sample similarity.

[0016] Furthermore, the method for calculating the correlation of the influencing factors of the distributed photovoltaic power output according to the sampling values of the influencing factors of the distributed photovoltaic power output on the historical day and the actual power output of the distributed photovoltaic on the historical day includes:

[0017] According to the sampled values of the distributed photovoltaic power output impact factor on the historical day and the actual power output of the distributed photovoltaic on the historical day, the direct correlation of the distributed photovoltaic power output impact factor is calculated through Formula 1:

[0018]

[0019] Wherein, represents the direct correlation of the distributed photovoltaic power output impact factor; T represents the historical output day of the distributed photovoltaic; s i,t represents all the sampled values of the distributed photovoltaic power output impact factor i on the historical day t; P t represents the actual power output of the distributed photovoltaic on the historical day t;

[0020] According to the sampled values of the distributed photovoltaic power output impact factor on the historical day, the correlation between the distributed photovoltaic power output impact factors is calculated through Formula 2:

[0021]

[0022] Wherein, R ij represents the correlation between the distributed photovoltaic power output impact factor i and the impact factor j; s j,t represents all the sampled values of the distributed photovoltaic output impact factor j on the historical day t;

[0023] According to the definition of correlation, the indirect correlation of each other impact factor j on the distributed photovoltaic power output through the impact factor i is obtained through Formula 3 as:

[0024]

[0025] Wherein, represents the indirect correlation of the distributed photovoltaic power output impact factor; J represents other impact factors except the impact factor i;

[0026] According to the direct correlation of the distributed photovoltaic power output impact factor and the direct and indirect correlations of the distributed photovoltaic power output impact factor, the interaction correlation of the distributed photovoltaic power output impact factor is obtained through Formula 4:

[0027]

[0028] Wherein, R i represents the interaction correlation of the distributed photovoltaic power output impact factor; α 1 , α 2 respectively represent the proportion components of the direct correlation and the indirect correlation within the impact factor i, and their values are related to the degree of influence of the impact factor i by other factors j. When the degree of indirect influence of the impact factor i on the distributed photovoltaic power output due to being affected by other factors is small, α2 < 0.5; conversely, when the degree to which the influence factor i indirectly affects the distributed photovoltaic power output due to the influence of other factors is large, α 2 ≥ 0.5.

[0029] Furthermore, the method for obtaining the quantization data of the intraday time factor according to the power output and light intensity at different times within a day includes:

[0030] Calculate the proportion of the distributed photovoltaic power output at different times in the historical day through Equation 5:

[0031]

[0032] where r τ represents the proportion of the power output at time τ in the historical day t; P tτ represents the power output at time τ in the historical day t; I tτ represents the light intensity at time τ in the historical day t;

[0033] Sort the different times τ in the historical day t according to the power magnitude. When the distributed photovoltaic power output reaches the maximum value, assign the maximum weight value τ to r τ and assign the second-largest weight value τ - 1 to r when obtaining the second-largest power output value in sequence, and assign the weight value 1 to the minimum power output value. Obtain the quantization data of the intraday time factor through Equation 6: τ

[0034]

[0035] where represents the quantization data of the intraday time factor, and w tτ represents the weight value;

[0036] The method for obtaining the quantization data of the daytime time factor according to the different distances between the prediction day and the historical day includes:

[0037] Obtain the quantization data of the daytime time factor through Equation 7:

[0038]

[0039] where represents the quantization data of the daytime time factor; ε be represents the daytime time factor constant; λ be represents the daytime time factor decay coefficient, presenting the characteristic of "larger near and smaller far" in the form of an exponential function.

[0040] Furthermore, the method for calculating the comprehensive correlation at different times in the historical day according to the correlation of the intraday time factor quantization data and the correlation of the daytime time factor quantization data includes:

[0041] The comprehensive correlation at different moments of the historical day is calculated by linearly weighting the correlation of the intraday time factor quantization data and the correlation of the daily time factor quantization data through Formula 8:

[0042]

[0043] Among them, C tτ represents the comprehensive correlation; c i represents the weight ratio of each influencing factor R of distributed photovoltaic i , represents the interaction correlation of the intraday time factor quantization data; represents the interaction correlation of the daily time factor quantization data.

[0044] Furthermore, the method for extracting the strongly correlated moments and weakly correlated moments in the historical intraday period based on the comprehensive correlation includes:

[0045] Based on the comprehensive correlation, those with a comprehensive correlation greater than the comprehensive correlation threshold are defined as strongly correlated moments through Formula 9, and vice versa, those with a comprehensive correlation less than or equal to the comprehensive correlation threshold are defined as weakly correlated moments:

[0046]

[0047] Among them, represents the comprehensive correlation threshold, n lim represents the minimum correlation moment threshold value, n s , n f respectively represent the strongly correlated moment and the weakly correlated moment;

[0048] When the number of weakly correlated moments is less than the minimum correlation moment threshold value, it is filled with the minimum correlation moment threshold value according to Formula 10, and the strongly correlated moment remains unchanged:

[0049]

[0050] Among them, n lim represents the minimum correlation moment threshold value, and n represents the value of the strongly and weakly correlated moments;

[0051] The method for obtaining the preliminary prediction data based on the comprehensive correlation of the strongly correlated moments and weakly correlated moments in the historical day, and the actual output of the distributed photovoltaic power at the strongly correlated moments and weakly correlated moments in the historical day includes:

[0052] The distributed photovoltaic power output at the strongly correlated moments directly participates in the prediction. For the prediction of the distributed photovoltaic power output at the weakly correlated moments, the average value of the distributed photovoltaic power outputs at n lim correlated moments participates in the prediction, and the first component of the preliminary predicted value of the distributed photovoltaic power at the τ-th moment in the next output day T+1 is obtained through Formula 11:

[0053]

[0054] Among them, represents the first component of the preliminary prediction value of the distributed photovoltaic power at the τ-th moment within the next output day T + 1; respectively represent the comprehensive correlation degree values at the strongly correlated moment and the weakly correlated moment in the historical days; respectively represent the actual output powers of the distributed photovoltaic power at the strongly correlated moment and the weakly correlated moment in the historical days; represents the average power output of the distributed photovoltaic at the weakly correlated moment in the historical day t;

[0055] Formula 12 predicts the second component of the preliminary prediction value of the distributed photovoltaic power at the τ-th moment within the next output day T + 1 through an extrapolation algorithm, and sums it with the first component of the preliminary prediction value of the distributed photovoltaic power within the next output day to obtain the preliminary prediction data:

[0056]

[0057] Among them, represents the actual output power of the distributed photovoltaic at each correlated moment τ in the historical day t; represents the actual output power of the distributed photovoltaic at each correlated moment τ - 1 in the historical day t; P T+1,τ represents the preliminary prediction data.

[0058] Furthermore, the method for error correction of the preliminary prediction data to obtain the final prediction data includes:

[0059] Perform clustering division on the preliminary prediction data through Formula 13, and divide the relative percentage error between the preliminary prediction value and the actual value according to the fuzzy c-means clustering algorithm:

[0060]

[0061] Among them, E rpe represents the relative percentage error between the predicted value and the actual value; P ac represents the actual value; E represents the state space; E 1 , …, E m represents the m clustering states obtained after fuzzy clustering;

[0062] Construct a Markov one-step state transition matrix through Formula 14:

[0063]

[0064] Among them, p 1 represents the one-step state transition matrix; p kqThe probability that state k transfers to state q, where k, q = 1, 2, …, m; N kq The frequency that state k transfers to state q; N k The total frequency of the occurrences of states;

[0065] Equation 14 satisfies the constraint of Equation 15:

[0066]

[0067] Equation 16 corrects the preliminary prediction data P through the partition of the clustered states and the one-step state transition matrix T+1,τ After being processed by the relative percentage error, E rpe The state it is in is corrected to obtain the final prediction data:

[0068]

[0069] Among them, Represents the final prediction data after Markov correction; E k-1 , E k Respectively represent the lower bound and upper bound of the interval where state E rpe Is located; ± is selected according to the state interval.

[0070] Furthermore, the method for preprocessing the final prediction data includes:

[0071] The wavelet transform and normalization processing adopted by Equation 17 are used to perform noise reduction and smoothing processing on the final prediction data based on the time-domain signal:

[0072]

[0073] Among them, ψ a,b(τ) Represents the mother wavelet; a represents the scale factor; b represents the translation factor; Represents the reconstructed distributed photovoltaic power output prediction data after wavelet denoising;

[0074] The normalization processing adopted by Equation 18 is used to eliminate the scale difference caused by noise in some time periods:

[0075]

[0076] Among them, Respectively represent the maximum power output value and the minimum power output value of the reconstructed distributed photovoltaic power output prediction data after wavelet denoising;

[0077] The methods for extracting the single-dimensional features, multi-dimensional features, and DTW distance of the preprocessed final prediction data include:

[0078] respectively corresponding to extracting single - dimensional features of five elements including maximum power, average power, power change rate, peak time, and degree of bending from distributed photovoltaic power output prediction data through Formula 19 to Formula 22:

[0079]

[0080] Through Formula 23, the probability density function and marginal probability density are used to measure two types of single - dimensional features, and the mutual information measure element of multi - dimensional features is extracted:

[0081]

[0082] Among them, U(X,Y) represents the joint probability density function of two single - dimensional vectors; u(x,y) represents the joint probability density of two single - dimensional vectors; u(x) and u(y) respectively represent the marginal probability densities of two single - dimensional vectors;

[0083] Through Formula 24, the Lyapunov exponent is used to measure, and the complexity measure element of multi - dimensional features is extracted:

[0084]

[0085] Among them, ρ represents the small - perturbation index of distributed photovoltaic power output prediction data; respectively represent the small - perturbation and average perturbation of distributed photovoltaic power output prediction data;

[0086] Through Formula 25, the DTW distance between various types of faults and normal operation of distributed photovoltaic power output prediction data is extracted:

[0087]

[0088] Among them, represents the DTW distance between various types of faults and normal operation of distributed photovoltaic; ω tτ represents the DTW weight in the τ - th period; respectively represent the power output prediction data of distributed photovoltaic under various types of faults and normal operation.

[0089] Furthermore, the methods for obtaining the fault types of distributed photovoltaic according to the extracted single - dimensional features, multi - dimensional features, and DTW distance include:

[0090] Normalize the maximum and minimum values of the eight elements of each extracted feature and map them to the range of [0, 1] to obtain the normalized values l e =[l 1 ,l 2 ,…,l 8 ;

[0091] The normalized values of the eight element features are standardized by a radar chart and represented in polar coordinates through Equation 26:

[0092]

[0093] where θ e represents the angle corresponding to the normalized value;

[0094] According to the eight elements on the radar chart, the polar coordinates can form a corresponding feature vector L e =[L 1 , L 2 , …, L 8 . If the normal template for the current distributed PV power processing is K e =[K 1 , K 2 , …, K 8 , the final fault matching is performed from the distance similarity and the angle similarity through Equation 27, and the Euclidean distance between the fault template and the normal template and the cosine similarity between the fault template and the normal template are obtained respectively:

[0095]

[0096] where d LK represents the Euclidean distance between the fault template and the normal template, and cosθ LK represents the cosine similarity between the fault template and the normal template;

[0097] When the Euclidean distance and the cosine similarity between the fault template and the normal template are both satisfied, the fault type of the distributed PV is obtained through Equation 28:

[0098]

[0099] where represents the fault type of the distributed PV; μ represents the fault type number; respectively represent the threshold values of the Euclidean distance and the cosine similarity between the fault template and the normal template.

[0100] Compared with the prior art, the beneficial effects of this application are:

[0101] The predicted data of the present invention takes into account the comprehensive correlation of various influencing factors including intraday time factor quantities and daily time factors, and extracts strong correlation moments and weak correlation moments based on the comprehensive correlation to reduce the error of the predicted data, effectively avoiding the problem that the predicted data only relies on data fitting and sample training, effectively improving the accuracy of the predicted data, extracting single-dimensional features, multi-dimensional features, and DTW distances of the predicted data, and obtaining the fault types of the predicted data according to the extracted feature elements, effectively avoiding problems such as overfitting and local optimality, and enhancing the adaptability of fault diagnosis; the present invention effectively avoids the problem that the predicted data only relies on data fitting and sample training, improves the accuracy of the predicted data, and enhances the adaptability of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0103] Figure 1 is a schematic flowchart of the method of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0104] The following specific examples illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0105] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Therefore, only the layers related to the present invention are shown in the diagrams, rather than being drawn according to the number, shape, and size of the layers in actual implementation. The type, quantity, and ratio of each layer in actual implementation can be arbitrarily changed, and the layer layout type may also be more complex.

[0106] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details.

[0107] Please refer to Figure 1 , a distributed photovoltaic power prediction set fault diagnosis model, including the following steps:

[0108] S1. Sample the possible impact factors of distributed photovoltaic power output on the historical day to obtain the sampling values of the impact factors of distributed photovoltaic power output on the historical day, and obtain the actual power output of the distributed photovoltaic on the historical day;

[0109] S2. Calculate the correlation of the impact factors of distributed photovoltaic power output according to the sampling values of the impact factors of distributed photovoltaic power output on the historical day and the actual power output of the distributed photovoltaic on the historical day;

[0110] S3. Obtain the power output of the distributed photovoltaic at different times within the historical day and the corresponding light intensity. According to the power output and light intensity at different times within the day, obtain the quantization data of the intra-day time factor; according to the different distances between the prediction day and the historical day, obtain the quantization data of the inter-day time factor;

[0111] S4. According to the quantization data of the intra-day time factor and the quantization data of the inter-day time factor, obtain the correlation of the quantization data of the intra-day time factor and the correlation of the quantization data of the inter-day time factor through step S2;

[0112] S5. Calculate the comprehensive correlation at different times of the historical day according to the correlation of the quantization data of the intra-day time factor and the correlation of the quantization data of the inter-day time factor;

[0113] S6. Extract the strongly correlated moments and weakly correlated moments within the historical day according to the comprehensive correlation, and obtain the preliminary prediction data according to the comprehensive correlation of the strongly correlated moments and weakly correlated moments in the historical day and the actual power output of the distributed photovoltaic at the strongly correlated moments and weakly correlated moments in the historical day;

[0114] S7. Perform error correction on the preliminary prediction data to obtain the final prediction data;

[0115] S8. Preprocess the final prediction data, and extract the single-dimensional features, multi-dimensional features and DTW distance of the preprocessed final prediction data;

[0116] S9. According to the extracted single-dimensional features, multi-dimensional features and DTW distance, obtain the fault type of the distributed photovoltaic through sample similarity.

[0117] The predicted data of the present invention takes into account the comprehensive correlation of various influencing factors including intraday time factor quantity and daily time factors, and extracts strong correlation moments and weak correlation moments according to the comprehensive correlation to reduce the error of the predicted data, effectively avoiding the problem that the predicted data only depends on data fitting and sample training prediction, effectively improving the accuracy of the predicted data, extracting single-dimensional features, multi-dimensional features and DTW distance of the predicted data, and obtaining the fault type of the predicted data according to the extracted feature elements, effectively avoiding problems such as overfitting and local optimum, and enhancing the adaptability of fault diagnosis; the present invention effectively avoids the problem that the predicted data only depends on data fitting and sample training prediction, improves the accuracy of the predicted data, and enhances the adaptability of fault diagnosis.

[0118] The factors affecting the output of distributed photovoltaic not only exist in each influencing factor itself, but also are reflected in the corresponding changes in the output of distributed photovoltaic caused by each influencing factor through other influencing factors. Therefore, the present invention constructs a correlation calculation model according to the series of influencing factors existing in the current distributed photovoltaic, including direct correlation degree calculation and indirect correlation degree calculation. Further, an interactive correlation degree calculation formula is established according to the obtained results to represent the actual effect of different influencing factors on the output of distributed photovoltaic.

[0119] The possible influencing factors on the output of distributed photovoltaic include: light radiation intensity, component conversion efficiency, atmospheric ambient temperature, on-site wind intensity, surface fouling degree. According to the sampling values of the influencing factors of distributed photovoltaic power output on historical days and the actual power output of distributed photovoltaic on historical days, the method for calculating the correlation of the influencing factors of distributed photovoltaic power output includes:

[0120] According to the sampling values of the influencing factors of distributed photovoltaic power output on historical days and the actual power output of distributed photovoltaic on historical days, the direct correlation of the influencing factors of distributed photovoltaic power output is calculated by formula 1:

[0121]

[0122] Among them, represents the direct correlation of the influencing factors of distributed photovoltaic power output; T represents the historical output day of distributed photovoltaic; s i,t represents all sampling values of the influencing factor i of distributed photovoltaic power output on the historical day t; P t represents the actual power output of distributed photovoltaic on the historical day t;

[0123] A single influencing factor can also affect the power output of distributed photovoltaic through other influencing factors. For example, the light radiation intensity itself, as an important factor affecting the power output of distributed photovoltaic, will also be indirectly affected by factors such as surface fouling degree and then affect the power output of distributed photovoltaic.

[0124] According to the sampled values of the distributed photovoltaic power output influencing factors on the historical day, the correlation between the distributed photovoltaic power output influencing factors is calculated by Formula 2:

[0125]

[0126] Wherein, R ij represents the correlation between the distributed photovoltaic power output influencing factor i and the influencing factor j; s j,t represents all the sampled values of the distributed photovoltaic output influencing factor j on the historical day t;

[0127] According to the definition of correlation, the indirect correlation of each other influencing factor j on the distributed photovoltaic power output through the influencing factor i is obtained by Formula 3 as:

[0128]

[0129] Wherein, represents the indirect correlation of the distributed photovoltaic power output influencing factor; J represents the other influencing factors except the influencing factor i;

[0130] According to the direct correlation of the distributed photovoltaic power output influencing factor and the direct and indirect correlation of the distributed photovoltaic power output influencing factor, the interaction correlation of the distributed photovoltaic power output influencing factor is obtained by Formula 4:

[0131]

[0132] Wherein, R i represents the interaction correlation of the distributed photovoltaic power output influencing factor; α 1 , α 2 respectively represent the proportion components of the direct correlation and the indirect correlation within the influencing factor i, and their values are related to the degree of influence of the influencing factor i by other factors j. When the degree of indirect influence of the influencing factor i on the distributed photovoltaic power output due to the influence of other factors is small, α 2 < 0.5; conversely, when the degree of indirect influence of the influencing factor i on the distributed photovoltaic power output due to the influence of other factors is large, α 2 ≥0.5.

[0133] The calculation of the correlation of the influencing factors provides a quantifiable index for power prediction from the external perspective of distributed photovoltaics; for more reasonable power output prediction of distributed photovoltaics, the present invention first analyzes from the perspective of the internal time correlation of distributed photovoltaics at a long time scale, and then calculates the comprehensive correlation degree by integrating the interaction correlation degree of the foregoing influencing factors;

[0134] From the intraday time scale, the proportion of the power generated at different times within a day in the total power generated throughout the day is significantly different. Therefore, to reflect such characteristics as the key factors for subsequent prediction, the method for obtaining the quantization data of the intraday time factor based on the power output and light intensity at different times within a day includes:

[0135] Calculate the proportion of the distributed photovoltaic power output at different times in the historical day through Equation 5:

[0136]

[0137] where r τ represents the proportion of the power output at time τ in the historical day t; P tτ represents the power output at time τ in the historical day t; I tτ represents the light intensity at time τ in the historical day t.

[0138] Sort the different times τ in the historical day t according to the power magnitude. When the distributed photovoltaic power output reaches the maximum value, assign the maximum weight value τ to r τ ; when obtaining the second largest power output value in sequence, assign the second largest weight value τ - 1 to r τ ; assign the weight value 1 to the minimum power output value, and obtain the quantization data of the intraday time factor through Equation 6:

[0139]

[0140] where represents the quantization data of the intraday time factor, and w tτ represents the weight value.

[0141] According to the quantization data of the intraday time factor, calculate the interaction correlation of the intraday time factor through Equations 1 to 4.

[0142] The method for obtaining the quantization data of the daily time factor according to the different distances between the prediction day and the historical day includes:

[0143] Obtain the quantization data of the daily time factor through Equation 7:

[0144]

[0145] where represents the quantization data of the daily time factor; ε be represents the daily time factor constant; λ be represents the daily time factor decay coefficient, presenting the characteristic of "larger nearby and smaller far away" in the form of an exponential function.

[0146] According to the quantization data of the daily time factor, calculate the interaction correlation of the daily time factor through Equations 1 to 4.

[0147] A method for calculating the comprehensive correlation at different moments of a historical day based on the correlation of intraday time factor quantization data and the correlation of daily time factor quantization data includes:

[0148] Performing a linear weighted calculation on the correlation of intraday time factor quantization data and the correlation of daily time factor quantization data through Formula 8 to calculate the comprehensive correlation at different moments of a historical day:

[0149]

[0150] where, C tτ represents the comprehensive correlation; c i represents the weight ratio of each influencing factor R of distributed photovoltaic i , represents the interactive correlation of intraday time factor quantization data; represents the interactive correlation of daily time factor quantization data.

[0151] The prediction of distributed photovoltaic power output is closely related to the comprehensive correlation at different moments in the sample historical day. Moments with strong comprehensive correlation account for a large proportion in power prediction, while moments with weak comprehensive correlation account for a relatively small proportion. Therefore, the present invention extracts strong correlation moments and weak correlation moments based on this comprehensive correlation. For strong correlation moments, directly take their power output to participate in the prediction; for weak correlation moments, fill them up according to the minimum correlation moment threshold value and participate in the prediction with the average power output;

[0152] A method for extracting strong correlation moments and weak correlation moments in a historical day based on comprehensive correlation includes:

[0153] According to the comprehensive correlation, those with a comprehensive correlation greater than the comprehensive correlation threshold are defined as strong correlation moments through Formula 9, and vice versa, those with a comprehensive correlation less than or equal to the comprehensive correlation threshold are defined as weak correlation moments:

[0154]

[0155] where, represents the comprehensive correlation threshold, n lim represents the minimum correlation moment threshold value, n s , n f respectively represent strong correlation moments and weak correlation moments;

[0156] When the number of weak correlation moments is less than the minimum correlation moment threshold value, it is defined through Formula 10 that the minimum correlation moment threshold value is taken for filling up, and the strong correlation moments remain unchanged:

[0157]

[0158] where, n limrepresents the minimum correlation time threshold value, and n represents the value of strong and weak correlation times;

[0159] The method for obtaining the preliminary prediction data according to the comprehensive correlation of strong and weak correlation times in the historical day and the actual output of distributed photovoltaic power at strong and weak correlation times in the historical day includes:

[0160] The distributed photovoltaic power output at strong correlation times directly participates in the prediction. For the distributed photovoltaic power output prediction at weak correlation times, the average value of the distributed photovoltaic power output at n lim correlation times participates in the prediction. The first component of the preliminary prediction value of the distributed photovoltaic power at the τ-th moment in the next output day T+1 is obtained through Formula 11:

[0161]

[0162] Among them, represents the first component of the preliminary prediction value of the distributed photovoltaic power at the τ-th moment in the next output day T+1; respectively represent the comprehensive correlation degree values of strong and weak correlation times in the historical day; respectively represent the actual outputs of distributed photovoltaic power at strong and weak correlation times in the historical day; represents the average power output of distributed photovoltaic at weak correlation times in the historical day t;

[0163] Formula 12 predicts the second component of the preliminary prediction value of the distributed photovoltaic power at the τ-th moment in the next output day T+1 through an extrapolation algorithm, and sums it with the first component of the preliminary prediction value of the distributed photovoltaic power at the -th moment in the next output day to obtain the preliminary prediction data:

[0164]

[0165] Among them, represents the actual output of distributed photovoltaic power at each correlation time τ in the historical day t; represents the actual output of distributed photovoltaic power at each correlation time τ-1 in the historical day t; P T+1,τ represents the preliminary prediction data.

[0166] The method for obtaining the final prediction data by performing error correction on the preliminary prediction data includes:

[0167] To further reduce the error caused by the average power output of distributed photovoltaic at weak correlation times, the present invention uses the Markov algorithm to correct the preliminary prediction data;

[0168] Cluster and divide the preliminary prediction data through Formula 13, and divide the relative percentage error between the preliminary predicted value and the actual value according to the fuzzy c-means clustering algorithm:

[0169]

[0170] Among them, E rpe represents the relative percentage error between the predicted value and the actual value; P ac represents the actual value; E represents the state space; E 1 ,…,E m represent m clustering states obtained after fuzzy clustering;

[0171] Construct a Markov one-step state transition matrix through Formula 14:

[0172]

[0173] Among them, p 1 represents the one-step state transition matrix; p kq represents the probability of transitioning from state k to state q, k, q = 1, 2, …, m; N kq represents the frequency of transitioning from state k to state q; N k represents the total frequency of the occurrence of states;

[0174] Formula 14 satisfies the constraint of Formula 15:

[0175]

[0176] Formula 16 corrects the state of the preliminary prediction data P T+1,τ after relative percentage error processing E rpe to obtain the final prediction data:

[0177]

[0178] Among them, represents the final prediction data corrected by Markov; E k-1 ,E k respectively represent the lower bound and upper bound of the interval where state E rpe is located; ± is selected according to the state interval.

[0179] The fault diagnosis of distributed photovoltaics is crucial for ensuring the performance and stability of the system. The physical detection method is not universal for fault diagnosis due to factors such as site and equipment. Therefore, this invention starts from the samples of distributed photovoltaic power output, analyzes the power output data of distributed photovoltaics in normal states and different fault conditions, establishes multi-dimensional characterizations of the power output sample data, and uses a sample similarity measurement scheme to match the corresponding fault diagnosis results.

[0180] The methods for preprocessing the final prediction data include:

[0181] To eliminate part of the noise in the distributed photovoltaic power output prediction data that affects fault discrimination, wavelet transform and normalization processing adopted by formula 17 are used to perform noise reduction and smoothing processing on the final prediction data based on time-domain signals:

[0182]

[0183] where ψ a,b(τ) represents the mother wavelet; a represents the scale factor; b represents the translation factor; represents the reconstructed distributed photovoltaic power output prediction data after wavelet denoising;

[0184] Normalization processing is adopted by formula 18 to eliminate the scale differences caused by noise in some time periods:

[0185]

[0186] where, respectively represent the maximum power output value and the minimum power output value of the reconstructed distributed photovoltaic power output prediction data after wavelet denoising;

[0187] The methods for extracting single-dimensional features, multi-dimensional features, and DTW distance of the final prediction data after preprocessing include:

[0188] Respectively, single-dimensional features of the distributed photovoltaic power output prediction data are extracted through five elements: maximum power, average power, power change rate, peak time, and degree of curvature, obtained by formulas 19 to 22:

[0189]

[0190] Maximum power output To reflect the possible maximum power differences of distributed photovoltaics when different faults occur;

[0191] Average power output To reflect the possible average output differences of distributed photovoltaics when different faults occur;

[0192] Power change rate (slope) to reflect the possible growth and decay differences of distributed PV when different faults occur;

[0193] Peak moment to reflect the moment differences of distributed PV reaching the maximum power when different faults occur;

[0194] Degree of bending to reflect the shape characteristics of the power output of distributed PV when different faults occur, that is, the degree of bending;

[0195] Refine the non - linear dependence relationship between single - dimension features;

[0196] Through formula 23, use the probability density function and marginal probability density to measure two types of single - dimension features, and extract the mutual information elements of multi - dimension features:

[0197]

[0198] Among them, U(X,Y) represents the joint probability density function of two single - dimension vectors; u(x,y) represents the joint probability density of two single - dimension vectors; u(x) and u(y) respectively represent the marginal probability densities of two single - dimension vectors;

[0199] To indicate the overall stability and fault sensitivity of the power output of distributed PV, use the Lyapunov exponent to measure through formula 24, and extract the complexity metric elements of multi - dimension features:

[0200]

[0201] Among them, ρ represents the small - perturbation index of the power output prediction data of distributed PV; respectively represent the small - perturbation and average perturbation of the power output prediction data of distributed PV;

[0202] The DTW distance is an algorithm for measuring the similarity of two sample data. Based on this, the present invention establishes a DTW distance library for various types of faults and normal operation of distributed PV to reliably diagnose subsequent faults;

[0203] Extract the DTW distances between various types of faults and normal operation of distributed PV in the power output prediction data of distributed PV through formula 25:

[0204]

[0205] Among them, represents the DTW distance between various types of faults and normal operation of distributed PV; ω tτ represents the DTW weight in the τ - th period; Respectively represent the power output prediction data of distributed PV under various types of faults and normal operation.

[0206] Construct the final similarity-based fault diagnosis model according to the five elements in the aforementioned single-dimensional features, the two elements in the multi-dimensional features, and the DTW distance element;

[0207] The method for obtaining the fault type of distributed PV according to the extracted single-dimensional features, multi-dimensional features, and DTW distance includes:

[0208] Referring to Equation 18, normalize the maximum and minimum values of the eight elements for each extracted feature and map them to the range [0, 1] to obtain the normalized values l of the eight-element features e =[l 1 ,l 2 ,…,l 8 ;

[0209] Standardize the normalized values of the eight-element features with a radar chart, and represent them in polar coordinates through Equation 26:

[0210]

[0211] where θ e represents the angle corresponding to the normalized value;

[0212] According to the eight elements on the radar chart, the polar coordinates can form the corresponding feature vector L e =[L 1 ,L 2 ,…,L 8 , if the normal template for the current distributed PV power processing is K e =[K 1 ,K 2 ,…,K 8 , perform the final fault matching from the distance similarity and the angle similarity through Equation 27 to obtain the Euclidean distance between the fault template and the normal template and the cosine similarity between the fault template and the normal template respectively:

[0213]

[0214] where d LK represents the Euclidean distance between the fault template and the normal template, and cosθ LK represents the cosine similarity between the fault template and the normal template;

[0215] When both the Euclidean distance and the cosine similarity between the fault template and the normal template are satisfied, obtain the fault type of the distributed PV through Equation 28:

[0216]

[0217] Among them, represents the fault type of distributed photovoltaic; μ represents the fault type number; respectively represent the threshold values of the Euclidean distance and cosine similarity between the fault template and the normal template.

[0218] The present invention proposes a calculation paradigm for the direct and indirect correlations of distributed photovoltaic impact factors considering sampling values and power output, and further constructs an interactive correlation calculation paradigm based on the degree of contribution of each impact factor indirectly affecting the distributed photovoltaic power output due to being affected by other factors;

[0219] At the same time, a distributed photovoltaic power prediction model based on time factor quantization and Markov correction is proposed. An intraday time factor and a daily time factor quantization model are constructed from the perspective of the inherent time correlation of distributed photovoltaic on a long time scale, and a comprehensive correlation model is constructed based on the interactive correlation of impact factors. On the one hand, the intraday and daily correlation features are used as the time attributes of the prediction data; on the other hand, the combination of interactive correlation and time factor quantization data provides rich attribute representations for the prediction data. Based on the comprehensive correlation, strong correlation moments and weak correlation moments are refined to form distributed photovoltaic power prediction data considering the differential ratio of correlation moments under Markov correction.

[0220] And a preprocessing scheme for distributed photovoltaic power output prediction data integrating wavelet transform and normalization is proposed. Single-dimensional features, multi-dimensional features, and features such as DTW distance of distributed photovoltaic power output prediction data are extracted. Accordingly, a feature vector radar chart model including Euclidean distance and cosine similarity is constructed, and a distributed photovoltaic fault diagnosis model design based on sample similarity is realized from two discrimination angles of distance similarity and angle similarity.

[0221] In the above embodiments, although the present invention has been described in combination with specific embodiments of the present invention, according to the previous description, many substitutions, modifications, and variations of these embodiments will be obvious to those of ordinary skill in the art. The embodiments of the present invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0222] The above embodiments are only illustrative of the principles and effects of the present invention and are not used to limit the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A distributed photovoltaic power prediction set fault diagnosis model, characterized by: The following steps are involved: S1. Sample the possible influencing factors of distributed photovoltaic power output on historical days, obtain the sampling values ​​of the influencing factors of distributed photovoltaic power output on historical days, and obtain the actual power output of distributed photovoltaic on historical days; S2. Calculate the correlation of the distributed photovoltaic power output influencing factors according to the sampling values ​​of the distributed photovoltaic power output influencing factors on the historical days and the actual power output of the distributed photovoltaic on the historical days; S3. Obtain the power output and corresponding light intensity of distributed photovoltaics at different times in historical days, and obtain the quantitative data of intraday time factors according to the power output and light intensity at different times in the day; obtain the quantitative data of daytime time factors according to the different distances between the predicted day and the historical day; S4, according to the intraday time factor quantified data and the inter-day time factor quantified data, obtaining the correlation of the intra-day time factor quantified data and the correlation of the inter-day time factor quantified data through step S2; S5. Calculate the comprehensive correlation at different times of the historical day based on the correlation of the quantitative data of the intraday time factor and the correlation of the quantitative data of the interday time factor; S6. Extract the strong correlation moments and weak correlation moments in the historical day according to the comprehensive correlation, and obtain preliminary prediction data according to the comprehensive correlation of the strong correlation moments and weak correlation moments in the historical day, and the actual output of distributed photovoltaic power at the strong correlation moments and weak correlation moments in the historical day; S7, performing error correction on the preliminary forecast data to obtain final forecast data; S8, preprocessing the final prediction data, and extracting single-dimensional features, multi-dimensional features, and DTW distance of the preprocessed final prediction data; S9. According to the extracted single-dimensional features, multi-dimensional features and DTW distance, the fault type of distributed photovoltaics is obtained through sample similarity.

2. A distributed photovoltaic power prediction set fault diagnosis model according to claim 1, characterized in that: According to the sampled values ​​of the distributed photovoltaic power output influencing factors on the historical days and the actual power output of the distributed photovoltaic on the historical days, the method for calculating the correlation of the distributed photovoltaic power output influencing factors includes: According to the sampling value of the distributed photovoltaic power output influencing factor on the historical day and the actual power output of the distributed photovoltaic on the historical day, the direct correlation of the distributed photovoltaic power output influencing factor is calculated by formula 1: in, represents the direct correlation of the factors affecting the power output of distributed photovoltaics; T represents the historical output day of distributed photovoltaics; s i,t represents all sampled values ​​of distributed photovoltaic power output influencing factor i on historical day t; P t represents the actual power output of distributed photovoltaic on historical day t; According to the sampling values ​​of the distributed photovoltaic power output influencing factors on historical days, the correlation between the distributed photovoltaic power output influencing factors is calculated by formula 2: Among them, R ij represents the correlation between the influencing factor i and the influencing factor j of distributed photovoltaic power output; s j,t represents all sampled values ​​of distributed photovoltaic output influencing factor j on historical day t; According to the definition of correlation, the indirect correlation of other influencing factors j on the distributed photovoltaic power output through influencing factor i is obtained through formula 3: in, represents the indirect correlation of the influencing factors of distributed photovoltaic power output; J represents other influencing factors except the influencing factor i; According to the direct correlation of the influencing factors of distributed photovoltaic power output and the direct correlation of the influencing factors of distributed photovoltaic power output, the interactive correlation of the influencing factors of distributed photovoltaic power output is obtained through formula 4: Among them, R i It represents the interactive correlation of the factors affecting distributed photovoltaic power output; α1 and α2 represent the proportion of direct correlation and indirect correlation within the influencing factor i, respectively. The value is related to the degree to which the influencing factor i is affected by other factors j. When the degree to which the influencing factor i is affected by other factors and indirectly affects the distributed photovoltaic power output is small, α2 < 0.5; on the contrary, when the degree to which the influencing factor i is affected by other factors and indirectly affects the distributed photovoltaic power output is large, α2 ≥ 0.

5.

3. A distributed photovoltaic power prediction set fault diagnosis model according to claim 2, characterized in that: According to the power output and light intensity at different times of the day, the methods for obtaining quantitative data of the intraday time factor include: The proportion of distributed photovoltaic power output at different times in the historical day is calculated by formula 5: Among them, r τ represents the power output proportion at time τ in historical day t; P tτ I represents the power output at time τ in historical day t; tτ represents the light intensity at time τ in historical day t; Sort the different moments τ in the historical day t by power size. When the distributed photovoltaic power output reaches the maximum value, r is assigned τ The maximum weight is τ, and r is assigned when the second largest power output value is obtained. τ The second largest weight is τ-1, and the minimum power output value is assigned a weight of 1. The quantitative data of the intraday time factor is obtained through formula 6: in, Represents the quantitative data of the intraday time factor, w tτ represents weight; According to the different distances between the forecast day and the historical day, the methods for obtaining the quantitative data of the daytime factor include: The quantitative data of the daytime factor is obtained by formula 7: in, Represents the quantitative data of the daytime factor; ε be represents the day time factor constant; λ be It represents the attenuation coefficient of the daytime factor, presenting the characteristic of "larger near and smaller far" in the form of an exponential function.

4. A distributed photovoltaic power prediction set fault diagnosis model according to claim 3, characterized in that: According to the correlation of the quantitative data of the intraday time factor and the correlation of the quantitative data of the interday time factor, the method of calculating the comprehensive correlation at different times of the historical day includes: The correlation of the intraday time factor quantitative data and the correlation of the interday time factor quantitative data are linearly weighted to calculate the comprehensive correlation at different times of the historical day through formula 8: Among them, C tτ represents comprehensive correlation; c i Represents each impact factor R of distributed photovoltaic i The weight ratio of Represents the mutual correlation of the quantitative data of the intraday time factor; Represents the interactive correlation of the quantitative data of the day time factor.

5. A distributed photovoltaic power prediction set fault diagnosis model according to claim 4, characterized in that: Methods for extracting strong correlation moments and weak correlation moments within historical days based on comprehensive correlation include: According to the comprehensive correlation, the strong correlation moment is defined as the time when the comprehensive correlation is greater than the comprehensive correlation threshold through formula 9, and the weak correlation moment is defined as the time when the comprehensive correlation is less than or equal to the comprehensive correlation threshold: in, Represents the comprehensive relevance threshold, n lim Represents the minimum association time threshold, n s ,n f Represent the strong correlation moment and the weak correlation moment respectively; Formula 10 defines that when the number of weakly associated moments is less than the minimum associated moment threshold, the minimum associated moment threshold is used to fill the gap, and the strongly associated moments remain unchanged: Among them, n lim represents the minimum correlation time threshold, and n represents the value of the strong and weak correlation time; According to the comprehensive correlation between the strong correlation moments and the weak correlation moments in the historical day, and the actual output of distributed photovoltaic power at the strong correlation moments and the weak correlation moments in the historical day, the method of obtaining preliminary prediction data includes: The distributed photovoltaic power output at the strongly correlated moment is directly predicted, and the distributed photovoltaic power output prediction at the weakly correlated moment is based on n lim The average value of the distributed photovoltaic power output at the associated moments is used for prediction. The first component of the preliminary prediction value of the distributed photovoltaic power at the τth moment in the next output day T+1 is obtained by formula 11: in, It represents the first component of the preliminary prediction value of distributed photovoltaic power at the time τ in the next output day T+1; They represent the comprehensive correlation values ​​of the strong correlation moments and weak correlation moments in the historical day respectively; They represent the actual output of distributed photovoltaic power at the strong correlation moment and the weak correlation moment in the historical day respectively; represents the average power output of distributed photovoltaic power generation at the weakly correlated moment in the historical day t; Formula 12 predicts the second component of the preliminary prediction value of the distributed photovoltaic power at the τth moment in the next output day T+1 through the extrapolation algorithm, and sums it with the first component of the preliminary prediction value of the distributed photovoltaic power at the τth moment in the next output day to obtain the preliminary prediction data: Among them, P tτ n represents the actual output of distributed photovoltaic power at each associated time τ in the historical day t; represents the actual output of distributed photovoltaic power at each associated time τ-1 in the historical day t; P T+1,τ Represents preliminary forecast data.

6. A distributed photovoltaic power prediction set fault diagnosis model according to claim 5, characterized in that: Methods for performing error correction on preliminary forecast data to obtain final forecast data include: The preliminary prediction data is clustered using formula 13, and the relative percentage errors between the preliminary prediction value and the actual value are divided according to the fuzzy mean clustering algorithm: Among them, E rpe Indicates the relative percentage error between the predicted value and the actual value; P ac represents the actual value; E represents the state space; E1,…,E m Represents the m clustering states obtained after fuzzy clustering; The Markov one-step state transfer matrix is ​​constructed by formula 14: Among them, p1 represents the one-step state transfer matrix; p kq represents the probability of state k transferring to state q, k,q=1,2,…,m; N kq Indicates the frequency of state k transferring to state q; N k Indicates the sum of the frequencies of the occurrence states; Formula 14 satisfies the constraints of Formula 15: Formula 16 uses the clustering state division and one-step state transfer matrix to predict the initial data P T+1,τ After relative percentage error processing, E rpe The current state is corrected to obtain the final prediction data: in, represents the final prediction data after Markov correction; E k-1 ,E k Respectively represent state E rpe The lower and upper bounds of the interval; ± is selected according to the state interval.

7. A distributed photovoltaic power prediction set fault diagnosis model according to claim 6, characterized in that: The methods for preprocessing the final prediction data include: The final prediction data based on the time domain signal is subjected to noise reduction and smoothing processing by wavelet transform and normalization processing adopted in formula 17: Among them, ψ a,b(τ) represents the mother wavelet; a represents the scale factor; b represents the translation factor; represents the distributed photovoltaic power output forecast data reconstructed after wavelet denoising; The scale difference caused by the noise in some time periods is eliminated by normalization processing through formula 18: in, They respectively represent the maximum power output value and the minimum power output value of the distributed photovoltaic power output prediction data reconstructed after wavelet denoising; Methods for extracting single-dimensional features, multi-dimensional features, and DTW distances of the preprocessed final prediction data include: The single-dimensional features of the distributed photovoltaic power output prediction data are extracted by using the five factors of maximum power, average power, power change rate, peak time, and bending degree through formulas 19 to 22 respectively: The probability density function and marginal probability density are used to measure two types of single-dimensional features through formula 23, and the mutual information elements of multi-dimensional features are extracted: Among them, U(X,Y) represents the joint probability density function of two single-dimensional vectors; u(x,y) represents the joint probability density of two single-dimensional vectors; u(x) and u(y) represent the marginal probability densities of two single-dimensional vectors respectively; Formula 24 is used to measure the complexity of extracting multi-dimensional features using the Lyapunov index: Where ρ represents the small disturbance index of distributed photovoltaic power output prediction data; They represent the small disturbance and average disturbance of the distributed photovoltaic power output forecast data respectively; The DTW distances between various types of faults and normal operation of distributed photovoltaic power output prediction data are extracted through formula 25: in, Indicates the DTW distance between various types of faults and normal operation of distributed photovoltaics; ω tτ represents the DTW weight of the τth period; Respectively represent the power output prediction data of distributed photovoltaics during various types of faults and normal operation.

8. A distributed photovoltaic power prediction set fault diagnosis model according to claim 7, characterized in that: According to the extracted single-dimensional features, multi-dimensional features and DTW distance, the method of obtaining the fault type of distributed photovoltaic includes: The maximum and minimum values ​​of the eight elements of each feature are normalized and mapped to the range of [0, 1] to obtain the normalized value l of the eight element features. e =[l1,l2,…,l8]; The normalized values ​​of the eight element features are used for radar chart standardization and are expressed in polar coordinates using formula 26: Among them, θ e Indicates the angle corresponding to the normalized value; According to the eight elements on the radar chart, the polar coordinates can form the corresponding feature vector L e =[L1,L2,…,L8], if the normal template of the current distributed photovoltaic power processing is K e =[K1, K2, …, K8], and the final fault matching is performed from the distance similarity and angle similarity through formula 27, and the Euclidean distance between the fault template and the normal template and the cosine similarity between the fault template and the normal template are obtained respectively: Among them, d LK represents the Euclidean distance between the fault template and the normal template, cosθ LK Represents the cosine similarity between the fault template and the normal template; When the Euclidean distance and cosine similarity between the fault template and the normal template are satisfied at the same time, the fault type of distributed photovoltaic is obtained by formula 28: in, Indicates the fault type of distributed photovoltaic; μ indicates the fault type number; They represent the threshold values ​​of the Euclidean distance and cosine similarity between the fault template and the normal template respectively.

Images