Photovoltaic shielding type and degree diagnosis method based on I-V curve and deep learning

Abstract

The invention discloses a photovoltaic shielding type and degree diagnosis method based on an I-V curve and deep learning. The method comprises the following implementation steps: (1) collecting I-V curve data of a photovoltaic string; (2) carrying out preprocessing and image coding on the acquired I-V curve data; (3) a double-flow deep neural network is adopted to diagnose the photovoltaic shielding type; and (4) estimating the shielding severity by combining a photovoltaic shielding diagnosis result and the mechanism model. The method can be used for solving the problem that the specific type and degree cannot be diagnosed in photovoltaic shielding detection, and is beneficial to adopting an effective power optimization method and an operation and maintenance strategy.

Description

Technical Field

[0001] The invention relates to a photovoltaic shading type and degree diagnosis method based on IV curve and deep learning, belonging to the field of new energy. Background Art

[0002] Shading is one of the common anomalies in the operation of photovoltaic systems, which will cause power loss and hot spot effect in photovoltaic systems. At present, common shading detection methods are mainly divided into image-based methods, methods based on string current, voltage and power data, and methods based on IV curves. Image-based methods can intuitively reflect the shading situation and are suitable for large-scale monitoring, but they require additional image acquisition equipment and complex image processing algorithms, and are easily affected by ambient light. Methods based on string current, voltage and power data detect shading by analyzing the changes in electrical parameters of photovoltaic strings, such as abnormal current, voltage or power. This type of method is suitable for real-time monitoring and has low cost, but its accuracy depends on the frequency of data acquisition and the complexity of data processing algorithms, and may not be sensitive to small-scale shading. The IV curve-based method detects shading by analyzing the changes in the shape of the IV curve of the photovoltaic array, which can reflect the changes in electrical performance in detail and is suitable for diagnosing complex shading faults. However, the current shading detection method can only realize the occurrence of shading, but cannot diagnose the specific shading type and estimate the severity in the actual operation of the photovoltaic system, and it is difficult to accurately detect small shading. Summary of the invention

[0003] To solve the above problems, the present invention provides a photovoltaic shading type and degree diagnosis method based on IV curve and deep learning. The method has high diagnostic performance, can realize the diagnosis and severity estimation of specific shading types, and has a certain robustness.

[0004] To achieve the above object, the technical solution adopted by the present invention is: a photovoltaic shading type and degree diagnosis method based on IV curve and deep learning, comprising the following steps:

[0005] Step 1: Collect IV curve data of PV strings;

[0006] Step 2: Preprocess and visualize the IV curve;

[0007] Step 3: Construct a two-stream deep neural network to diagnose the photovoltaic shading type;

[0008] Step 4: Combine the occlusion type diagnosis results and the mechanism model to estimate the occlusion severity.

[0009] Furthermore, the specific method of step 1 is: collecting the IV curve of the photovoltaic string by using an IV curve scanning device integrated in the inverter or additionally installed.

[0010] Furthermore, the step 2 specifically includes the following steps:

[0011] Step 2.1: De-noise the collected IV curve to remove the burr points on the curve;

[0012] Step 2.2: Convert the IV curve under actual operation to standard condition (STC) by the following formula, with irradiance of 1000W / m 2 , temperature is 25℃:

[0013]

[0014] Among them, G and T are the irradiance and temperature in the actual state, β i is the short-circuit current temperature coefficient, β v is the open circuit voltage temperature coefficient, K g is the irradiance correction factor for voltage, I STC and V STC is the current and voltage in the STC state, I out and V out is the current and voltage in the actual state, G STC and T STC are the irradiance and temperature under STC state respectively.

[0015] Step 2.3: Interpolate the IV curve after STC transformation and resample from different resolutions to the same resolution, with a total of 402 sampling points and the same voltage value interval.

[0016] Step 2.4: Encode the preprocessed IV curve to enhance the photovoltaic shading characteristics. For a current sequence C = (c1, c2, ..., c N ), first perform normalization, the formula is as follows:

[0017]

[0018] Among them, c i is the sampling point of the current sequence, is the sampling point of the normalized current series;

[0019] Then, the normalized sequence is converted from the Cartesian coordinate system to the polar coordinate system, and the formulas for the polar coordinate angle and image encoding can be expressed as:

[0020]

[0021] Among them, θ i is the polar coordinate angle, IMAGE is the image encoding of the IV curve, which can be represented by an RGB image.

[0022] Furthermore, the step 3 specifically includes the following steps:

[0023] Step 3.1: Design LSTM modules to extract various occlusion features reflected in the IV curve sequence data. The number of LSTM layers in the two-stream deep neural network is 2, the input scale of the first layer is 1, the hidden layer size is 128, and the output scale is 512. The hidden layer and output scale of the second layer are 128 and 512 respectively.

[0024] Step 3.2: Design a 2D-CNN module and add two convolutional attention modules after the maximum pooling layer to improve the feature extraction capability of the encoded image. At the same time, introduce a global maximum pooling module to reduce the parameters of the model and increase the training speed.

[0025] Step 3.3: Integrate the features extracted by LSTM and 2D-CNN respectively, fuse the IV curve temporal features and the features of the encoded image, and combine with the Softmax module to achieve accurate classification of occlusion types.

[0026] Furthermore, the step 4 specifically includes the following steps:

[0027] Step 4.1: Calculate the power loss rate SR caused by aging based on the IV curve data collected under normal conditions aging , eliminating the impact of aging power loss on occlusion severity estimation.

[0028] Step 4.2: Based on the occlusion diagnosis results obtained in step 3, calculate the average occlusion rate in each case to measure the overall occlusion severity. The calculation formula is as follows:

[0029]

[0030] Among them, P shading is the power under shading condition, P normal is the power in normal state;

[0031] Step 4.3: Combine the mechanism model and IV curve of each shading to calculate the number of PV modules or submodules affected and the dust accumulation rate under different shading conditions, and use the following formulas to calculate:

[0032]

[0033] Among them, N module is the number of modules in the photovoltaic string, V oc,module is the open circuit voltage of a normal module under standard conditions, V oc,submodule Open circuit voltage of normal submodule under standard conditions, N module,shaded is the number of obscured modules, Nsubmodule,shaded The number of obscured submodules, V lpp is the voltage of the first extreme point of the IV curve, k is the characterization coefficient, which is 0.75 after considering the aging effect, SR dust is the dust accumulation rate, I sc,dust is the short-circuit current under dust accumulation condition, I sc It is the short-circuit current of the string under normal conditions.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) Based on a dual-stream deep neural network integrating LSTM and 2D-CNN, the present invention makes full use of the original IV curve and the transformed image information, deeply explores the characteristics of different shading types, and accurately diagnoses the photovoltaic shading type.

[0036] (2) Based on the photovoltaic mechanism model and the key point information of the IV curve, the present invention further evaluates the severity of each shading type, providing reliable guidance for designing power optimization methods for photovoltaic systems under different shading conditions and formulating operation and maintenance plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is an overall flow chart of the photovoltaic shading type and degree diagnosis method based on IV curve and deep learning provided by the present invention;

[0038] Figure 2 It is a diagram of the structure of a two-stream deep neural network;

[0039] Figure 3 is a flowchart of occlusion severity assessment;

[0040] Figure 4 It is the IV curve diagram of various types of photovoltaic shading. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0042] Figure 1 It is the overall flow chart of the method of the present invention, and the specific steps are as follows:

[0043] Step 1: Collect the IV curve of the PV string using an IV curve scanning device integrated into the inverter or installed additionally.

[0044] Step 2: Preprocess and visualize the IV curve. The specific steps are as follows:

[0045] Step 2.1: De-noise the collected IV curve to remove the burr points on the curve.

[0046] Step 2.2: Convert the IV curve under actual operation to standard condition (STC) by the following formula, with irradiance of 1000W / m 2 , temperature is 25℃:

[0047]

[0048] Among them, G and T are the irradiance and temperature in the actual state, β i is the short-circuit current temperature coefficient, β v is the open circuit voltage temperature coefficient, K g is the irradiance correction factor for voltage, I STC and V STC is the current and voltage in the STC state, I out and V out is the current and voltage in the actual state, G STC and T STC are the irradiance and temperature under STC state respectively.

[0049] Step 2.3: Interpolate the IV curve after STC transformation and resample from different resolutions to the same resolution, with a total of 402 sampling points and the same voltage value interval.

[0050] Step 2.4: Encode the preprocessed IV curve into images to enhance the photovoltaic shading characteristics. For a current sequence C = (c1, c2, ..., c N ), first perform normalization, the formula is as follows:

[0051]

[0052] Among them, c i is the sampling point of the current sequence, is the sampling point of the normalized current series;

[0053] Then, the normalized sequence is converted from the Cartesian coordinate system to the polar coordinate system, and the formulas for the polar coordinate angle and image encoding can be expressed as:

[0054]

[0055] Among them, θ i is the polar coordinate angle, IMAGE is the image encoding of the IV curve, which can be represented by an RGB image.

[0056] Step 3: Construct a two-stream deep neural network to diagnose the photovoltaic shading type. The specific steps are as follows:

[0057] Step 3.1: Design LSTM modules to extract various occlusion features reflected in the IV curve sequence data. The number of LSTM layers in the two-stream deep neural network is 2, the input scale of the first layer is 1, the hidden layer size is 128, and the output scale is 512. The hidden layer and output scale of the second layer are 128 and 512 respectively.

[0058] Step 3.2: Design a 2D-CNN module and add two convolutional attention modules after the maximum pooling layer to improve the feature extraction capability of the encoded image. At the same time, introduce a global maximum pooling module to reduce the parameters of the model and increase the training speed.

[0059] Step 3.3: Integrate the features extracted by LSTM and 2D-CNN respectively, fuse the IV curve temporal features and the features of the encoded image, and combine with the Softmax module to achieve accurate classification of occlusion types.

[0060] Step 4: Combine the occlusion type diagnosis results and the mechanism model to estimate the occlusion severity. The specific steps are as follows:

[0061] Step 4.1: Calculate the power loss rate SR caused by aging based on the IV curve data collected under normal conditions aging , eliminating the impact of aging power loss on occlusion severity estimation.

[0062] Step 4.2: Based on the occlusion diagnosis results obtained in step 3, calculate the average occlusion rate in each case to measure the overall occlusion severity. The calculation formula is as follows:

[0063]

[0064] Among them, P shading is the power under shading condition, P normal is the power in normal state;

[0065] Step 4.3: Combine the mechanism model and IV curve of each shading to calculate the number of PV modules or submodules affected and the dust accumulation rate under different shading conditions, and use the following formulas to calculate:

[0066]

[0067]

[0068] Among them, N module is the number of modules in the photovoltaic string, V oc,module is the open circuit voltage of a normal module under standard conditions, V oc,submodule Open circuit voltage of normal submodule under standard conditions, N module,shaded is the number of obscured modules, Nsubmodule,shaded The number of obscured submodules, V lpp is the voltage of the first extreme point of the IV curve, k is the characterization coefficient, which is 0.75 after considering the aging effect, SR dust is the dust accumulation rate, I sc,dust is the short-circuit current under dust accumulation condition, I sc It is the short-circuit current of the string under normal conditions.

[0069] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

[0070] The above description is only a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope of protection required by the present invention.

Claims

1. A photovoltaic shading type and degree diagnosis method based on IV curve and deep learning, characterized by: The following steps are involved: Step 1: Collect IV curve data of PV strings; Step 2: Preprocess and visualize the IV curve; Step 3: Construct a two-stream deep neural network to diagnose the photovoltaic shading type; Step 4: Combine the occlusion type diagnosis results and the mechanism model to estimate the occlusion severity.

2. The photovoltaic shading type and degree diagnosis method based on IV curve and deep learning according to claim 1 is characterized in that: The step 2 specifically includes the following steps: Step 2.1: De-noise the collected IV curve to remove the burr points on the curve; Step 2.2: Convert the IV curve under actual operation to the standard state STC, with an irradiance of 1000W / m 2 , the temperature is 25℃, the specific formula is as follows: Among them, G and T are the irradiance and temperature in the actual state, β i is the short-circuit current temperature coefficient, β v is the open circuit voltage temperature coefficient, K g is the irradiance correction factor for voltage, I STC and V STC is the current and voltage in the STC state, I out and V out is the current and voltage in the actual state, G STC and T STC are the irradiance and temperature under STC state respectively; Step 2.3: Interpolate the IV curve after STC transformation and resample from different resolutions to the same resolution, with a total of 402 sampling points and the same voltage value interval; Step 2.4: Encode the preprocessed IV curve into images to enhance the photovoltaic shielding characteristics; for a current sequence C of an IV curve = (c1, c2, ..., c N ), first perform normalization, the formula is as follows: Among them, c i is the sampling point of the current sequence, is the sampling point of the normalized current series; Then, the normalized current sequence is converted from the Cartesian coordinate system to the polar coordinate system, and the formulas for the polar coordinate angle and image encoding are expressed as: Among them, θ i is the polar coordinate angle, IMAGE is the image encoding of the IV curve, represented by an RGB image.

3. The photovoltaic shading type and degree diagnosis method based on IV curve and deep learning according to claim 1 is characterized in that: The step 3 specifically comprises the following steps: Step 3.1: Design an LSTM module to extract various occlusion features reflected in the IV curve sequence data; the number of LSTM layers in the two-stream deep neural network is 2, the input scale of the first layer is 1, the hidden layer size is 128, and the output scale is 512. The hidden layer and output scale of the second layer are 128 and 512 respectively; Step 3.2: Design a 2D-CNN module and add two convolutional attention modules after the maximum pooling layer to improve the feature extraction capability of the encoded image. At the same time, introduce a global maximum pooling module to reduce the parameters of the model and improve the training speed. Step 3.3: Integrate the features extracted by LSTM and 2D-CNN respectively, fuse the IV curve temporal features and the features of the encoded image, and combine with the Softmax module to achieve accurate classification of occlusion types.

4. The photovoltaic shading type and degree diagnosis method based on IV curve and deep learning according to claim 1 is characterized in that: The step 4 specifically comprises the following steps: Step 4.1: Calculate the power loss rate SR caused by aging based on the IV curve data collected under normal conditions aging ,eliminates the influence of aging power loss on occlusion severity estimation; Step 4.2: Based on the occlusion diagnosis results obtained in step 3, calculate the average occlusion rate in each case to measure the overall occlusion severity. The calculation formula is as follows: Among them, P shading is the power under shading condition, P normal is the power in normal state; Step 4.3: Combine the mechanism model and IV curve of each shading to calculate the number of PV modules or submodules affected and the dust accumulation rate under different shading conditions, and use the following formulas to calculate them respectively: Among them, N module is the number of modules in the photovoltaic string, V oc,module is the open circuit voltage of a normal module under standard conditions, V oc,submodule Open circuit voltage of normal submodule under standard conditions, N module,shaded is the number of obscured modules, N submodule,shaded The number of obscured submodules, V lpp is the voltage of the first extreme point of the IV curve, k is the characterization coefficient, which is 0.75 after considering the aging effect, SR dust is the dust accumulation rate, I sc,dust is the short-circuit current under dust accumulation condition, I sc It is the short-circuit current of the string under normal conditions.

Images