Photovoltaic power station power loss calculation method based on neural network

Abstract

The invention provides a photovoltaic power station electric quantity loss calculation method based on a neural network, and the calculation method specifically comprises the steps: obtaining the operation data and meteorological data of a photovoltaic power station, and constructing a data set based on the operation data and the meteorological data; on the basis of a neural network model, calculating the power generation amount of the photovoltaic power station under the weather condition of the day according to the constructed data set; defining a power loss type; and performing electric quantity loss source judgment based on the operation data of the photovoltaic power station, determining the electric quantity loss type corresponding to the photovoltaic power station, and calculating the corresponding electric quantity loss in combination with the calculated required electric quantity and the actual electric quantity. According to the method, the neural network model is utilized to realize high-precision prediction of the power generation capacity, errors caused by environmental uncertainty are reduced, and on the basis, source analysis is performed on the lost power, so that different types of lost power are clearly distinguished and quantified, and the calculation accuracy of the lost power is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and in particular to a method for calculating power loss in a photovoltaic power station based on a neural network. Background Art

[0002] In the actual operation of photovoltaic power stations, due to the influence of power restrictions, equipment failures, system losses, component dust accumulation, etc., there will be a certain degree of power loss, which will affect the power generation efficiency and economic benefits of the power station. Accurate calculation and analysis of power loss in power stations not only helps to evaluate the actual power generation efficiency of power stations, reveal problems and room for improvement in power station operation, but also guides technical personnel to formulate operation strategies and improve the economic benefits of power stations.

[0003] Among the existing methods for calculating the power loss of photovoltaic power stations, the model inverter method mainly calculates the required power generation of the equipment with power loss based on the overall equivalent utilization hours of the model inverter, and then subtracts the actual power generation of the equipment with power loss from the required power generation of the equipment with power loss to obtain the power loss. When the model inverter is abnormal, its overall equivalent utilization hours will lose its reference significance, and it is difficult to ensure the accuracy of the calculation of power loss. When calculating the power loss by using the method of obtaining the required power generation and power loss based on the theoretical power generation efficiency of the power station, since the theoretical power generation efficiency needs to be based on a relatively ideal meteorological environment, this also leads to the fact that the power loss calculated by this method includes the loss caused by uncontrollable factors such as weather, which affects the subsequent analysis of the source of power loss, and thus affects the accuracy of the power loss calculation results.

[0004] It can be seen that the existing photovoltaic power station loss calculation method is easily affected by abnormal interference from the data source, and it is difficult to accurately analyze the source of loss. The accuracy of the loss calculation is difficult to guarantee, and it is difficult to achieve good results in the actual power station loss analysis and calculation.

[0005] In summary, how to better calculate and analyze the power loss of a photovoltaic power station is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art that the calculation of lost electricity is easily affected by abnormal interference from the data source, and it is difficult to accurately analyze the source of the loss and to ensure the accuracy of the calculation of lost electricity. A method for calculating lost electricity in a photovoltaic power station based on a neural network is provided, which utilizes a neural network model to achieve high-precision prediction of the amount of electricity to be generated, reduces errors caused by environmental uncertainty, and on this basis, performs source analysis on the lost electricity, thereby clearly distinguishing and quantifying different types of lost electricity and ensuring the accuracy of the lost electricity calculation.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A method for calculating power loss of a photovoltaic power station based on a neural network, comprising:

[0009] Step 1: Obtain the operation data and meteorological data of the photovoltaic power station, and build a data set based on the operation data and meteorological data;

[0010] Step 2: Based on the neural network model, the power generation of the photovoltaic power station under the meteorological conditions of the day is calculated according to the constructed data set;

[0011] Step 3: Define the type of power loss;

[0012] Step 4: Determine the source of power loss based on the operating data of the photovoltaic power station, determine the type of power loss corresponding to the photovoltaic power station, and calculate the corresponding power loss based on the calculated power generation and actual power generation.

[0013] Furthermore, the types of power loss defined in step three include power restriction loss, fault loss and error loss.

[0014] Furthermore, the step 4 of determining the source of power loss based on the operation data of the photovoltaic power station and determining the type of power loss includes:

[0015] Obtain power restriction records and fault records of the photovoltaic power station based on the operation data of the photovoltaic power station;

[0016] If there is a power-limiting record of the photovoltaic power station and no fault record, it is determined that the current power loss type includes power-limiting loss and error loss;

[0017] If there is a fault record of the photovoltaic power station and no unlimited power record, it is determined that the current power loss type includes fault loss and error loss;

[0018] If there are fault records and power-limiting records of the photovoltaic power station, the current power loss type is determined to include power-limiting loss, fault loss, and error loss;

[0019] If there is no fault record or power restriction record of the photovoltaic power station, it is determined that the current power loss includes error loss.

[0020] Furthermore, when it is determined that the current type of power loss includes power restriction loss and error loss, the power restriction loss is calculated based on the total radiation during the power restriction period corresponding to the power restriction record, the grid-connected capacity of the photovoltaic power station and the standard power generation efficiency of the photovoltaic power station, and the error loss is calculated based on the expected power generation, the actual power generation and the power restriction loss.

[0021] Furthermore, when it is determined that the current type of power loss includes fault loss and error loss, the fault power loss is calculated based on the total radiation during the fault period corresponding to the fault record, the grid-connected capacity of the fault area, and the standard power generation efficiency of the photovoltaic power station, and the error loss is calculated based on the expected power generation, the actual power generation and the fault loss.

[0022] Furthermore, when it is determined that the current power loss type includes power restriction loss, fault loss and error loss:

[0023] The power loss due to power restriction is calculated based on the total radiation amount during the power restriction period corresponding to the power restriction record, the grid-connected capacity of the photovoltaic power station and the standard power generation efficiency of the photovoltaic power station;

[0024] Calculate the power loss caused by the fault according to the total radiation during the fault period corresponding to the fault record, the grid-connected capacity of the fault area, and the standard power generation efficiency of the photovoltaic power station;

[0025] The error loss is calculated based on the expected power generation, actual power generation, power loss due to power restriction and power loss due to fault.

[0026] Furthermore, when it is determined that the current power loss includes error loss, the error loss is calculated based on the power generation to be generated and the actual power generation.

[0027] Furthermore, before constructing the data set based on the operating data and the meteorological data in step 1, the operating data and the meteorological data are screened, including:

[0028] Calculate the daily power generation efficiency of the photovoltaic power station based on the daily operation data and meteorological data, and sort the daily power generation efficiency in ascending order;

[0029] Select several daily power generation efficiencies according to the sorting results, and calculate the average of all the selected daily power generation efficiencies;

[0030] The calculated mean value is used as the standard power generation efficiency of the photovoltaic power station, and the dates with daily power generation efficiency lower than the standard power generation efficiency are screened out, and the operating data and meteorological data corresponding to the screened dates are discarded.

[0031] Furthermore, the step 1 of constructing a data set based on the operation data and the meteorological data includes:

[0032] Perform feature derivation processing on meteorological data and construct feature sets based on the derived features;

[0033] Standardize the constructed feature set;

[0034] The standardized feature set is screened using the random forest algorithm, and a data set is constructed based on the screened features and running data.

[0035] Furthermore, the step 2 is based on the neural network model to calculate the power generation of the photovoltaic power station under the meteorological conditions of the day according to the constructed data set, including:

[0036] Divide the constructed data set to obtain the training set and test set of the neural network model;

[0037] Train the neural network model according to the training set;

[0038] Based on the test set, according to the trained neural network model, the corresponding power generation under the meteorological conditions of the day is output.

[0039] The beneficial effects of the present invention are:

[0040] (1) Using the neural network model, taking into account the variable meteorological conditions, we can achieve high-precision prediction of the required power generation, reduce the error caused by environmental uncertainty, and improve the accuracy of the calculation of lost power. On the basis of accurate prediction of required power generation, we can further refine the classification calculation of lost power, so as to clearly distinguish and quantify different types of lost power, ensure the accuracy of the calculation of lost power, and provide strong support for power station operation and maintenance and optimization strategy formulation.

[0041] (2) When constructing the data set for power loss calculation, the average of some daily power generation efficiencies is selected as the standard power generation efficiency for data screening, which effectively avoids the impact of single equipment abnormalities, accurately identifies the inefficient operation status of the power station, and provides a reliable benchmark for power loss analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of a process of the present invention;

[0043] Figure 2 It is a schematic diagram of the network structure of a neural network model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0045] Example:

[0046] A method for calculating the power loss of photovoltaic power stations based on neural networks, such as Figure 1 As shown, including:

[0047] Step 1: Obtain the operation data and meteorological data of the photovoltaic power station, and build a data set based on the operation data and meteorological data;

[0048] Step 2: Based on the neural network model, the power generation of the photovoltaic power station under the meteorological conditions of the day is calculated according to the constructed data set;

[0049] Step 3: Define the type of power loss;

[0050] Step 4: Determine the source of power loss based on the operating data of the photovoltaic power station, determine the type of power loss corresponding to the photovoltaic power station, and calculate the corresponding power loss based on the calculated power generation and actual power generation.

[0051] The operating data of the photovoltaic power station obtained includes the daily power generation, installed capacity, geographical location information, and records of equipment failures, power restrictions, etc.

[0052] Since the power generation of a photovoltaic power station is related to natural factors such as solar radiation, temperature, and precipitation, the meteorological data includes the daily solar radiation, 2m temperature, 2m relative humidity, 10m wind speed, 10m wind direction, precipitation, total cloud cover, and snowfall at the location of the photovoltaic power station.

[0053] And considering that in the actual operation process of photovoltaic power stations, they are sometimes operated below their power generation capacity due to scheduling power restrictions, equipment failures, component dust accumulation and other external factors, which leads to differences between actual power generation and expected power generation. Therefore, before constructing a data set for training a neural network model for predicting expected power generation, it is necessary to filter the acquired data to select data whose actual power generation is as close to its theoretical expected power generation as possible. This can ensure the accuracy and reliability of the neural network model training, thereby better calculating the expected power generation of photovoltaic power stations under different meteorological conditions.

[0054] Therefore, in this embodiment, when the actual power generation of the photovoltaic power station is close to its theoretical power generation, it is regarded as a normal operating state. When the actual power generation is far less than the theoretical power generation, it is regarded as an inefficient operating state. When performing data screening, the operating data of the photovoltaic power station in the inefficient operating state and the meteorological data corresponding to the operating data are eliminated.

[0055] Specifically, before constructing the data set based on the operating data and meteorological data in step 1, the operating data and meteorological data are screened, including:

[0056] Calculate the daily power generation efficiency of the photovoltaic power station based on the daily operation data and meteorological data, and sort the daily power generation efficiency in ascending order;

[0057] Select several daily power generation efficiencies according to the sorting results, and calculate the average of all the selected daily power generation efficiencies;

[0058] The calculated mean value is used as the standard power generation efficiency of the photovoltaic power station, and the dates with daily power generation efficiency lower than the standard power generation efficiency are screened out, and the operating data and meteorological data corresponding to the screened dates are discarded.

[0059] Among them, the calculation formula for the daily power generation efficiency of the photovoltaic power station is:

[0060]

[0061] Among them, PR is the daily power generation efficiency, E is the daily power generation recorded by the photovoltaic power station, G is the daily total radiation at the geographical location of the photovoltaic power station, P is the installed capacity of the photovoltaic power station, and G STC is the irradiance under standard conditions, taking the constant as 1kW / m 2 .

[0062] According to the above formula, the corresponding daily power generation efficiency is calculated based on the daily operation data of the photovoltaic power station, and the daily power generation efficiency is sorted in ascending order from small to large, which is recorded as the set S = [PR1, PR2, ..., PR n ]. Taking the position Q3 of 75% of the set as the screening basis, select the daily power generation efficiency before Q3, and calculate the average of the selected daily power generation efficiency as the standard power generation efficiency. The calculation formula of the standard power generation efficiency of the photovoltaic power station is:

[0063]

[0064] Among them, PR 标准 is the standard power generation efficiency of the photovoltaic power station, n is the total number of elements in the set S, PR i is the daily power generation efficiency at the i-th position in the set S.

[0065] By calculating the average daily power generation efficiency of the top 75% positions as the standard power generation efficiency of the power station, the impact of abnormalities in a single device can be effectively avoided, the inefficient operating state of the power station can be accurately identified, and a reliable benchmark can be provided for power loss analysis.

[0066] Based on the above data screening, a data set is constructed based on operation data and meteorological data, including:

[0067] Perform feature derivation processing on meteorological data and construct feature sets based on the derived features;

[0068] Standardize the constructed feature set;

[0069] The standardized feature set is screened using the random forest algorithm, and a data set is constructed based on the screened features and running data.

[0070] In order to better capture the nonlinear relationship between power generation and meteorological data, the meteorological data is further derived to increase the dimension and information content of the features.

[0071] Specifically, meteorological data can be integrated into the original feature set F 原始= [f1,f2,f3,f4,f5,f6,f7,f8], which are solar radiation, 2m temperature, 2m relative humidity, 10m wind speed, 10m wind direction, precipitation, total cloud cover and snowfall. The derived features are as follows:

[0072] 1) Time feature: Generate month feature f9 and season feature f based on date 10 , in order to capture the impact of seasonal factors on power generation, where f9 is the month of the date, March to May is spring, June to August is summer, September to November is autumn, and December to February is winter. Set the seasonal feature f 10 The values ​​in spring, summer, autumn and winter are 1, 2, 3 and 4 respectively.

[0073] 2) Polynomial features: Generate polynomial features of 2m temperature and 2m relative humidity to capture the nonlinear relationship between temperature, relative humidity and power generation, where:

[0074] Polynomial characteristic f of 2m temperature 11 =f2 2 ;

[0075] Polynomial characteristic f of 2m relative humidity 12 =f3 2 .

[0076] 3) Cross-signature: Generate the cross-signature f of 2m temperature and irradiance 13 , 2m relative humidity and irradiation cross characteristics f 14 , in order to capture the impact of different temperature irradiation and humidity irradiation conditions on power generation, among which:

[0077] f 13 =f1×f2;

[0078] f 14 =f1×f3.

[0079] 4) Precipitation level feature: Generate precipitation level feature f 15 , in order to reduce the noise in the precipitation data, the 24-hour precipitation level classification is shown in Table 1. Based on the daily precipitation, the levels of sporadic light rain, light rain, moderate rain, heavy rain, heavy rain, heavy rain, and extremely heavy rain are set to 1, 2, 3, 4, 5, 6, and 7 respectively.

[0080] Table 1 24-hour precipitation level classification table

[0081] grade Precipitation (unit: mm) Scattered light rain <0.1 light rain 0.1～9.9 moderate rain 10.0～24.9 heavy rain 25.0～49.9 rainstorm 50.0～99.9 Heavy rain 100.0～249.9 Heavy rain ≥250.0

[0082] After completing feature derivation, the new feature set is:

[0083] F=[f1,f2,f3,f4,f5,f6,f7,f8,f9,f 10 ,f 11 ,f 12 ,f 13 ,f 14 ,f 15 ].

[0084] After obtaining the new feature set F, the features in the feature set F are first standardized. In this embodiment, the features are standardized by a normalization method, and the feature data is compressed to between 0 and 1, thereby converting it into dimensionless data. The normalization formula is:

[0085]

[0086] Where: x max and x min are the maximum and minimum values ​​in the feature set F, respectively, x is the initial value, and x′ is the normalized value.

[0087] After completing the standardization process, the feature set F after the standardization process is further subjected to feature selection through the random forest algorithm.

[0088] Random forest is a tree-based embedded feature selection model that selects important features by evaluating the importance of each feature. Specifically, the random forest algorithm is used to select features from the standardized feature set F, including:

[0089] Generate multiple decision trees;

[0090] The feature importance is calculated by counting the usage frequency of each feature in the tree model and its contribution to the reduction of impurity;

[0091] According to the feature importance, the top k most important features [f′1,f′2,f′3,…f′ k ] as the feature set for subsequent neural network model training.

[0092] Construct a dataset based on the selected feature set and running data.

[0093] The step 2 is based on the neural network model and the constructed data set to calculate the power generation of the photovoltaic power station under the meteorological conditions of the day, including:

[0094] Divide the constructed data set to obtain the training set and test set of the neural network model;

[0095] Train the neural network model according to the training set;

[0096] Based on the test set, according to the trained neural network model, the corresponding power generation under the meteorological conditions of the day is output.

[0097] In this embodiment, a deep neural network model (DNN) is specifically used to calculate the required power generation of a photovoltaic power station. DNN is a variant of an artificial neural network, which consists of multiple neural network layers, usually including an input layer, multiple hidden layers, and an output layer. Each layer contains multiple neurons, each neuron contains multiple weights and an activation function. When the input data is passed to the neuron, it is weighted and summed, and then the summation result is passed to the activation function for nonlinear transformation. DNN learns more advanced and abstract feature representations through multiple levels of nonlinear transformation. DNN training is usually performed through a back-propagation algorithm, and the network weights are adjusted according to the difference between the output result and the true label.

[0098] In this embodiment, the feature set [f′1,f′2,f′3,…f′ k ] as the input of the neural network model, and the power generation of the photovoltaic power station as the output of the neural network model. The network structure of the neural network model of this embodiment is as follows: Figure 2 shown.

[0099] Before training the neural network model, the constructed data set is first divided into a ratio of 8:2 to obtain a training set and a test set respectively. The meteorological data-related features in the training set are used as the input set of the DNN, and the samples in the training set are further trained by minimizing the MSE loss. The loss function is used to calculate the loss value by combining the output power generation and the actual power generation. Then, based on the optimizer (such as SGD), the weight parameters are slightly updated along the gradient direction, and the forward and backward propagation processes are repeated, and iterated repeatedly until the corresponding weight parameter model with the minimum loss value is obtained.

[0100] The loss function MSE formula is:

[0101]

[0102] Among them, y i and y′ i are the actual power generation and the expected power generation output by the model, and m is the sample point.

[0103] Obtain the optimal DNN model after training convergence, test the test set based on the model, and output the corresponding power generation.

[0104] In addition to using deep neural networks (DNNs) to calculate the power generation required by photovoltaic power stations under different meteorological conditions, the DNN model can also be replaced by other regression models, such as XGBoost, SVR, etc. In addition, multiple regression models can also be used for combined prediction.

[0105] The types of power loss defined in step 3 include power cut loss, fault loss and error loss.

[0106] Among them, power curtailment loss refers to the amount of electricity that can be generated but is not generated by photovoltaic power stations due to factors such as power load and peak regulation capacity of the power system, grid constraints, and safe and stable operation;

[0107] Fault loss refers to the amount of electricity that can be generated but is not generated due to equipment failure, defect or maintenance.

[0108] Error losses include the amount of electricity that could have been generated but was not generated by the photovoltaic power station due to system losses and panel contamination.

[0109] The fourth step of judging the source of power loss based on the operation data of the photovoltaic power station and determining the type of power loss includes:

[0110] Obtain power restriction records and fault records of the photovoltaic power station based on the operation data of the photovoltaic power station;

[0111] If there is a power-limiting record of the photovoltaic power station and no fault record, it is considered that the power loss on that day is mainly composed of power-limiting loss, and the current power loss type is judged to include power-limiting loss and error loss;

[0112] If there is a fault record of the photovoltaic power station and no unlimited power record, it is considered that the power loss on that day is mainly composed of fault loss, and the current power loss type is judged to include fault loss and error loss;

[0113] If there are fault records and power-limiting records of the photovoltaic power station, it is considered that the power loss on that day is mainly composed of fault loss and power-limiting loss, and the current power loss type is judged to include power-limiting loss, fault loss and error loss;

[0114] If there is no fault record or power restriction record of the photovoltaic power station, it is considered that the power loss on that day is mainly composed of error loss, and it is determined that the current power loss includes error loss.

[0115] When it is determined that the current power loss type includes power restriction loss and error loss, the total radiation amount G1 during the power restriction period, the grid-connected capacity P1 of the photovoltaic power station and the standard power generation efficiency PR of the photovoltaic power station are calculated according to the power restriction record. 标准 Calculate the power loss due to power restriction, and calculate the error loss based on the power generation to be generated, the actual power generation and the power loss due to power restriction. The calculation formula for power loss due to power restriction is:

[0116] Power loss due to power restriction = (P1×G1 / G STC )×PR 标准 ;

[0117] Among them, G STC The irradiation is under standard conditions.

[0118] At this time, the error loss power is:

[0119] Error power loss = required power generation - actual power generation - power loss due to power restriction.

[0120] When it is determined that the current power loss type includes fault loss and error loss, the total radiation G2 during the fault period corresponding to the fault record, the grid-connected capacity P2 of the fault area, and the standard power generation efficiency PR of the photovoltaic power station are calculated. 标准 Calculate the power loss due to the fault, and calculate the error loss based on the power generation to be generated, the actual power generation and the power loss due to the fault. The calculation formula for the power loss due to the fault is:

[0121] Fault power loss = (P2×G2 / G STC )×PR 标准 .

[0122] At this time, the error loss power is:

[0123] Error power loss = expected power generation - actual power generation - power loss due to fault.

[0124] When it is determined that the current power loss type includes power restriction loss, fault loss and error loss:

[0125] The power loss due to power restriction is calculated based on the total radiation amount during the power restriction period corresponding to the power restriction record, the grid-connected capacity of the photovoltaic power station and the standard power generation efficiency of the photovoltaic power station;

[0126] Calculate the power loss caused by the fault according to the total radiation during the fault period corresponding to the fault record, the grid-connected capacity of the fault area, and the standard power generation efficiency of the photovoltaic power station;

[0127] The error loss is calculated based on the expected power generation, actual power generation, power loss due to power restriction and power loss due to fault.

[0128] At this time, the calculation method of fault loss and power restriction loss is the same as above, error loss = required power generation - actual power generation - fault loss - power restriction loss.

[0129] When it is determined that the current power loss includes error loss, the error loss is calculated based on the power generation to be generated and the actual power generation.

[0130] At this time, the calculation formula for the error loss is:

[0131] Other power loss = expected power generation - actual power generation.

[0132] Based on the calculation results of the lost electricity, technical improvements can be made to the photovoltaic power station according to actual conditions, or strategies can be provided for future operation and maintenance.

[0133] Take the calculation of power loss of a grid-connected distributed photovoltaic power station in a certain area as an example:

[0134] The operation data and meteorological data of the photovoltaic power station from January to December of a certain year are obtained. The daily power generation and installed capacity are recorded as E and P respectively. The daily solar radiation, 2m temperature, 2m relative humidity, 10m wind speed, 10m wind direction, precipitation, total cloud cover and snowfall are f1, f2, f3, f4, f5, f6, f7 and f8 respectively. The power generation unit is kWh, the installed capacity unit is kW, and the solar radiation unit is kWh / m 2 .

[0135] Based on daily power generation E, installed capacity P, solar radiation f1 and irradiation G under standard conditions STC , calculate the daily power generation efficiency

[0136] Sort the daily power generation efficiency from small to large, and take the average of the power generation efficiencies in the top 75% as the standard power generation efficiency PR of the power station. 标准 ;

[0137] When the power generation efficiency on a certain day is less than the standard power generation efficiency, it is considered that the power station is operating inefficiently and the data for that day is deleted.

[0138] Perform feature derivation processing on meteorological data to generate monthly feature f9 and seasonal feature f 10 , 2m temperature polynomial characteristic f 11 、2m relative humidity polynomial characteristic f 12 , 2m temperature and irradiation cross characteristics f 13 , 2m relative humidity and irradiation cross characteristics f 14 and precipitation level characteristics f 15 , get the feature set F.

[0139] For the feature set F = [f1,f2,f3,f4,f5,f6,f7,f8,f9,f 10 ,f 11 ,f 12 ,f 13 ,f 14 ,f 15 ] is normalized, and the feature set F is screened using the random forest algorithm to obtain the k screened features [f′1,f′2,f′3,…f′ k ], combined with the running data to build a dataset for model training.

[0140] Among them, k features [f′1,f′2,f′3,…f′ k ] corresponds to the meteorological data characteristics of one day. The data set includes daily meteorological data characteristics and corresponding operation data.

[0141] The data set is divided into training set and test set in a ratio of 8:2, and the input of the neural network model is set as the meteorological data features of the corresponding date [f′1,f′2,f′3,…f′ k ], the model output is the expected power generation E0, and the model training is carried out according to the operation data of the corresponding date.

[0142] Based on the PyTorch deep learning framework, the DNN network is trained to calculate the required power generation and obtain the optimal PTH model weights.

[0143] Based on the trained model and the meteorological data features in the test set, the expected power generation of the test set is output.

[0144] Based on the operating data of the test set and the power generation expected by the DNN model, the daily power loss in the test set is calculated as follows:

[0145] If there is a power-limiting record but no fault record in the operating data of the day, it is considered that the power loss of the day is mainly composed of power-limiting losses. The accumulated radiation data during the power-limiting period is G1, the grid-connected capacity is P1, and the power loss due to power-limiting is:

[0146]

[0147] E 误差损失 =E0-EE 限电损失 .

[0148] If there is a fault record but no unlimited power record in the operating data of the day, it is considered that the power loss of the day is mainly composed of fault losses. The accumulated radiation data during the fault period is G2, the grid-connected capacity of the fault area is P2, and the power loss due to the fault is:

[0149]

[0150] E 误差损失 =E0-EE 故障损失 .

[0151] If there are fault records and power-limiting records in the operating data of the day, it is considered that the power loss on that day is mainly composed of fault loss and power-limiting loss. 限电损失 +E 故障损失 =E0-E.

[0152] If there is no unlimited power record and no fault record in the operation data of that day, it is considered that the power loss of that day is mainly composed of error loss, and the error power loss is:

[0153] E 误差损失 =E0-E.

[0154] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A method for calculating power loss in a photovoltaic power station based on a neural network, characterized in that: include: Step 1: Obtain the operation data and meteorological data of the photovoltaic power station, and build a data set based on the operation data and meteorological data; Step 2: Based on the neural network model, the power generation of the photovoltaic power station under the meteorological conditions of the day is calculated according to the constructed data set; Step 3: Define the type of power loss; Step 4: Determine the source of power loss based on the operating data of the photovoltaic power station, determine the type of power loss corresponding to the photovoltaic power station, and calculate the corresponding power loss based on the calculated power generation and actual power generation.

2. A method for calculating power loss in a photovoltaic power station based on a neural network according to claim 1, characterized in that: The types of power loss defined in step 3 include power cut loss, fault loss and error loss.

3. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 2, characterized in that: The fourth step of judging the source of power loss based on the operation data of the photovoltaic power station and determining the type of power loss includes: Obtain power restriction records and fault records of the photovoltaic power station based on the operation data of the photovoltaic power station; If there is a power-limiting record of the photovoltaic power station and no fault record, it is determined that the current power loss type includes power-limiting loss and error loss; If there is a fault record of the photovoltaic power station and no unlimited power record, it is determined that the current power loss type includes fault loss and error loss; If there are fault records and power-limiting records of the photovoltaic power station, the current power loss type is determined to include power-limiting loss, fault loss, and error loss; If there is no fault record or power restriction record of the photovoltaic power station, it is determined that the current power loss includes error loss.

4. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 3, characterized in that: When it is determined that the current type of power loss includes power restriction loss and error loss, the power restriction loss is calculated based on the total radiation during the power restriction period corresponding to the power restriction record, the grid-connected capacity of the photovoltaic power station and the standard power generation efficiency of the photovoltaic power station, and the error loss is calculated based on the expected power generation, the actual power generation and the power restriction loss.

5. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 3, characterized in that: When it is determined that the current power loss type includes fault loss and error loss, the fault loss power is calculated based on the total radiation during the fault period corresponding to the fault record, the grid-connected capacity of the fault area, and the standard power generation efficiency of the photovoltaic power station, and the error loss is calculated based on the expected power generation, the actual power generation and the fault loss power.

6. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 3, characterized in that: When it is determined that the current power loss type includes power restriction loss, fault loss and error loss: the power restriction loss is calculated according to the total radiation amount in the power restriction period corresponding to the power restriction record, the grid-connected capacity of the photovoltaic power station and the standard power generation efficiency of the photovoltaic power station; Calculate the power loss caused by the fault according to the total radiation during the fault period corresponding to the fault record, the grid-connected capacity of the fault area, and the standard power generation efficiency of the photovoltaic power station; The error loss is calculated based on the expected power generation, actual power generation, power loss due to power restriction and power loss due to fault.

7. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 3, characterized in that: When it is determined that the current power loss includes error loss, the error loss is calculated based on the power generation to be generated and the actual power generation.

8. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 1, characterized in that: Before constructing the data set based on the operating data and meteorological data in step 1, the operating data and meteorological data are screened, including: Calculate the daily power generation efficiency of the photovoltaic power station according to the daily operation data, and sort the daily power generation efficiency in ascending order; According to the sorting results, several daily power generation efficiencies are selected, and the average of all selected daily power generation efficiencies is calculated; the calculated average is taken as the standard power generation efficiency of the photovoltaic power station, and the dates with daily power generation efficiency lower than the standard power generation efficiency are screened out, and the operating data and meteorological data corresponding to the screened out dates are eliminated.

9. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 1, characterized in that: The data set is constructed based on the operation data and meteorological data described in step 1, including: Perform feature derivation processing on meteorological data and construct feature sets based on the derived features; Standardize the constructed feature set; The standardized feature set is screened using the random forest algorithm, and a data set is constructed based on the screened features and running data.

10. The method for calculating the power loss of a photovoltaic power station based on a neural network according to claim 1, characterized in that: The step 2 is based on the neural network model and the constructed data set to calculate the power generation of the photovoltaic power station under the meteorological conditions of the day, including: Divide the constructed data set to obtain the training set and test set of the neural network model; Train the neural network model according to the training set; Based on the test set and the trained neural network model, the corresponding power generation under the meteorological conditions of the day is output.

Images