Photovoltaic equipment power generation loss prediction method based on principal component analysis and Resnet network

Abstract

The present invention provides a photovoltaic equipment power generation loss prediction method using a principal component analysis method and a Resnet network, comprising the following steps: S1 data processing: calculating the equipment type index of different photovoltaic equipment types, and obtaining factors affecting the photovoltaic equipment power generation loss; S2 establishment of a prediction and evaluation model: among the factors affecting the photovoltaic equipment power generation loss, a principal component factor is constructed by a principal component analysis method, and the principal component factor is used as a model input parameter, and is constructed according to a Resnet network method, and a main sequence prediction and evaluation model and a detail sequence prediction and evaluation model are respectively constructed; S3 acquisition of data one day before the prediction day: obtaining the photovoltaic equipment power generation loss sequence one day before the prediction day, and decomposing it into a main part and a detail part; S4 output of sequence values: predicting using a prediction and evaluation model to obtain a main part prediction value and a detail part prediction value; S5 acquisition of prediction values: reconstructing the two prediction values ​​using wavelets to obtain a prediction value of the photovoltaic equipment power generation loss.

Description

Technical Field

[0001] The invention belongs to the technical field of photovoltaic equipment power generation loss prediction, and in particular relates to a photovoltaic equipment power generation loss prediction method based on a principal component analysis method and a Resnet network. Background Art

[0002] Solar photovoltaic power generation technology uses the photoelectric effect of semiconductor materials to directly convert solar energy into electrical energy. Photovoltaic power generation has the advantages of not consuming fuel, not being restricted by geographical location, flexible in scale, pollution-free, safe and reliable, and simple to maintain. Grid-connected photovoltaic power generation systems are the mainstream trend of photovoltaic power generation systems. Large-scale photovoltaic grid-connected power generation is an effective way to make full use of solar energy. At present, large-scale photovoltaic grid-connected systems have been widely used.

[0003] With the extensive application of photovoltaic grid-connected systems, the number of photovoltaic stations has increased rapidly. New energy companies are paying more and more attention to the economic benefits and operation and maintenance costs generated by photovoltaic stations. The core of effective control of economic benefits and operation and maintenance costs lies in whether the power generation benefits of photovoltaic equipment can be maximized and the equipment operation and maintenance costs can be minimized. The core of solving the problem lies in effectively identifying the power generation loss of photovoltaic equipment and the health status of the equipment, ensuring the healthy operation of photovoltaic equipment as much as possible, and maximizing the power generation benefits. Based on the above problems, it is urgent to propose a method to effectively predict the power generation loss of photovoltaic equipment, and at the same time reverse the health status of the equipment, accurately locate abnormal equipment, and provide technical support for enterprises to improve the economic benefits of photovoltaic stations and control operation and maintenance costs.

[0004] The traditional prediction of photovoltaic equipment power loss is mainly divided into mathematical statistics prediction method and artificial intelligence prediction method, and the equipment health status identification is mainly based on the discrete rate, deviation rate, junction box temperature and inverter conversion efficiency of photovoltaic branch current and voltage. At present, there are few research methods that combine photovoltaic power loss with equipment health status identification, and the existing research methods have their inherent advantages and disadvantages. There is no recognized ideal intelligent algorithm to solve the above problems. Therefore, combining various intelligent algorithms to form a hybrid algorithm is an effective attempt to improve the prediction of photovoltaic power loss and identify the health status of equipment.

[0005] Due to the large number of photovoltaic equipment, from box-type transformers, inverters to combiner boxes, it is difficult to manually identify the power loss of equipment and locate abnormal equipment, which makes it difficult for new energy companies to control the economic benefits and operation and maintenance costs of photovoltaic stations. At the same time, long-term power loss and equipment abnormalities of photovoltaic equipment are likely to cause safety risks. Summary of the invention

[0006] The present invention provides a photovoltaic equipment power generation loss prediction method based on a principal component analysis method and a Resnet network, in order to find a method for improving the prediction accuracy of photovoltaic equipment power generation loss and the identification of equipment health status.

[0007] In order to achieve the above technical effects, the present invention is implemented through the following technical solutions.

[0008] The photovoltaic equipment power generation loss prediction method based on principal component analysis method and Resnet network includes the following steps:

[0009] S1 data processing: calculate the equipment type index of different types of photovoltaic equipment and obtain the factors affecting the power generation loss of photovoltaic equipment;

[0010] S2 Establishment of prediction and evaluation model: Among the factors affecting the power generation loss of photovoltaic equipment, the principal component factors are constructed by principal component analysis, and the principal component factors are used as model input parameters. According to the Resnet network method, a main sequence prediction and evaluation model and a detail sequence prediction and evaluation model are constructed respectively;

[0011] S3 Acquisition of data one day before the forecast day: Acquire the power generation loss sequence of photovoltaic equipment one day before the forecast day, and use the ensemble empirical mode decomposition method to decompose the power generation loss sequence of photovoltaic equipment one day before the forecast day into the main part and the detail part;

[0012] Output of S4 sequence value: For the main part and the detail part, use the corresponding prediction evaluation model to predict, and obtain the prediction value of the main part and the prediction value of the detail part;

[0013] S5 Obtaining the prediction value: The prediction value of the main part and the prediction value of the detail part are reconstructed by wavelet to obtain the prediction value of the power generation loss of the photovoltaic equipment, and then obtain the prediction value of the power generation loss sequence of the photovoltaic equipment on the predicted day.

[0014] In this technical solution, the principal component analysis method and the Resnet network are combined at the same time, and thus the accuracy is higher than other prediction methods. The cumulative contribution of variables is calculated by the principal component analysis method, which can solve the problem of variable response information overlap and obtain the best variable subset. Compared with general neural network algorithms, the Resnet neural network algorithm can effectively solve the problem of variable information loss caused by gradient disappearance through residual iteration, and avoid the phenomenon of model overfitting caused by increasing the number of network layers.

[0015] In this technical solution, due to the discontinuity of the photovoltaic power generation loss value, the photovoltaic power generation loss sequence fluctuates greatly. After the photovoltaic power generation loss sequence is decomposed into a main sequence and a detail sequence by ensemble empirical mode decomposition, the detail sequence can reflect the details of the photovoltaic power generation loss sequence. Splitting a single sequence into a main sequence and a detail sequence can effectively distinguish the changes in data characteristics caused by differences in data fluctuation amplitudes, and improve the accuracy of model prediction.

[0016] In this technical solution, the loss index in the power transmission process, the ensemble empirical mode decomposition method and the prediction model based on principal component analysis and Resnet network are integrated to propose a power generation loss prediction and processing method for power generation equipment suitable for photovoltaic systems. It can effectively predict the photovoltaic system and reduce the impact of the photovoltaic system on the power grid.

[0017] Preferably, in the data processing step S1, the device type index of different photovoltaic device types is calculated by using a normalization method, according to the size of power generation loss data of different photovoltaic devices, mapping the device type to a device type index to reflect the impact of the device type on photovoltaic power generation loss.

[0018] In this technical solution, the purpose of using the normalization method is to eliminate the differences between different indicator data due to different dimensional dimensions, lay the foundation for the calculation of equipment type index and normalization of model parameters, and improve the data quality of the input model.

[0019] Preferably, the calculation of the device type index of each device type is specifically as follows:

[0020] Grouping PV power loss curves according to equipment type;

[0021] Calculate the point-by-point average of each set of photovoltaic circuit loss curves to obtain the average power generation loss curve for each type of equipment;

[0022] The average power generation loss curve of each equipment type is used to calculate the equipment type index of each equipment type using the normalization method.

[0023] In this technical solution, a curve is first drawn and then normalized through the curve, so that the data is more simplified, the subsequent calculations are less, and the process is simple, fast, direct and clear.

[0024] Preferably, in the data processing of step S1, obtaining the internal factor data affecting the power generation loss of the photovoltaic equipment is specifically: obtaining the junction box temperature of the photovoltaic equipment to be predicted, the total output power of the junction box, the inverter conversion efficiency, the inverter temperature, the total inverter input power, the inverter output power, the box transformer winding temperature and the photovoltaic module efficiency as the internal factor data.

[0025] In the present technical solution, the above data is used as the internal factor, which is the data directly generated by the internal operation of the photovoltaic device itself, and thus directly reflects the internal operation of the photovoltaic device, and is the internal factor itself.

[0026] Preferably, the establishment of the prediction and evaluation model in step S2 is specifically as follows:

[0027]

[0028] Among them, A in the photovoltaic equipment power generation loss prediction model 1 Refers to the main part of the day before the forecast date, B 1 is the device type index of the day before the forecast, B 2 is the equipment type index on the forecast day, C 1 ,C 2 They are the influencing factor data of the day before and on the forecast day, and D is the details of the day before the forecast day. is the predicted value of the main part of the equipment power generation loss sequence to be predicted on the day. It is the predicted value of the detailed part of the power generation loss series of the equipment to be tested on the day;

[0029] It is a photovoltaic equipment power generation loss main sequence prediction and estimation model established based on the main part of the day before the forecast date, the equipment type index of the day before the forecast date, the equipment type index of the forecast date, the core influencing factors of the day before the forecast date, and the core influencing factors of the forecast date;

[0030] and The photovoltaic equipment power generation loss detail sequence prediction and estimation models are established based on the detail sequence of the day before the prediction day, the equipment type index of the day before the prediction day, the equipment type index of the prediction day, other influencing factors of the day before the prediction day, and other influencing factors of the prediction day.

[0031] In this technical solution, two forecasting and evaluation models are formed, and then comprehensive forecasting is carried out from multiple aspects. At the same time, factor data of the day before the forecast day is collected. The purpose is to rely on the month-on-month development pattern of historical data for data forecasting, which can effectively improve the quality and accuracy of model forecasting.

[0032] In this technical solution, the original data will be split by the algorithm to obtain a main sequence and three detail sequences. The main sequence is represented by A 1 Instead, the three detail sequences are represented by D 1 ,D 2 ,D 3 Alternative representation.

[0033] Preferably, in the establishment of the prediction and evaluation model in step S2, the principal component analysis method is used to construct the principal component factors as follows: the orthogonal transformation method of the principal component analysis is used to convert the variables that may be correlated in the factors affecting the power generation loss of the photovoltaic equipment into linearly uncorrelated variables, and the converted variables will be used as the principal component factors of the power generation loss of the photovoltaic equipment.

[0034] In this technical solution, there is a certain correlation between the factors affecting the power generation loss of photovoltaic equipment, and such factors will cause information overlap when reflecting the target object information. Therefore, it is necessary to remove the closely related variables through principal component analysis, and establish new variables such as: total output power of the junction box, inverter conversion efficiency, total inverter input power, inverter output power, box transformer winding temperature, photovoltaic module efficiency, etc., so that the target object information reflected by the new variables does not overlap.

[0035] Preferably, in the acquisition of the data one day before the forecast day in step S3, the method based on the ensemble empirical mode decomposition is specifically as follows: finding all the maximum points of the power generation loss data sequence X(t) and fitting them with a cubic spline interpolation function to form an upper envelope of the original data;

[0036] Find all the minimum points of the power generation loss data series, and fit all the minimum points through the cubic spline interpolation function to form the lower envelope of the data;

[0037] The mean of the upper envelope and the lower envelope is denoted as ml. The original data sequence X(t) is subtracted from the mean envelope ml to obtain a new data sequence hl:

[0038] X(t)-ml=hl

[0039] Using the ensemble empirical mode decomposition method, the photovoltaic equipment power generation loss sequence P predicted the day before is decomposed into a main sequence A 1 and three detail sequences D 1 , D 2 and D 3 , A 1 Refers to the low-frequency sequence after the collective empirical mode decomposition, that is, the main part of the equipment power generation loss sequence predicted one day before, D 1 , D 2 and D 3 They are the three high-frequency sequences after the collective empirical mode decomposition, i.e., the detailed parts of the equipment power generation loss sequence one day before the forecast;

[0040] In this technical solution, a set simulation static decomposition is adopted, which mainly decomposes the fluctuations and trends of different scales in the signal step by step to form data sequences with different characteristic scales, namely, intrinsic mode function (IMF) components. The lowest frequency IMF component represents the time series of the overall trend or mean of the original signal.

[0041] In this technical solution, the intrinsic mode components obtained by the ensemble empirical mode decomposition usually concentrate the most significant and important information in the original sequence, and different intrinsic mode components contain different time scales. The ensemble empirical mode decomposition can be used to extract modal components with specific characteristics from complex sequences, and the simulation sequence is decomposed to obtain 1 low-frequency sequence and 3 high-frequency sequences. The frequencies corresponding to the split sequences are different from high to low scales.

[0042] Preferably, the output of the sequence value in step S4 is specifically: using the Resnet18 network to iterate the residual layer multiple times to obtain a minimum residual value, that is, a result with the minimum error between the predicted value and the true value. The smaller the residual value, the more accurate the model prediction result.

[0043] Preferably, the main part prediction value is obtained as follows: the output power of the photovoltaic equipment one day before the prediction day, the output power of the photovoltaic equipment on the prediction day, the equipment type index one day before the prediction day and the equipment type index on the prediction day are used as inputs of the Resnet network model, and the main part of the power generation loss sequence corresponding to the prediction day is output to form the main part prediction value.

[0044] In this technical solution, the core factors such as the power, index and type of the equipment itself form the main body, which in turn determines the accuracy of the preliminary prediction.

[0045] Preferably, the details are as follows: using the combiner box temperature, component efficiency, inverter temperature, inverter conversion efficiency, and transformer winding temperature of the day before the prediction, the combiner box temperature, component efficiency, inverter temperature, inverter conversion efficiency, and transformer winding temperature of the day before the prediction, and the equipment type index of the day before the prediction and the equipment type index of the day as inputs of the Resnet network model, and outputting the details of the corresponding power generation loss sequence of the day respectively.

[0046] In this technical solution, various small parameters in operation are used as details, and the equipment type index of the day is used as both the main body and the details, which fully reflects the principle of prudence and improves the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flow chart of a method for predicting power generation loss of photovoltaic equipment using a principal component analysis method and a Resnet network provided by the present invention;

[0048] Figure 2 The flowchart of Embodiment 3 provided by the present invention;

[0049] Figure 3A flow chart of calculating the device type index of each device type in Embodiment 3 provided by the present invention;

[0050] Figure 4 This is a schematic diagram of the Resnet network in Example 3 provided by the present invention. DETAILED DESCRIPTION

[0051] The present invention is described in detail below in conjunction with the various embodiments shown in the accompanying drawings, but it should be noted that these embodiments are not limitations of the present invention, and any equivalent transformations or substitutions in functions, methods, or structures made by ordinary technicians in the field based on these embodiments are all within the scope of protection of the present invention.

[0052] In the description of the present embodiment, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the invention.

[0053] In addition, the terms "first", "second", "third", etc. are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more. The terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood through specific circumstances.

[0054] Example 1

[0055] The photovoltaic equipment power generation loss prediction method based on principal component analysis method and Resnet network includes the following steps:

[0056] S1 data processing: calculate the equipment type index of different types of photovoltaic equipment and obtain the internal factor data that affects the power generation loss of photovoltaic equipment;

[0057] S2 Establishment of prediction and evaluation model: Among the factors affecting the power generation loss of photovoltaic equipment, the principal component factors are constructed by principal component analysis, and the principal component factors are used as model input parameters. According to the Resnet network method, a main sequence prediction and evaluation model and a detail sequence prediction and evaluation model are constructed respectively;

[0058] S3 Acquisition of data one day before the forecast day: Acquire the power generation loss sequence of photovoltaic equipment one day before the forecast day, and use the ensemble empirical mode decomposition method to decompose the power generation loss sequence of photovoltaic equipment one day before the forecast day into the main part and the detail part;

[0059] Output of S4 sequence value: For the main part and the detail part, use the corresponding prediction evaluation model to predict, and obtain the prediction value of the main part and the prediction value of the detail part;

[0060] S5 Obtaining the prediction value: The prediction value of the main part and the prediction value of the detail part are reconstructed by wavelet to obtain the prediction value of the power generation loss of the photovoltaic equipment, and then obtain the prediction value of the power generation loss sequence of the photovoltaic equipment on the predicted day.

[0061] In this embodiment, the wavelet reconstruction mainly adds the main part and the detail part point by point to obtain the predicted value of the photovoltaic equipment power generation loss sequence on the predicted day. Specifically, the predicted value of the main sequence and the detail sequence is reconstructed by wavelet to obtain the predicted value of the photovoltaic equipment power generation loss. The wavelet reconstruction mainly adds the main part and the detail part point by point to obtain the predicted value of the photovoltaic equipment power generation loss sequence on the predicted day.

[0062] The wavelet reconstruction formula is as follows:

[0063]

[0064] In this embodiment, the principal component analysis method and the Resnet network are combined at the same time, and thus the accuracy is higher than other prediction methods. The cumulative contribution of variables is calculated by the principal component analysis method, which can solve the problem of variable response information overlap and obtain the best variable subset. Compared with general neural network algorithms, the Resnet neural network algorithm can effectively solve the problem of variable information loss caused by gradient disappearance through residual iteration, and avoid the phenomenon of model overfitting caused by increasing the number of network layers.

[0065] In this embodiment, due to the discontinuity of the photovoltaic power generation loss value, the photovoltaic power generation loss sequence fluctuates greatly. After the photovoltaic power generation loss sequence is decomposed into a main sequence and a detail sequence by ensemble empirical mode decomposition, the detail sequence can reflect the details of the photovoltaic power generation loss sequence. Splitting a single sequence into a main sequence and a detail sequence can effectively distinguish the changes in data characteristics caused by differences in data fluctuation amplitudes, thereby improving the accuracy of model prediction.

[0066] In this embodiment, the loss index in the power transmission process, the ensemble empirical mode decomposition method and the prediction model based on principal component analysis and Resnet network are integrated to propose a method for predicting the power loss of power generation equipment suitable for photovoltaic systems. The method can effectively predict the photovoltaic system and reduce the impact of the photovoltaic system on the power grid.

[0067] Example 2

[0068] In this embodiment, the steps in Embodiment 1 are mainly refined.

[0069] Preferably, in the data processing step S1, the device type index of different photovoltaic device types is calculated by using a normalization method, according to the size of power generation loss data of different photovoltaic devices, mapping the device type to a device type index to reflect the impact of the device type on photovoltaic power generation loss.

[0070] In this embodiment, the purpose of using the normalization method is to eliminate the differences between different indicator data due to different dimensional dimensions, lay the foundation for equipment type index calculation and model parameter normalization, and improve the data quality of the input model.

[0071] Preferably, the calculation of the device type index of each device type is specifically as follows:

[0072] Grouping PV power loss curves according to equipment type;

[0073] Calculate the point-by-point average of each set of photovoltaic circuit loss curves to obtain the average power generation loss curve for each type of equipment;

[0074] The average power generation loss curve of each equipment type is used to calculate the equipment type index of each equipment type using the normalization method.

[0075] In this embodiment, a curve is first drawn and then normalized using the curve, so that the data is more simplified, the subsequent calculations are less, and the process is simple, fast, direct and clear.

[0076] Preferably, in the data processing of step S1, obtaining the internal factor data affecting the power generation loss of the photovoltaic equipment is specifically: obtaining the junction box temperature of the photovoltaic equipment to be predicted, the total output power of the junction box, the inverter conversion efficiency, the inverter temperature, the total inverter input power, the inverter output power, the box transformer winding temperature and the photovoltaic module efficiency as the internal factor data.

[0077] In the present technical solution, the above data is used as the internal factor, which is the data directly generated by the internal operation of the photovoltaic device itself, and thus directly reflects the internal operation of the photovoltaic device, and is the internal factor itself.

[0078] Preferably, the establishment of the prediction and evaluation model in step S2 is specifically as follows:

[0079]

[0080] Among them, A in the photovoltaic equipment power generation loss prediction model 1 Refers to the main part of the day before the forecast date, B 1 is the device type index of the day before the forecast, B 2 is the equipment type index on the forecast day, C 1 ,C 2 They are the influencing factor data of the day before and on the forecast day, and D is the details of the day before the forecast day. is the predicted value of the main part of the equipment power generation loss sequence to be predicted on the day. It is the predicted value of the detailed part of the power generation loss series of the equipment to be tested on the day;

[0081] It is a photovoltaic equipment power generation loss main sequence prediction and estimation model established based on the main part of the day before the forecast date, the equipment type index of the day before the forecast date, the equipment type index of the forecast date, the influencing factors of the day before the forecast date, and the influencing factors of the forecast date;

[0082] and The photovoltaic equipment power generation loss detail sequence prediction and estimation models are established based on the detail sequence of the day before the prediction day, the equipment type index of the day before the prediction day, the equipment type index of the prediction day, the influencing factors of the day before the prediction day, and the influencing factors of the prediction day.

[0083] In this embodiment, two prediction and evaluation models are formed, and then comprehensive prediction is carried out from multiple aspects. At the same time, factor data of the day before the prediction is collected, with the purpose of relying on the month-on-month development law of historical data for data prediction, which can effectively improve the quality and accuracy of model prediction. In this embodiment, the original data will be split by the algorithm to obtain a main sequence and three detail sequences. The main sequence is represented by A 1 Instead, the three detail sequences are represented by D 1 ,D 2 ,D 3 Alternative representation.

[0084] Preferably, in the establishment of the prediction and evaluation model in step S2, the principal component analysis method is used to construct the principal component factors as follows: the orthogonal transformation method of the principal component analysis is used to convert the variables that may be correlated in the factors affecting the power generation loss of the photovoltaic equipment into linearly uncorrelated variables, and the converted variables will be used as the principal component factors of the power generation loss of the photovoltaic equipment.

[0085] In this embodiment, there is a certain correlation between the factors affecting the power generation loss of the PV equipment, and such factors will cause information overlap when reflecting the target object information. Therefore, it is necessary to remove the closely related variables through principal component analysis, and establish new variables such as: total output power of the junction box, inverter conversion efficiency, total inverter input power, inverter output power, box transformer winding temperature, photovoltaic module efficiency, etc., so that the target object information reflected by the new variables does not overlap.

[0086] Preferably, in the acquisition of the data one day before the forecast day in step S3, the method based on the ensemble empirical mode decomposition is specifically as follows: finding all the maximum points of the power generation loss data sequence X(t) and fitting them with a cubic spline interpolation function to form an upper envelope of the original data;

[0087] Find all the minimum points of the power generation loss data series, and fit all the minimum points through the cubic spline interpolation function to form the lower envelope of the data;

[0088] The mean of the upper envelope and the lower envelope is denoted as ml. The original data sequence X(t) is subtracted from the mean envelope ml to obtain a new data sequence hl:

[0089] X(t)-ml=hl

[0090] Using the ensemble empirical mode decomposition method, the photovoltaic equipment power generation loss sequence P predicted the day before is decomposed into a main sequence A 1 and three detail sequences D 1 , D 2 and D 3 .

[0091] In this embodiment, a set simulation static decomposition is adopted, which mainly decomposes the fluctuations and trends of different scales in the signal step by step to form data sequences with different characteristic scales, namely, intrinsic mode function (IMF) components. The lowest frequency IMF component represents the time series of the overall trend or mean of the original signal.

[0092] In this embodiment, the intrinsic mode components obtained by the ensemble empirical mode decomposition usually concentrate the most significant and important information in the original sequence, and different intrinsic mode components contain different time scales. The ensemble empirical mode decomposition can be used to extract modal components with specific characteristics from complex sequences, and the simulation sequence is decomposed to obtain 1 low-frequency sequence and 3 high-frequency sequences. The frequencies corresponding to the decomposed sequences are different from high to low scales.

[0093] Preferably, the output of the sequence value in step S4 is specifically: using the Resnet18 network to iterate the residual layer multiple times to obtain a minimum residual value, that is, a result with the minimum error between the predicted value and the true value, and using this result as the output result.

[0094] The specific parameters of the ResNet18 network are shown in Table 1:

[0095] Table 1 ResNet18 network parameters

[0096]

[0097]

[0098] Preferably, the main part prediction value is obtained as follows: the output power of the photovoltaic equipment one day before the prediction day, the output power of the photovoltaic equipment on the prediction day, the equipment type index one day before the prediction day and the equipment type index on the prediction day are used as inputs of the Resnet network model, and the main part of the power generation loss sequence corresponding to the prediction day is output to form the main part prediction value.

[0099] In this embodiment, the core factors such as the power, index and type of the device itself form the main body, which in turn determines the accuracy of the preliminary prediction.

[0100] Preferably, the details are as follows: using the combiner box temperature, component efficiency, inverter temperature, inverter conversion efficiency, and transformer winding temperature of the day before the prediction, the combiner box temperature, component efficiency, inverter temperature, inverter conversion efficiency, and transformer winding temperature of the day before the prediction, and the equipment type index of the day before the prediction and the equipment type index of the day as inputs of the Resnet network model, and outputting the details of the corresponding power generation loss sequence of the day respectively.

[0101] In this embodiment, various small parameters in operation are used as details, and the equipment type index of the day is used as both the main body and the details, which fully reflects the principle of prudence and improves the accuracy of the prediction.

[0102] Example 3

[0103] This embodiment is introduced with respect to a specific solution.

[0104] In this embodiment, the following specific steps are included:

[0105] Construction of photovoltaic equipment power generation loss prediction model: After determining the main component factors affecting the power generation loss of photovoltaic equipment, the present invention constructs data features and uses them as model input parameters, and then uses empirical mode decomposition, Resnet neural network and wavelet reconstruction technology to construct a photovoltaic equipment power generation loss prediction model;

[0106] The ResNet neural network protects the integrity of information by directly transferring input information to the output. The entire network only needs to learn the difference between input and output, simplifying the learning objectives and difficulty.

[0107] The steps specifically include the following steps:

[0108] Data preprocessing, normalizing the influencing factors of the input model;

[0109] Principal component analysis: Principal component analysis is mainly used to analyze the factors affecting the power generation loss of photovoltaic equipment. Through the orthogonal transformation method, the variables that may be correlated in the factors affecting the power generation loss of photovoltaic equipment are converted into linearly uncorrelated variables. The converted variables will be used as the principal component factors of the power generation loss of photovoltaic equipment to achieve dimensionality reduction;

[0110] Data feature construction: calculate the indicators composed of several influencing factors, and select the main factors based on the calculation results;

[0111] Empirical mode analysis: The power generation loss sequence one day before the forecast day is decomposed into a main sequence and three detail sequences using the empirical mode method;

[0112] The main sequence, detail sequence, equipment type index, and equipment power generation loss influencing factors are respectively input into the photovoltaic equipment power generation loss prediction model, and the ResNet neural network is used to predict the main part and detail part of the equipment power generation loss sequence.

[0113] The predicted values ​​of the main part and the detailed part of the photovoltaic equipment power generation loss sequence are combined by using wavelet reconstruction technology;

[0114] Compare the predicted values ​​with the actual data, calculate the error index and perform error analysis;

[0115] The photovoltaic equipment power generation loss prediction model is updated according to the results of the error analysis, and the health status of different equipment is identified based on the power loss prediction value, and different treatment plans are adopted accordingly.

[0116] In a preferred embodiment of the present invention, the data of the several influencing factors include: junction box temperature, total output power of the junction box, inverter conversion efficiency, inverter temperature, total inverter input power, inverter output power, box transformer winding temperature, photovoltaic module efficiency, etc.

[0117] In the above method, preferably, calculating the device type index in the process of photovoltaic device power generation transmission includes:

[0118] The specific formula for calculating the average power generation loss value of photovoltaic box transformers, photovoltaic inverters, photovoltaic combiner boxes and other equipment is as follows;

[0119] Daily average power generation loss = ((sunlight duration * total power of photovoltaic array) - (sunlight duration * total power of photovoltaic array * power generation efficiency)) / n (n: number of photovoltaic arrays)

[0120] The ratio relationship between these types of average power generation losses is then converted into an equipment type index.

[0121] Equipment type index = average power generation loss value of each equipment / total power generation loss value

[0122] In a preferred embodiment of the present invention, the decomposition of the equipment power generation loss sequence of the day before the prediction day is achieved based on the ensemble empirical mode decomposition method, and the prediction merging of the main part and the detail part of the equipment power generation loss sequence of the prediction day is achieved based on wavelet reconstruction.

[0123] In the above method, preferably, the preset photovoltaic equipment power generation loss prediction model is:

[0124]

[0125] Among them, A in the photovoltaic equipment power generation loss prediction model 1 Refers to the main part of the day before the forecast date, B 1 is the device type index of the day before the forecast, B 2 is the equipment type index on the forecast day, C 1 ,C 2 They are the influencing factor data of the day before and on the forecast day, and D is the details of the day before the forecast day. is the predicted value of the main part of the equipment power generation loss sequence to be predicted on the day. It is the predicted value of the detailed part of the power generation loss series of the equipment to be tested on the day;

[0126] It is a photovoltaic equipment power generation loss main sequence prediction and estimation model established based on the main part of the day before the forecast date, the equipment type index of the day before the forecast date, the equipment type index of the forecast day, the influencing factors of the day before the forecast date, and the influencing factors of the forecast day.

[0127] and The photovoltaic equipment power generation loss detail sequence prediction and estimation models are established based on the detail sequence of the day before the prediction day, the equipment type index of the day before the prediction day, the equipment type index of the prediction day, the influencing factors of the day before the prediction day, and the influencing factors of the prediction day.

[0128] In the above method, preferably, the photovoltaic equipment power generation loss main series prediction and estimation model and the photovoltaic equipment power generation loss detail series prediction and estimation model are both established based on the Resnet network method.

[0129] In the above method, preferably, the subject and detail sequence are split using empirical mode decomposition, the subject and detail are predicted using Resnet neural network prediction, and the subject and detail are merged using wavelet reconstruction technology.

[0130] In the above method, preferably, the daily photovoltaic equipment power generation loss sequence, main sequence and detail sequence all refer to the power generation loss data of the power generation equipment from 6 to 20 o'clock every day, and the time resolution is 15 minutes.

[0131] In the above method, preferably, in the photovoltaic equipment power generation loss main sequence prediction and estimation model, the main part of the hourly equipment power generation loss sequence of the photovoltaic equipment power generation loss of the previous day, the influencing factors of the previous day, the influencing factors of the current day, the equipment type index of the previous day and the equipment type index of the current day are first taken as the input of the Resnet network prediction model, and the output corresponds to the main equipment power generation loss sequence of the current day.

[0132] The above method preferably comprises, in the photovoltaic equipment power generation loss detail sequence prediction and estimation model, firstly taking the detail part of the hourly equipment power generation loss sequence of the photovoltaic equipment power generation loss of the previous day, the equipment type index of the previous day and the equipment type index of the current day as the input of the Resnet network prediction model, and the output respectively corresponds to the detail equipment power generation loss sequence of the current day.

[0133] It can be seen from the above technical solution that the present application discloses a mathematical model method for predicting power generation loss of photovoltaic power generation equipment.

[0134] In the prediction stage, this method needs to know the temperature of the PV equipment combiner box, the total output power of the combiner box, the inverter conversion efficiency, the inverter temperature, the total inverter input power, the inverter output power, the box transformer winding temperature, and the PV module efficiency on the previous day.

[0135] The purpose of obtaining the relevant data of the previous day is to use the data as the input parameter of the photovoltaic power generation equipment power loss prediction model to predict the power generation loss value of the predicted day.

[0136] After obtaining these conditions, the variables are input into the prediction model for prediction, so that the power generation loss data of the photovoltaic equipment on the prediction day can be obtained.

[0137] Furthermore, in this embodiment, the specific process is as follows.

[0138] See also Figure 2 A schematic flow chart of a method for predicting power generation loss of photovoltaic power generation equipment disclosed in an embodiment of the present invention is shown.

[0139] The main steps in this embodiment are:

[0140] 1. Analysis of factors affecting the power generation loss of photovoltaic equipment: The power generation loss of photovoltaic equipment is mainly affected by two factors: the external environment and the internal equipment health status. External environmental factors include solar radiation, the tilt angle of photovoltaic modules, ambient temperature, panel dust accumulation, snow shadow area, line loss, etc. Internal equipment health status factors include combiner box temperature, combiner box total output power, inverter conversion efficiency, inverter temperature, inverter total input power, inverter output power, box transformer winding temperature, photovoltaic module efficiency, module matching loss, etc.

[0141] 2. Analysis of key influencing factors: This invention mainly uses principal component analysis to summarize and extract the introduced multi-faceted variables and complex factors. The purpose of principal component analysis is to use the least number of principal components to explain the maximum amount of variance and avoid multicollinearity caused by too many factors. The main factors finally retained by principal component analysis include junction box temperature, junction box total output power, inverter conversion efficiency, inverter temperature, inverter total input power, inverter output power, box transformer winding temperature, photovoltaic module efficiency, etc.

[0142] 3. Construction of photovoltaic equipment power generation loss prediction model: After determining the main component factors affecting the power generation loss of photovoltaic equipment, the present invention constructs data features and uses them as model input parameters, and then uses empirical mode decomposition, Resnet neural network and wavelet reconstruction technology to construct a photovoltaic equipment power generation loss prediction model;

[0143] The ResNet neural network protects the integrity of information by directly transferring input information to the output. The entire network only needs to learn the difference between input and output, simplifying the learning objectives and difficulty.

[0144] The model building steps are as follows:

[0145] Data preprocessing: normalizing the influencing factors of the input model to avoid data differences caused by indicator dimensions;

[0146] Principal component analysis to achieve dimensionality reduction;

[0147] Data feature construction, calculation of several indicators, etc.

[0148] The power generation loss sequence one day before the forecast day is decomposed into a main sequence and three detail sequences using the empirical mode method;

[0149] The main sequence, detail sequence, equipment type index, and equipment power generation loss influencing factors are respectively input into the photovoltaic equipment power generation loss prediction model, and the ResNet neural network is used to predict the main part and detail part of the equipment power generation loss sequence.

[0150] The predicted values ​​of the main part and the detailed part of the photovoltaic equipment power generation loss sequence are combined by using wavelet reconstruction technology;

[0151] Compare the predicted values ​​with the actual data, calculate the error index and perform error analysis;

[0152] The photovoltaic equipment power generation loss prediction model is updated according to the results of the error analysis, and the health status of different equipment is identified based on the power loss prediction value, and different treatment plans are adopted accordingly.

[0153] The method specifically includes:

[0154] S11: Calculate the device type index for different photovoltaic device types.

[0155] Since the equipment type is a noun variable, it cannot be used directly in the model building process. In order to use the equipment type as input for modeling and thus improve the prediction accuracy, the present invention is based on a normalization method. According to the size of the power generation loss data under different equipment types, the equipment type is mapped to an equipment type index to reflect the impact of the equipment type on the photovoltaic power generation loss.

[0156] S12: Obtaining data on internal factors that affect power generation loss of photovoltaic equipment;

[0157] The combiner box temperature of the photovoltaic equipment, the total output power of the combiner box, the inverter conversion efficiency, the inverter temperature, the total inverter input power, the inverter output power, the box transformer winding temperature, the photovoltaic module efficiency, etc. are the main influencing factors for analyzing the power loss during the power transmission process.

[0158] Therefore, when predicting the power generation loss of the photovoltaic equipment on the prediction day, the combiner box temperature, the total output power of the combiner box, the inverter conversion efficiency, the inverter temperature, the total inverter input power, the inverter output power, the box transformer winding temperature, and the photovoltaic module efficiency on the day before the prediction day and the prediction day are considered.

[0159] S13: The photovoltaic equipment power generation loss sequence one day before the forecast day is decomposed into a main part and a detailed part using an ensemble-based empirical mode decomposition method.

[0160] The ensemble empirical mode decomposition method described above performs empirical mode decomposition after superimposing white noise of finite amplitude multiple times in the observed signal, and considers the mean of the ensemble average of each order IMF obtained by the multiple empirical mode decompositions as the real IMF component of the ensemble empirical mode decomposition method. Through this ensemble average method, the various scales of the signal can be clearly separated without any prior knowledge.

[0161] The decomposition method of the ensemble empirical mode decomposition method is based on the following assumptions: ⑴ The data has at least two extreme values, a maximum value and a minimum value; ⑵ The local time domain characteristics of the data are uniquely determined by the time scale between the extreme points; ⑶ If the data has no extreme points but has inflection points, the extreme values ​​can be obtained by differentiating the data once or multiple times, and then the decomposition result can be obtained by integration.

[0162] Decomposition process:

[0163] Find all the maximum points of the power generation loss data sequence X(t) and use the cubic spline interpolation function to fit the upper envelope of the original data;

[0164] Find all the minimum points of the power generation loss data series, and fit all the minimum points through the cubic spline interpolation function to form the lower envelope of the data;

[0165] The mean of the upper envelope and the lower envelope is denoted as ml. The original data sequence X(t) is subtracted from the mean envelope ml to obtain a new data sequence h:

[0166] X(t)-ml=hl

[0167] Using the ensemble empirical mode decomposition method, the photovoltaic equipment power generation loss sequence P predicted the day before is decomposed into a subject A 1 and three detail sequences D 1 , D 2 and D 3 .

[0168] The detailed process is as follows:

[0169] For the original signal x(t), first find all its local maxima and minima, and then use the cubic spline interpolation method to form the upper envelope u 1 (t) and the lower envelope u 2 (t), then the local mean envelope m 1 (t) can be expressed as:

[0170]

[0171] The original signal x(t) minus the local mean envelope m 1 (t), we can get the first vector h 1 , the mathematical expression is:

[0172] h 1 (t) = x(t) - m 1 (t) (1.1.2)

[0173] Such as h 1If (t) does not meet the IMF conditions, it is regarded as a new x(t), and the calculation steps of (1.1.1) and (1.1.2) are repeated. After k repetitions, h that meets the IMF requirements is obtained. 1k (t), the first IMF component

[0174] h 1k (t) = h 1(k-1) (t)-m 1k (t) (1.1.3)

[0175] In actual operation, repeating the above process too much will turn the IMF into a pure frequency modulation signal with constant amplitude, thus losing its practical significance.

[0176] Therefore, the standard deviation SD (generally 0.2-0.3) can be used as the criterion for stopping the screening process. When SD reaches a certain threshold, the screening is stopped. The calculation formula of SD is:

[0177]

[0178] If the first IMF component h 1k (t) = c 1 , then the other remaining amount r 1 (t) can be expressed as:

[0179] r 1 (t) = x(t) - C 1 (1.1.5)

[0180] R 1 (t) Do the same "screening" process as in (1.1.1) to (1.1.5) to obtain C2, C3, ..., until r 1 (t) basically presents a monotonic trend or |r 1 (t)| is very small, then the original signal is reconstructed as:

[0181]

[0182] Add a white noise sequence to the original signal:

[0183] x i (t) = x(t) + n i (t) (1.1.7)

[0184] In the formula, x i (t) is the new signal obtained by adding the ith white noise to the original signal data x(t), n i (t) is white noise.

[0185] According to the empirical mode decomposition method described in equations (1.1.1) to (1.1.2), the signal with white noise added is decomposed into various IMFs. The component C of the IMF response of the i-th component is obtained. ij (t) and the remaining component r i (t).

[0186] Finally, the IMF components of the response obtained by the above decomposition are averaged to obtain the final decomposition result, namely:

[0187]

[0188] In the formula, C j (t) is the jth component IMF obtained, N is the number of white noise sequences, C ij (t) represents the jth component IMF after adding the i-th white noise processing.

[0189] In the example of the present invention, the original photovoltaic equipment power generation loss sequence is not directly used to model the model, but the original photovoltaic equipment power generation loss sequence is decomposed by the ensemble empirical mode decomposition method to obtain the subject A 1 and three detail sequences D 1 , D 2 and D 3 , and then model the main sequence and detail sequence respectively. Due to the discontinuity and intermittency of photovoltaic equipment power generation loss, the photovoltaic equipment power generation loss sequence fluctuates greatly. Through the ensemble empirical mode decomposition method, the photovoltaic equipment power generation loss sequence can be decomposed into a low-frequency sequence (main sequence) and a high-frequency sequence (detail sequence). The high-frequency sequence can reflect the details of the photovoltaic equipment power generation loss sequence.

[0190] S14: For the main part and the detail part of the photovoltaic equipment power generation loss sequence, respectively, a prediction model based on the Resnet network is used to predict the sequence values ​​of each component.

[0191] Resnet is a convolutional neural network proposed by four scholars from Microsoft Research. The residual network is easy to optimize and can improve accuracy by increasing the depth. The internal residual block uses skip connections to alleviate the gradient vanishing problem caused by increasing the depth of the deep neural network.

[0192] The Resnet network will make a reference (X) for each input layer, learn to form a residual function, and greatly deepen the number of network layers. In the residual block in the figure above, it has two layers, as shown in the following expression, where σ represents the nonlinear function ReLU.

[0193]

[0194] Then through a shortcut and the second ReLU, the output y is obtained.

[0195]

[0196] When the input and output dimensions need to be changed (such as changing the number of channels), a linear transformation Ws can be performed on x during the shortcut, as shown below.

[0197]

[0198] By changing the number of channels multiple times, we can eventually obtain a result with the minimum residual value, that is, the minimum error between the predicted value and the true value, so as to achieve the optimal prediction result.

[0199] S15: The predicted values ​​of the main body and detail sequences are reconstructed using wavelet to obtain the predicted value of the power generation loss of the photovoltaic device.

[0200] In the prediction and estimation model of the main sequence of photovoltaic equipment power generation loss, the main part of the hourly equipment power generation loss sequence of the photovoltaic equipment power generation loss of the previous day and the loss index in the power generation transmission process are first taken as the input of the neural network model, and the output corresponds to the hourly equipment power generation loss sequence of the day.

[0201] In the prediction and estimation model of the detailed sequence of photovoltaic equipment power generation loss, the detailed part of the hourly equipment power generation loss sequence of the photovoltaic equipment power generation loss of the previous day and the loss index in the power generation transmission process are first taken as the input of the prediction model based on principal component analysis and Resnet network, and the output corresponds to the hourly equipment power generation loss sequence of the day.

[0202] The main part and detail part of the predicted daily photovoltaic equipment power generation loss are obtained, and wavelet reconstruction is used, that is, the main part and the detail part are added point by point to obtain the predicted value of the predicted daily photovoltaic equipment power generation loss sequence.

[0203] In this embodiment, the power generation loss sequence of the photovoltaic equipment one day before the forecast day is obtained, and is decomposed into a main part and a detail part using an ensemble empirical mode decomposition method. The ensemble empirical mode decomposition mainly decomposes the fluctuations and trends of different scales in the signal step by step to form data sequences with different characteristic scales, namely, intrinsic mode function (IMF) components. The lowest frequency IMF component represents the time series of the overall trend or mean of the original signal.

[0204] It can be seen from the above embodiments that the present application discloses a data analysis method for predicting the power generation loss of photovoltaic system power generation equipment. The method is based on factors such as combiner box temperature, combiner box total output power, inverter conversion efficiency, inverter temperature, inverter total input power, inverter output power, transformer winding temperature, photovoltaic module efficiency, etc., and then uses the ensemble empirical mode decomposition method and the combined prediction model based on principal component analysis and Resnet network prediction model to obtain the predicted value of photovoltaic equipment power generation loss on the prediction day;

[0205] Compared with the prior art, the present invention integrates the loss index in the power transmission process, the ensemble empirical mode decomposition method, and the prediction model based on principal component analysis and Resnet network, and proposes a power generation loss prediction and processing method for power generation equipment suitable for photovoltaic systems. It can effectively predict the photovoltaic system and reduce the impact of the photovoltaic system on the power grid.

[0206] In the example of the present invention, a photovoltaic equipment power generation loss prediction model is constructed by using a principal component analysis method and a Resnet network prediction model.

[0207] Based on the business experience of photovoltaic experts, the model first selects the junction box temperature, the total output power of the junction box, the inverter conversion efficiency, the inverter temperature, the total input power of the inverter, the inverter output power, the box transformer winding temperature, the photovoltaic module efficiency, etc. as the factors affecting the power generation loss of photovoltaic equipment. Then, the principal component analysis method is used to select the useful features such as the total output power of the junction box, the inverter conversion efficiency, the total input power of the inverter, and the inverter output power to calculate the characteristic value of the indicator, and use this as the model input parameter to use the Resnet network for power loss prediction. The shortcut in the Resnet network can effectively avoid the overfitting phenomenon caused by the excessive number of network layers, and effectively improve the accuracy of the prediction of power generation loss of photovoltaic equipment. Then, the health status of different equipment is identified based on the power loss prediction value, providing support for equipment operation and maintenance for photovoltaic station business personnel.

[0208] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. Any equivalent implementation methods or changes that do not deviate from the technical spirit of the present invention should be included in the scope of protection of the present invention.

[0209] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations of the claims that fall within the meaning and scope of equivalents be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0210] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. Photovoltaic equipment power generation loss prediction method based on principal component analysis and Resnet network, It is characterized in that The following steps are involved: S1 data processing: calculate the equipment type index of different types of photovoltaic equipment and obtain the internal factor data that affects the power generation loss of photovoltaic equipment; S2 Establishment of prediction and evaluation model: Among the factors affecting the power generation loss of photovoltaic equipment, the principal component factors are constructed by principal component analysis, and the principal component factors are used as model input parameters. According to the Resnet network method, a main sequence prediction and evaluation model and a detail sequence prediction and evaluation model are constructed respectively; S3 Acquisition of data one day before the forecast day: Acquire the power generation loss sequence of photovoltaic equipment one day before the forecast day, and use the ensemble empirical mode decomposition method to decompose the power generation loss sequence of photovoltaic equipment one day before the forecast day into the main part and the detail part; Output of S4 sequence value: For the main part and the detail part, use the corresponding prediction evaluation model to predict, and obtain the prediction value of the main part and the prediction value of the detail part; S5 Obtaining the prediction value: The prediction value of the main part and the prediction value of the detail part are reconstructed by wavelet to obtain the prediction value of the power generation loss of the photovoltaic equipment, and then obtain the prediction value of the power generation loss sequence of the photovoltaic equipment on the predicted day.

2. According to the principal component analysis method and Resnet network of claim 1, It is characterized in that In the data processing step S1, the device type index of different photovoltaic device types is calculated specifically by using a normalization method, according to the size of the power generation loss data of different photovoltaic devices, mapping the device type to a device type index to reflect the impact of the device type on the photovoltaic power generation loss.

3. The photovoltaic equipment power generation loss prediction method based on the principal component analysis method and Resnet network according to claim 2, It is characterized in that The device type index of each device type is calculated as follows: Grouping PV power loss curves according to equipment type; Calculate the point-by-point average of each set of photovoltaic circuit loss curves to obtain the average power generation loss curve for each type of equipment; The average power generation loss curve of each equipment type is used to calculate the equipment type index of each equipment type using the normalization method.

4. The photovoltaic equipment power generation loss prediction method based on the principal component analysis method and Resnet network according to claim 1, It is characterized in that In the data processing of step S1, the internal factor data affecting the power generation loss of the photovoltaic equipment is obtained specifically by obtaining the junction box temperature of the photovoltaic equipment to be predicted, the total output power of the junction box, the inverter conversion efficiency, the inverter temperature, the total input power of the inverter, the inverter output power, the box transformer winding temperature and the photovoltaic module efficiency as the internal factor data.

5. The photovoltaic equipment power generation loss prediction method based on the principal component analysis method and Resnet network according to claim 1, It is characterized in that The establishment of the prediction and evaluation model in step S2 is specifically as follows: Among them, A in the photovoltaic equipment power generation loss prediction model 1 Refers to the main part of the day before the forecast date, B 1 is the device type index of the day before the forecast, B 2 is the equipment type index on the forecast day, C 1 ,C 2 They are the influencing factor data of the day before and on the forecast day, and D is the details of the day before the forecast day. is the predicted value of the main part of the equipment power generation loss sequence to be predicted on the day. It is the predicted value of the detailed part of the daily equipment power generation loss series; It is a photovoltaic equipment power generation loss main series prediction and estimation model established based on the main part of the day before the forecast date, the equipment type index of the day before the forecast date, the equipment type index of the forecast date, the influencing factors of the day before the forecast date, and the influencing factors of the forecast date; and The photovoltaic equipment power generation loss detail sequence prediction and estimation models are established based on the detail sequence of the day before the prediction day, the equipment type index of the day before the prediction day, the equipment type index of the prediction day, the influencing factors of the day before the prediction day, and the influencing factors of the prediction day.

6. The photovoltaic equipment power generation loss prediction method based on the principal component analysis method and Resnet network according to claim 5, It is characterized in that In the establishment of the prediction and evaluation model in step S2, the principal component analysis method is used to construct the principal component factors specifically as follows: the orthogonal transformation method of the principal component analysis is used to convert the variables that may be correlated in the factors affecting the power generation loss of the photovoltaic equipment into linearly uncorrelated variables, and the converted variables will be used as the principal component factors of the power generation loss of the photovoltaic equipment.

7. The photovoltaic equipment power generation loss prediction method based on the principal component analysis method and Resnet network according to claim 5, It is characterized in that In the acquisition of the data of the day before the prediction day in step S3, the method based on the ensemble empirical mode decomposition is specifically as follows: finding all the maximum points of the power generation loss data sequence X(t) and fitting them with a cubic spline interpolation function to form the upper envelope of the original data; Find all the minimum points of the power generation loss data series, and fit all the minimum points through the cubic spline interpolation function to form the lower envelope of the data; The mean of the upper envelope and the lower envelope is denoted as ml. The original data sequence X(t) is subtracted from the mean envelope ml to obtain a new data sequence hl: X(t)-ml=hl Using the ensemble empirical mode decomposition method, the photovoltaic equipment power generation loss sequence P predicted the day before is decomposed into a main sequence A 1 and three detail sequences D 1 , D 2 and D 3 .

8. The photovoltaic equipment power generation loss prediction method based on the principal component analysis method and Resnet network according to claim 1, It is characterized in that The output of the sequence value in step S4 is specifically: using the Resnet18 network to iterate the residual layer multiple times to obtain a minimum residual value, that is, the result with the minimum error between the predicted value and the true value, and using this result as the output result.

9. The photovoltaic equipment power generation loss prediction method using the principal component analysis method and Resnet network according to claim 8, It is characterized in that In the output of the sequence value, the main part prediction value is obtained specifically as follows: the output power of the photovoltaic equipment one day before the prediction day, the output power of the photovoltaic equipment on the prediction day, the equipment type index one day before the prediction day and the equipment type index on the prediction day are used as the input of the Resnet network model, and the main part of the power generation loss sequence corresponding to the prediction day is output to form the main part prediction value.

10. The photovoltaic equipment power generation loss prediction method using the principal component analysis method and Resnet network according to claim 9, It is characterized in that The details are as follows: using the combiner box temperature, component efficiency, inverter temperature, inverter conversion efficiency, and transformer winding temperature of the day before the prediction, the combiner box temperature, component efficiency, inverter temperature, inverter conversion efficiency, and transformer winding temperature of the day before the prediction, and the equipment type index of the day before the prediction and the equipment type index of the day as inputs of the Resnet network model, and outputting the details of the corresponding power generation loss sequence of the day respectively.

Images