CN-121997259-A - Photovoltaic device stability prediction method and related device based on multi-mode model

CN121997259ACN 121997259 ACN121997259 ACN 121997259ACN-121997259-A

Abstract

The embodiment of the application provides a photovoltaic device stability prediction method and a related device based on a multi-mode model, which are characterized in that a multi-mode sample set is input into a trained multi-mode deep neural network model to output a high-dimensional feature vector representing the state of a comprehensive device, then the high-dimensional feature vector is input into a time sequence predictor model to obtain a photoelectric conversion efficiency PCE predicted value of each time node in the whole life cycle of the photovoltaic device to be detected, a performance attenuation curve is drawn to realize the accurate pre-judgment of the whole life cycle performance of the photovoltaic device, the stability rating of the photovoltaic device to be detected can be generated according to the performance attenuation curve, the key factors influencing the stability of the device are positioned when the stability rating is detected to be in an unstable state, a device optimization proposal is generated, the stability evaluation period is greatly shortened, the test cost is reduced, and a high-efficiency technical support is provided for device research and development and mass production optimization.

Inventors

Du Daixuan
ZHOU CHUANZHE
WU XIAOXUE
HUANG LIANG
SHI YUXIAN

Assignees

港华能源创科(深圳)有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. A method for predicting stability of a photovoltaic device based on a multimodal model, the method comprising: Collecting multi-source data of a photovoltaic device to be tested, wherein the multi-source data comprises at least two of image data, spectrum data, initial photoelectric data and an environmental stress parameter sequence; Respectively executing modal adaptation preprocessing on the multi-source data to obtain a multi-modal sample set; the multi-modal deep neural network model comprises modal sub-networks and a feature fusion layer which are arranged in parallel, wherein each modal sub-network is used for carrying out feature extraction on the training sample set, and the feature fusion layer is used for fusing the extracted features of each modal sub-network to obtain the high-dimensional feature vector; inputting the high-dimensional feature vector into a time sequence predictor model to obtain a PCE predicted value of each time node in the whole life cycle of the photovoltaic device to be tested and drawing a performance attenuation curve; generating a stability rating of the photovoltaic device to be tested according to the performance decay curve, wherein the stability rating comprises a stable state and an unstable state; Detecting that the stability rating is in the unstable state, and determining key factors influencing the stability of the device according to the performance attenuation curve and the multi-source data, wherein the key factors comprise at least one of a defect area, a sensitive spectrum and environmental influence factors; and generating device optimization improvement suggestions according to the key factors.
2. The method of claim 1, wherein the image data is used to reflect grain morphology and defect distribution of the photovoltaic device to be tested, the spectral data is used to reflect material structure and stress information of the device, the initial photovoltaic data includes an open circuit voltage Voc, a short circuit current density Jsc, a fill factor FF, and an initial photovoltaic conversion efficiency PCE, and the environmental stress parameter sequence includes a temperature T, a relative humidity RH, and an illumination intensity during testing or use; The system comprises a modal sub-network, a spectrum sub-network and a numerical sub-network, wherein the image sub-network adopts a convolutional neural network CNN or a lightweight CNN to extract image characteristics, the image characteristics comprise texture and structure characteristics, the spectrum sub-network adopts a spectrum analysis network or a coding network to extract spectrum characteristics, the spectrum characteristics comprise spectrum peak positions and intensity characteristics, and the numerical sub-network adopts a full-connection network or a convolutional network to extract numerical characteristics of the initial photoelectric data and the environmental stress parameter sequence.
3. The method of claim 1, wherein the feature fusion layer uses a multi-head self-attention mechanism of a transfomer network to realize the associated modeling of cross-modal features through self-attention weight calculation, or And after the multi-mode variation self-encoder VAE is adopted to encode the mode characteristics into hidden variables which are distributed in a combined way, a fused high-dimensional characteristic vector is generated through a decoding process.
4. The method according to claim 1, wherein the time series predictor model adopts a long-short-term memory network LSTM or a modified structure thereof, the modified structure comprising a bidirectional LSTM, a gated loop unit GRU; The specific process of generating the photoelectric conversion efficiency PCE predicted value of each time node in the whole life cycle of the photovoltaic device to be tested and drawing the performance attenuation curve by the time sequence predictor model comprises the following steps: Inputting the high-dimensional feature vector at the time t and the corresponding environmental stress parameter into the LSTM or an improved structure thereof; processing input data at the time t through a plurality of memory gating units, and outputting a device performance predicted value at the time t+1; Information integration is carried out on the environmental stress parameter at the time t+1 and the performance predicted value at the time t+1, so that input data at the time t+1 is formed; Updating t to be t+1, and repeatedly executing the steps of inputting the high-dimensional feature vector at the moment t and the corresponding environmental stress parameter into the LSTM or the improved structure thereof and the following steps, and sequentially iterating to obtain PCE predicted values at the moment t+2 and t+3. And associating PCE predicted values of all time nodes in the whole life cycle according to time axis sequence to form a continuous performance decay curve.
5. The method of claim 1, wherein prior to generating the stability rating for the photovoltaic device under test from the performance decay curve, the method further comprises: And calculating a decision coefficient R2, an average absolute error MAE and a root mean square error RMSE on a verification set of the multi-mode sample set to verify the model prediction accuracy.
6. A method according to claim 3, wherein said determining key factors affecting device stability from said performance decay curve and said multi-source data set comprises: Identifying an attenuation abnormal interval in the performance attenuation curve, wherein the attenuation abnormal interval comprises an attenuation rate mutation time point, an acceleration attenuation interval and a final failure threshold time point; The multisource data corresponding to the attenuation curve abnormal interval is obtained, and the target image defect area and the target sensitive spectrum band with the largest attenuation contribution are visually positioned through the attention weight output by the feature fusion layer; analyzing the relevance between the environmental stress parameter and the attenuation curve abnormal section by adopting a control variable method, and determining a target environmental influence factor of dominant attenuation; and combining a preset multi-mode characteristic-failure mechanism mapping library, fusing the target image defect area, the target sensitive spectrum band and the target environment influence factors, and outputting the priority ordering of the key factors.
7. The method of claim 6, wherein generating device optimization improvement suggestions based on the key factors comprises: Acquiring a preset optimization measure rule base, wherein the optimization measure rule base comprises a mapping relation between the key factors and the device optimization improvement suggestions; acquiring corresponding device optimization improvement suggestions according to the key factors and the optimization measure rule base; and determining the priority order of each device optimization improvement suggestion according to the priority order of the key factors.
8. A photovoltaic device stability prediction apparatus based on a multimodal model, the apparatus comprising: The acquisition unit is used for acquiring multi-source data of the photovoltaic device to be detected, wherein the multi-source data comprises at least two of image data, spectrum data, initial photoelectric data and an environmental stress parameter sequence; The system comprises a processing unit, a multi-mode deep neural network model, a performance attenuation curve and a performance optimization factor, wherein the processing unit is used for respectively executing mode adaptation pretreatment on multi-source data to obtain a multi-mode sample set, inputting a training sample set in the multi-mode sample set into the trained multi-mode deep neural network model, outputting a high-dimensional characteristic vector representing the state of a comprehensive device, the multi-mode deep neural network model comprises mode sub-networks and characteristic fusion layers which are arranged in parallel, each mode sub-network is used for carrying out characteristic extraction on the training sample set, each characteristic fusion layer is used for carrying out fusion on the characteristics extracted by each mode sub-network to obtain the high-dimensional characteristic vector, inputting the high-dimensional characteristic vector into a time sequence prediction sub-model to obtain a photoelectric conversion efficiency PCE predicted value of each time node in the whole life cycle of the photovoltaic device to be tested, drawing the performance attenuation curve, generating the stability rating of the photovoltaic device to be tested according to the performance attenuation curve, wherein the stability rating comprises a stable state and an unstable state, the stability rating is detected to be the unstable state, a key factor influencing the stability of the device is determined according to the performance attenuation curve and the multi-source data, and the key factor influencing the stability is generated according to the key factor, the key factor and the factor optimizing factor is generated according to the environment.
9. A server comprising a processor and a memory, the memory having stored therein a computer program, the processor executing the step instructions of the method of any of claims 1-7 when the computer program in the memory is invoked by the processor.
10. A computer-readable storage medium, on which a computer program/instruction is stored which, when executed by a processor, carries out the steps of the method according to any one of claims 1-7.

Description

Photovoltaic device stability prediction method and related device based on multi-mode model Technical Field The application belongs to the technical field of photovoltaics, and particularly relates to a photovoltaic device stability prediction method and a related device based on a multi-mode model. Background Photovoltaic devices (e.g., perovskite cells) have high photoelectric conversion efficiency, but their efficiency is easily attenuated and stability is poor. How to predict and delay the efficiency decay of perovskite batteries and improve the operation stability of devices is a key technical problem to be solved in the current field. In the prior art, the reliability of the photovoltaic device is researched, the degradation of the device is mostly evaluated and predicted by adopting single-mode data or means, the full-chain change of the perovskite battery from the material structural state, the photoelectric performance and the final failure cannot be completely described, and the complex rule of efficiency attenuation caused by the interaction of a plurality of attenuation factors is difficult to accurately capture. Disclosure of Invention The application provides a photovoltaic device stability prediction method and a related device based on a multi-mode model, which realize the accurate pre-judgment and stability rating of the full life cycle performance of a photovoltaic device through multi-mode data fusion and time sequence prediction technology, can locate core influencing factors in an unstable state and generate targeted optimization suggestions, greatly shorten the stability evaluation period, reduce the test cost and provide high-efficiency technical support for device research and development and mass production optimization. In a first aspect, an embodiment of the present application provides a method for predicting stability of a photovoltaic device based on a multi-mode model, where the method includes collecting multi-source data of the photovoltaic device to be measured, where the multi-source data includes at least two of image data, spectrum data, initial photoelectric data, and an environmental stress parameter sequence; the method comprises the steps of respectively executing modal adaptation pretreatment on multi-source data to obtain a multi-modal sample set, inputting a training sample set in the multi-modal sample set into a trained multi-modal deep neural network model, outputting a high-dimensional feature vector representing the state of a comprehensive device, wherein the multi-modal deep neural network model comprises modal subnetworks and feature fusion layers which are arranged in parallel, each modal subnetwork is used for carrying out feature extraction on the training sample set, each feature fusion layer is used for carrying out fusion on features extracted by each modal subnetwork to obtain the high-dimensional feature vector, inputting the high-dimensional feature vector into a time sequence prediction sub-model to obtain a photoelectric conversion efficiency PCE predicted value of each time node in the whole life cycle of the photovoltaic device to be tested, drawing a performance attenuation curve, generating a stability rating of the photovoltaic device to be tested according to the performance attenuation curve, wherein the stability rating comprises a stable state and an unstable state, detecting the stability rating is the unstable state, determining key factors influencing the stability of the device according to the performance attenuation curve and the multi-source data, the key factors comprise a defect area, an environmental impact factor and an improvement factor according to the performance attenuation curve and an optimization spectral factor. In a second aspect, an embodiment of the present application provides a photovoltaic device stability prediction apparatus based on a multi-mode model, where the apparatus includes an acquisition unit, configured to acquire multi-source data of a photovoltaic device to be tested, where the multi-source data includes at least two of image data, spectrum data, initial photoelectric data, and an environmental stress parameter sequence; the system comprises a processing unit, a multi-mode deep neural network model, a performance attenuation curve and a performance optimization factor, wherein the processing unit is used for respectively executing mode adaptation pretreatment on multi-source data to obtain a multi-mode sample set, inputting a training sample set in the multi-mode sample set into the trained multi-mode deep neural network model, outputting a high-dimensional characteristic vector representing the state of a comprehensive device, the multi-mode deep neural network model comprises mode sub-networks and characteristic fusion layers which are arranged in parallel, each mode sub-network is used for carrying out characteristic extraction on the training sample set, each characteristic fusion layer is us