Multi-DNN technology non-inductive shunting method based on photovoltaic industry
Through the multi-DNN technology inductive shunt method, the in-depth analysis and prediction of photovoltaic data is solved, and the problem that traditional data processing methods are difficult to adapt to complex data characteristics is achieved, efficient and intelligent data shunt and processing is achieved, and data value utilization and system stability are improved.
Patent Information
- Application Number
- CN202411933902.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-04-18
AI Technical Summary
Traditional photovoltaic data processing methods are difficult to adapt to massive, multi-source, heterogeneous data characteristics with complex spatial and temporal correlations, resulting in insufficient utilization of data value and insufficient accuracy and timeliness of power generation efficiency prediction and equipment fault diagnosis.
The multi-DNN technology inductive shunt method is adopted to collect and integrate photovoltaic equipment data, extract representative features using feature engineering, build high-dimensional feature vectors, and establish multiple DNN prediction models, including power generation performance prediction models and equipment health monitoring models, for in-depth analysis and prediction.
It realizes efficient, intelligent diversion and processing of photovoltaic data, improves the accuracy of power generation performance prediction and the accuracy of equipment health monitoring, enhances the utilization of data value, and ensures the stability and real-time nature of the system.
Smart Images
Figure CN119848506A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of photovoltaic technology, and in particular to a multi-DNN technology non-sensing shunting method based on the photovoltaic industry. Background Art
[0002] As the photovoltaic industry is booming, the amount of data generated by photovoltaic power stations is growing explosively. These data range from the photovoltaic panel's photoelectric conversion efficiency, current and voltage characteristics, to meteorological data such as ambient light intensity, temperature, humidity, wind direction, and operating parameters and status data of equipment such as inverters and transformers. Traditional data processing methods have exposed many defects when faced with such massive, multi-source, heterogeneous and complex spatiotemporal correlation data. For example, the rule-based diversion method is difficult to adapt to the dynamic changes of data characteristics, and the hard-coded diversion logic lacks flexibility and cannot be automatically optimized according to the actual data situation; a single data analysis model cannot fully mine the information of different levels and dimensions in photovoltaic data, resulting in insufficient utilization of data value, and greatly reducing the accuracy and timeliness when performing tasks such as power generation efficiency prediction and equipment fault diagnosis. Therefore, how to establish an innovative, efficient and intelligent data diversion and processing method to meet the growing data management needs of the photovoltaic industry is a technical problem that needs to be solved at this stage. Summary of the invention
[0003] The purpose of the present invention is to solve the above problems and design a multi-DNN technology non-sensing shunting method based on the photovoltaic industry.
[0004] To achieve the above purpose, the technical solution of the present invention is that, further, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the multi-DNN technology non-sensing shunting method includes the following steps:
[0005] Collecting photovoltaic equipment data, integrating the photovoltaic equipment data, classifying and storing the data according to data type, collection time, and equipment number, and establishing a photovoltaic equipment data set;
[0006] For different types of data in the photovoltaic equipment dataset, using feature engineering technology to extract representative and distinguishing features to obtain a feature photovoltaic equipment dataset;
[0007] The features of the characteristic photovoltaic equipment data set are integrated and normalized to construct a high-dimensional feature vector, each of which represents the comprehensive data features of the photovoltaic power station at a specific moment, thereby obtaining a vector photovoltaic equipment data set;
[0008] Establishing a plurality of DNN prediction models, wherein the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model;
[0009] The vector photovoltaic device data set is input into an intelligent shunt controller, and the controller is used to push the feature vectors in parallel to each trained DNN model for analysis to obtain analysis results.
[0010] Furthermore, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, photovoltaic equipment data is collected, the photovoltaic equipment data is integrated, and classified and stored according to data type, collection time, and equipment number to establish a photovoltaic equipment data set, including:
[0011] Collect photovoltaic equipment data, which includes at least the output power, voltage, and current curves of photovoltaic cells, real-time monitoring data of solar irradiance, ambient temperature, and humidity, as well as operating status data of various electrical equipment in the power station.
[0012] Furthermore, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the representative and distinguishing features of different types of data in the photovoltaic equipment data set are extracted using feature engineering technology to obtain a feature photovoltaic equipment data set, including:
[0013] Extract the slope characteristics of the power-time curve in the photovoltaic panel data to characterize the changing trend of power generation efficiency and obtain the photovoltaic panel characteristic data;
[0014] Calculate the voltage-current ratio characteristics under different light intensities to reflect the electrical characteristics of the panel and obtain electrical characteristic data;
[0015] Construct the temperature-humidity coupling characteristics and light intensity fluctuation characteristics in meteorological data to obtain meteorological characteristic data;
[0016] The frequency spectrum characteristics of the equipment operation sound and the temperature change rate characteristics of key components of the equipment data are extracted to obtain the equipment characteristic data.
[0017] Furthermore, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the features of the characteristic photovoltaic equipment data set are fused and normalized to construct a high-dimensional feature vector, each feature vector represents the comprehensive data features of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained, including:
[0018] The features of different sources and types in the characteristic photovoltaic equipment dataset are fused to obtain a fused dataset;
[0019] A weighted summation method is used to assign corresponding weight coefficients to features of different importance and influence in the fused data set;
[0020] The fused data set is normalized using the min-max normalization method to obtain a normalized data set;
[0021] A high-dimensional feature vector is constructed, each of which represents the comprehensive data characteristics of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained.
[0022] Furthermore, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, multiple DNN prediction models are established, and the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model, including:
[0023] Establish a DNN power generation performance prediction model, use the input layer of the DNN power generation performance prediction model to receive the photovoltaic panel data and meteorological data related feature vectors, perform deep mining on the data through multiple hidden layers, and use the output layer to predict the power generation curve in the future period;
[0024] Establish a DNN device health monitoring model, input the device operation data feature vector into the DNN device health monitoring model, and output the device failure probability and possible failure type;
[0025] Establishing a system optimization suggestion DNN model, using the system optimization suggestion DNN model to integrate all types of data feature vectors, and outputting optimization strategy suggestions for the overall operation of the photovoltaic power station;
[0026] The historical data of photovoltaic power plants are used to train each DNN model, and the weights and bias parameters of the model are optimized based on adaptive learning rate adjustment and regularization technology to prevent overfitting.
[0027] Furthermore, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the vector photovoltaic device data set is input into the intelligent shunting controller, and the controller is used to push the feature vector in parallel to each trained DNN model for analysis to obtain the analysis results, including:
[0028] The DNN power generation performance prediction model outputs the predicted power generation curve and the corresponding confidence level, and the DNN equipment health monitoring model outputs the equipment failure probability and confidence level;
[0029] The intelligent diversion controller makes diversion decisions based on the confidence of each DNN model and the predefined dynamic diversion strategy.
[0030] Furthermore, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the intelligent shunting controller makes shunting decisions based on the confidence of each DNN model and a predefined dynamic shunting strategy, including:
[0031] If the confidence of the power generation performance prediction DNN model is higher than the set threshold and the predicted power generation will fluctuate significantly, the controller will divert the data to the power generation optimization adjustment module.
[0032] Furthermore, in the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the intelligent shunting controller makes shunting decisions based on the confidence of each DNN model and a predefined dynamic shunting strategy, and also includes:
[0033] If the confidence of the equipment health monitoring DNN model is high and indicates that the equipment has a potential failure risk, the data will be directed to the equipment fault diagnosis and maintenance system.
[0034] Furthermore, in a system for implementing the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the system includes the following modules:
[0035] A data collection module is used to collect photovoltaic equipment data, integrate the photovoltaic equipment data, classify and store the data according to data type, collection time, and equipment number, and establish a photovoltaic equipment data set;
[0036] A feature extraction module, used to extract representative and distinguishing features from different types of data in the photovoltaic device data set by using feature engineering technology to obtain a feature photovoltaic device data set;
[0037] A vector conversion module is used to fuse and normalize the features of the characteristic photovoltaic equipment data set to construct a high-dimensional feature vector, each of which represents the comprehensive data features of the photovoltaic power station at a specific moment, to obtain a vector photovoltaic equipment data set;
[0038] A model building module, used to build multiple DNN prediction models, wherein the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model;
[0039] The result analysis module is used to input the vector photovoltaic device data set into the intelligent shunt controller, and use the controller to push the feature vector in parallel to each trained DNN model for analysis to obtain the analysis result.
[0040] Furthermore, in a system for implementing the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the system includes the following submodules:
[0041] The extraction submodule is used to extract the slope characteristics of the power-time curve in the photovoltaic panel data to characterize the change trend of power generation efficiency and obtain photovoltaic panel characteristic data;
[0042] A calculation submodule is used to calculate the voltage-current ratio characteristics under different light intensities to reflect the electrical characteristics of the panel and obtain electrical characteristic data;
[0043] Construct submodules to construct temperature-humidity coupling characteristics and light intensity fluctuation characteristics in meteorological data to obtain meteorological characteristic data;
[0044] A submodule is obtained, which is used to extract the frequency spectrum characteristics of the equipment operation sound and the temperature change rate characteristics of key components of the equipment data to obtain the equipment characteristic data.
[0045] Furthermore, in a system for implementing the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the system includes the following submodules:
[0046] A fusion submodule is used to fuse the features of different sources and types in the characteristic photovoltaic equipment data set to obtain a fused data set;
[0047] A weighting submodule, used for assigning corresponding weight coefficients to features of different importance and influence in the fused data set by using a weighted summation method;
[0048] The normalization submodule is used to normalize the fused data set using the min-max normalization method to obtain a normalized data set;
[0049] The submodule is obtained to construct a high-dimensional feature vector, each feature vector represents the comprehensive data characteristics of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained.
[0050] Furthermore, in a system for implementing the above-mentioned multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the system includes the following submodules:
[0051] The first submodule is used to establish a DNN power generation performance prediction model, using the input layer of the DNN power generation performance prediction model to receive photovoltaic panel data and meteorological data related feature vectors, deeply mine the data through multiple hidden layers, and the output layer predicts the power generation curve in the future period;
[0052] The second establishment submodule is used to establish a DNN device health monitoring model, input the device operation data feature vector into the DNN device health monitoring model, and output the device failure probability and possible failure type;
[0053] The third submodule is used to establish a system optimization suggestion DNN model, and use the system optimization suggestion DNN model to integrate all types of data feature vectors to output optimization strategy suggestions for the overall operation of the photovoltaic power station;
[0054] The optimization submodule is used to train each DNN model using historical data of photovoltaic power plants, and optimize the weight and bias parameters of the model based on adaptive learning rate adjustment and regularization technology to prevent overfitting.
[0055] Its beneficial effects are that, by collecting photovoltaic equipment data, the photovoltaic equipment data is integrated, classified and stored according to data type, collection time, and equipment number, and a photovoltaic equipment data set is established; for different types of data in the photovoltaic equipment data set, representative and distinguishing features are extracted using feature engineering technology to obtain a feature photovoltaic equipment data set; the features of the feature photovoltaic equipment data set are fused and normalized to construct a high-dimensional feature vector, each feature vector represents the comprehensive data features of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained; multiple DNN prediction models are established, and the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model; the vector photovoltaic equipment data set is input into an intelligent shunt controller, and the controller is used to push the feature vectors in parallel to each trained DNN model for analysis to obtain analysis results. 1. High intelligence and accuracy: Through multiple specially designed DNN models, in-depth analysis of photovoltaic data can accurately mine the complex relationships and potential values behind the data, and can provide highly accurate results for power generation performance prediction, equipment health monitoring, and system optimization recommendations. 2. Efficient and non-inductive shunting: The intelligent shunting controller combines advanced parallel processing, caching mechanism and intelligent scheduling algorithm to achieve efficient data shunting, and has almost no impact on the normal operation of the photovoltaic power station during the shunting process, ensuring the real-time data processing and system stability. 3. Dynamic adaptability: Based on the feedback update mechanism of the DNN model, it can continuously adjust and optimize the model according to the actual situation during the operation of the photovoltaic power station, adapt to the dynamic changes of data characteristics, equipment aging, environmental changes and other factors, always maintain the efficiency and accuracy of data processing, and extend the service life and effectiveness of the entire data processing system. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.
[0057] Figure 1 This is a schematic diagram of a first embodiment of a multi-DNN technology-based photovoltaic industry non-sensing shunting method in an embodiment of the present invention;
[0058] Figure 2 It is a schematic diagram of a second embodiment of a multi-DNN technology non-sensing shunting method based on the photovoltaic industry in an embodiment of the present invention;
[0059] Figure 3 It is a schematic diagram of a third embodiment of a multi-DNN technology non-sensing shunting method based on the photovoltaic industry in an embodiment of the present invention. DETAILED DESCRIPTION
[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0061] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.
[0062] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown, a multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the multi-DNN technology non-sensing shunting method includes the following steps:
[0063] Step 101: Collect photovoltaic equipment data, integrate the photovoltaic equipment data, classify and store them according to data type, collection time, and equipment number, and establish a photovoltaic equipment data set;
[0064] Specifically, in this embodiment, photovoltaic equipment data is collected, and the photovoltaic equipment data at least includes the output power, voltage, and current curves of the photovoltaic cells, real-time monitoring data of solar irradiance, ambient temperature, and humidity, and operating status data of various electrical equipment in the power station.
[0065] Step 102: for different types of data in the photovoltaic equipment data set, extract representative and distinguishing features using feature engineering technology to obtain a feature photovoltaic equipment data set;
[0066] Specifically, in this embodiment, the slope characteristics of the power-time curve in the photovoltaic panel data are extracted to characterize the change trend of power generation efficiency, and the photovoltaic panel characteristic data is obtained;
[0067] Calculate the voltage-current ratio characteristics under different light intensities to reflect the electrical characteristics of the panel and obtain electrical characteristic data;
[0068] Construct the temperature-humidity coupling characteristics and light intensity fluctuation characteristics in meteorological data to obtain meteorological characteristic data;
[0069] The frequency spectrum characteristics of the equipment operation sound and the temperature change rate characteristics of key components of the equipment data are extracted to obtain the equipment characteristic data.
[0070] Step 103: fuse and normalize the features of the characteristic photovoltaic equipment data set to construct a high-dimensional feature vector, each of which represents the comprehensive data features of the photovoltaic power station at a specific moment, to obtain a vector photovoltaic equipment data set;
[0071] Specifically, in this embodiment, features of different sources and types in the characteristic photovoltaic equipment data set are fused to obtain a fused data set;
[0072] The weighted summation method is used to assign corresponding weight coefficients to the features of different importance and influence in the fusion data set;
[0073] The fused data set is normalized using the min-max normalization method to obtain a normalized data set;
[0074] A high-dimensional feature vector is constructed, each of which represents the comprehensive data characteristics of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained.
[0075] The features of the characteristic photovoltaic equipment data set are fused and normalized to construct a high-dimensional feature vector. Each feature vector represents the comprehensive data features of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained. Specifically, the fusion operation is first carried out on the features of different sources and types in the characteristic photovoltaic equipment data set. For example, for the features that reflect the power generation performance of photovoltaic panels, such as the output power value and power conversion efficiency under different light intensities, the features related to the environment, such as the real-time light intensity, ambient temperature, humidity, wind direction and wind speed and other meteorological data features, and the operating status characteristics of various equipment in the photovoltaic power station, including the operating frequency of the inverter, the load rate of the transformer, the current stability of the junction box, etc., they are organically integrated according to their internal logical relationships.
[0076] In the fusion process, a weighted summation method is used to assign corresponding weight coefficients to features of different importance and influence. For example, higher weights are assigned to features of light intensity and photovoltaic panel temperature that have a direct impact on power generation efficiency, while relatively lower weights are assigned to some auxiliary features, such as wind direction and wind speed, which have little impact on power generation in specific regions and seasons. Through such weighted summation, a fusion feature set is initially constructed.
[0077] Next, the fused feature set is normalized. Common normalization methods, such as min-max normalization, are used to map the value range of each feature to a specific interval, usually the [0,1] interval. For a certain feature value x, its normalized calculation method is: (x-min(x)) / (max(x)-min(x)), where min(x) and max(x) are the minimum and maximum values of the feature in the entire data set, respectively. Through normalization, the adverse effects of different features on subsequent data analysis and model training due to large differences in dimensions and numerical ranges can be eliminated, so that all features can be compared and calculated on the same numerical scale.
[0078] After the above fusion and normalization processing, these processed features are combined in order to construct a high-dimensional feature vector. Each high-dimensional feature vector completely covers the comprehensive data feature information of the photovoltaic power station at a specific moment, from power generation equipment to environmental conditions to the operating status of supporting equipment.
[0079] Step 104: Establish multiple DNN prediction models, where the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model;
[0080] Specifically, in this embodiment, a DNN power generation performance prediction model is established, and the input layer of the DNN power generation performance prediction model is used to receive the photovoltaic panel data and the feature vectors related to the meteorological data, and the data is deeply mined through multiple hidden layers, and the output layer predicts the power generation curve in the future period;
[0081] Establish a DNN device health monitoring model, input the device operation data feature vector into the DNN device health monitoring model, and output the device failure probability and possible failure type;
[0082] Establish a system optimization suggestion DNN model, use the system optimization suggestion DNN model to integrate all types of data feature vectors, and output optimization strategy suggestions for the overall operation of the photovoltaic power station;
[0083] The historical data of photovoltaic power plants are used to train each DNN model, and the weights and bias parameters of the model are optimized based on adaptive learning rate adjustment and regularization technology to prevent overfitting.
[0084] Step 105: input the vector photovoltaic device data set into the intelligent shunt controller, and use the controller to push the feature vectors in parallel to each trained DNN model for analysis to obtain the analysis results.
[0085] Specifically, in this embodiment, the DNN power generation performance prediction model outputs the predicted power generation curve and the corresponding confidence level, and the DNN equipment health monitoring model outputs the equipment failure probability and confidence level;
[0086] The intelligent diversion controller makes diversion decisions based on the confidence of each DNN model and the predefined dynamic diversion strategy;
[0087] If the confidence of the power generation performance prediction DNN model is higher than the set threshold and the predicted power generation will fluctuate significantly, the controller will divert the data to the power generation optimization adjustment module;
[0088] If the confidence of the equipment health monitoring DNN model is high and indicates that the equipment has a potential failure risk, the data will be directed to the equipment fault diagnosis and maintenance system.
[0089] For example, the present invention is implemented in a large centralized photovoltaic power station located in a desert area. First, an advanced sensor network is installed in various areas of the photovoltaic power station, including high-precision photovoltaic power sensors, light intensity sensors, temperature sensors, humidity sensors, and special monitoring sensors for inverters, transformers and other equipment. The collected data is transmitted to the server cluster of the data center through a high-speed optical fiber network to build a raw data pool.
[0090] Perform feature engineering on the data in the original data pool. For example, for photovoltaic panel data, analyze the changing characteristics of the power-time curve in different time periods, extract its slope change characteristics as the trend characteristics of power generation efficiency, and calculate the voltage-current ratio characteristics under different light intensities; for meteorological data, construct the joint distribution characteristics of temperature-humidity and the fluctuation frequency characteristics of light intensity; for equipment data, extract the frequency spectrum characteristics of the inverter operation sound and the change rate characteristics of the transformer oil temperature, etc. After normalizing these features, construct a high-dimensional feature vector.
[0091] Three main DNN models were constructed: power generation performance prediction DNN model, equipment health monitoring DNN model and system optimization recommendation DNN model. These models were trained using the historical data of the photovoltaic power station over the past three years. During the training process, adaptive learning rate adjustment technology was used to dynamically adjust the learning rate according to the model convergence situation, and L2 regularization was used to prevent overfitting.
[0092] In the intelligent non-sensing shunt stage, the feature vector constructed in real time is input into the intelligent shunt controller. For example, when the power generation performance prediction DNN model outputs the power generation curve for the next 24 hours and the confidence reaches 0.85 (the threshold is set to 0.8), and the predicted power will drop by 15%, the intelligent shunt controller will shunt the data to the power generation optimization adjustment module. This module adjusts the tilt angle of the photovoltaic panel according to the data, increases the light receiving area, and optimizes the power factor setting of the inverter to improve the power generation efficiency. At the same time, if the equipment health monitoring DNN model outputs that the probability of failure of a certain inverter is 0.6 (the fault warning threshold is set to 0.5) and the confidence is 0.9, the data will be directed to the equipment fault diagnosis and maintenance system, and the technicians will conduct in-depth troubleshooting and maintenance work based on the data.
[0093] Each processing module feeds back the processing results and newly generated data to the corresponding DNN model. For example, the power generation optimization adjustment module feeds back the adjusted power generation data, panel angle adjustment data, etc. to the power generation performance prediction DNN model, and the model uses incremental learning technology to update itself to adapt to changes in the operating status of the photovoltaic power station.
[0094] Its beneficial effects are that by collecting photovoltaic equipment data, integrating the photovoltaic equipment data, classifying and storing them according to data type, collection time, and equipment number, a photovoltaic equipment data set is established; for different types of data in the photovoltaic equipment data set, representative and distinguishing features are extracted using feature engineering technology to obtain a feature photovoltaic equipment data set; the features of the feature photovoltaic equipment data set are fused and normalized to construct a high-dimensional feature vector, each feature vector represents the comprehensive data features of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained; multiple DNN prediction models are established, and the DNN prediction model includes at least a DNN power generation performance prediction model and a DNN equipment health monitoring model; the vector photovoltaic equipment data set is input into the intelligent shunt controller, and the controller is used to push the feature vectors in parallel to each trained DNN model for analysis to obtain the analysis results. 1. High intelligence and accuracy: Through multiple specially designed DNN models, in-depth analysis of photovoltaic data can accurately mine the complex relationships and potential values behind the data, and can provide highly accurate results for power generation performance prediction, equipment health monitoring, and system optimization recommendations. 2. Efficient and non-inductive shunting: The intelligent shunting controller combines advanced parallel processing, caching mechanism and intelligent scheduling algorithm to achieve efficient data shunting, and has almost no impact on the normal operation of the photovoltaic power station during the shunting process, ensuring the real-time data processing and system stability. 3. Dynamic adaptability: Based on the feedback update mechanism of the DNN model, it can continuously adjust and optimize the model according to the actual situation during the operation of the photovoltaic power station, adapt to the dynamic changes of data characteristics, equipment aging, environmental changes and other factors, always maintain the efficiency and accuracy of data processing, and extend the service life and effectiveness of the entire data processing system.
[0095] See also Figure 2 In a multi-DNN technology non-sensing shunting method based on the photovoltaic industry, the representative and discriminative features of different types of data in the photovoltaic equipment data set are extracted using feature engineering technology to obtain the characteristic photovoltaic equipment data set, which includes the following steps:
[0096] Step 201, extracting the slope characteristics of the power-time curve in the photovoltaic panel data to characterize the change trend of power generation efficiency, and obtaining photovoltaic panel characteristic data;
[0097] Step 202, calculating the voltage-to-current ratio characteristics under different light intensities to reflect the electrical characteristics of the panel and obtain electrical characteristic data;
[0098] Step 203: construct temperature-humidity coupling characteristics and light intensity fluctuation characteristics in meteorological data to obtain meteorological characteristic data;
[0099] Step 204 , extracting the frequency spectrum characteristics of the equipment operation sound and the temperature change rate characteristics of key components from the equipment data to obtain equipment characteristic data.
[0100] See also Figure 3 ,In a multi-DNN technology senseless shunting method based on the photovoltaic industry, the features of the ,feature photovoltaic equipment data set are fused and normalized to construct a high-dimensional ,feature vector. Each feature vector represents the comprehensive data ,feature of a photovoltaic power station at a specific moment. ,Getting the vector photovoltaic equipment data set includes the following steps:
[0101] Step 301, fusing features of different sources and types in a characteristic photovoltaic device data set to obtain a fused data set;
[0102] Step 302: using a weighted summation method to assign corresponding weight coefficients to features of different importance and influence in the fused data set;
[0103] Step 303: normalize the fused data set using the min-max normalization method to obtain a normalized data set;
[0104] Step 304: construct a high-dimensional feature vector, each feature vector represents the comprehensive data feature of the photovoltaic power station at a specific moment, and obtain a vector photovoltaic equipment data set.
[0105] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A multi-DNN technology non-sensing shunting method based on the photovoltaic industry, characterized in that: The multi-DNN technology non-sensing shunting method comprises the following steps: Collecting photovoltaic equipment data, integrating the photovoltaic equipment data, classifying and storing the data according to data type, collection time, and equipment number, and establishing a photovoltaic equipment data set; For different types of data in the photovoltaic equipment dataset, using feature engineering technology to extract representative and distinguishing features to obtain a feature photovoltaic equipment dataset; The features of the characteristic photovoltaic equipment data set are integrated and normalized to construct a high-dimensional feature vector, each of which represents the comprehensive data features of the photovoltaic power station at a specific moment, thereby obtaining a vector photovoltaic equipment data set; Establishing a plurality of DNN prediction models, wherein the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model; The vector photovoltaic device data set is input into an intelligent shunt controller, and the controller is used to push the feature vectors in parallel to each trained DNN model for analysis to obtain analysis results.
2. The multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1 is characterized in that: The photovoltaic equipment data is collected, the photovoltaic equipment data is integrated, and the photovoltaic equipment data is classified and stored according to the data type, collection time, and equipment number to establish a photovoltaic equipment data set, including: Collect photovoltaic equipment data, which includes at least the output power, voltage, and current curves of photovoltaic cells, real-time monitoring data of solar irradiance, ambient temperature, and humidity, as well as operating status data of various electrical equipment in the power station.
3. The multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1 is characterized in that: The method of extracting representative and distinguishing features from different types of data in the photovoltaic equipment dataset using feature engineering technology to obtain a feature photovoltaic equipment dataset includes: Extract the slope characteristics of the power-time curve in the photovoltaic panel data to characterize the changing trend of power generation efficiency and obtain the photovoltaic panel characteristic data; Calculate the voltage-current ratio characteristics under different light intensities to reflect the electrical characteristics of the panel and obtain electrical characteristic data; Construct the temperature-humidity coupling characteristics and light intensity fluctuation characteristics in meteorological data to obtain meteorological characteristic data; The frequency spectrum characteristics of the equipment operation sound and the temperature change rate characteristics of key components of the equipment data are extracted to obtain the equipment characteristic data.
4. The multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1 is characterized in that: The features of the characteristic photovoltaic equipment data set are fused and normalized to construct a high-dimensional feature vector, each of which represents the comprehensive data features of the photovoltaic power station at a specific moment, to obtain a vector photovoltaic equipment data set, including: The features of different sources and types in the characteristic photovoltaic equipment dataset are fused to obtain a fused dataset; A weighted summation method is used to assign corresponding weight coefficients to features of different importance and influence in the fused data set; The fused data set is normalized using the min-max normalization method to obtain a normalized data set; A high-dimensional feature vector is constructed, each of which represents the comprehensive data characteristics of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained.
5. The multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1, characterized in that: The establishing of multiple DNN prediction models, wherein the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model, includes: Establish a DNN power generation performance prediction model, use the input layer of the DNN power generation performance prediction model to receive the photovoltaic panel data and meteorological data related feature vectors, perform deep mining on the data through multiple hidden layers, and use the output layer to predict the power generation curve in the future period; Establish a DNN device health monitoring model, input the device operation data feature vector into the DNN device health monitoring model, and output the device failure probability and possible failure type; Establishing a system optimization suggestion DNN model, using the system optimization suggestion DNN model to integrate all types of data feature vectors, and outputting optimization strategy suggestions for the overall operation of the photovoltaic power station; The historical data of photovoltaic power plants are used to train each DNN model, and the weights and bias parameters of the model are optimized based on adaptive learning rate adjustment and regularization technology to prevent overfitting.
6. The multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1, characterized in that: The vector photovoltaic device data set is input into the intelligent shunt controller, and the controller is used to push the feature vectors to each trained DNN model in parallel for analysis to obtain analysis results, including: The DNN power generation performance prediction model outputs the predicted power generation curve and the corresponding confidence level, and the DNN equipment health monitoring model outputs the equipment failure probability and confidence level; The intelligent diversion controller makes diversion decisions based on the confidence of each DNN model and the predefined dynamic diversion strategy.
7. The multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 6, characterized in that: The intelligent diversion controller makes diversion decisions based on the confidence of each DNN model and a predefined dynamic diversion strategy, including: If the confidence of the power generation performance prediction DNN model is higher than the set threshold and the predicted power generation will fluctuate significantly, the controller will divert the data to the power generation optimization adjustment module.
8. The multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 6, characterized in that: The intelligent diversion controller makes diversion decisions based on the confidence of each DNN model and a predefined dynamic diversion strategy, and also includes: If the confidence of the equipment health monitoring DNN model is high and indicates that the equipment has a potential failure risk, the data will be directed to the equipment fault diagnosis and maintenance system.
9. A system for implementing a multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1, characterized in that: The system includes the following modules: A data collection module is used to collect photovoltaic equipment data, integrate the photovoltaic equipment data, classify and store the data according to data type, collection time, and equipment number, and establish a photovoltaic equipment data set; A feature extraction module, used to extract representative and distinguishing features from different types of data in the photovoltaic device data set by using feature engineering technology to obtain a feature photovoltaic device data set; A vector conversion module is used to fuse and normalize the features of the characteristic photovoltaic equipment data set to construct a high-dimensional feature vector, each of which represents the comprehensive data features of the photovoltaic power station at a specific moment, to obtain a vector photovoltaic equipment data set; A model building module, used to build multiple DNN prediction models, wherein the DNN prediction models at least include a DNN power generation performance prediction model and a DNN equipment health monitoring model; The result analysis module is used to input the vector photovoltaic device data set into the intelligent shunt controller, and use the controller to push the feature vector in parallel to each trained DNN model for analysis to obtain the analysis result.
10. A system for implementing a multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1, characterized in that: The system includes the following submodules: The extraction submodule is used to extract the slope characteristics of the power-time curve in the photovoltaic panel data to characterize the change trend of power generation efficiency and obtain photovoltaic panel characteristic data; A calculation submodule is used to calculate the voltage-current ratio characteristics under different light intensities to reflect the electrical characteristics of the panel and obtain electrical characteristic data; Construct submodules to construct temperature-humidity coupling characteristics and light intensity fluctuation characteristics in meteorological data to obtain meteorological characteristic data; A submodule is obtained, which is used to extract the frequency spectrum characteristics of the equipment operation sound and the temperature change rate characteristics of key components of the equipment data to obtain the equipment characteristic data.
11. A system for implementing a multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1, characterized in that: The system includes the following submodules: A fusion submodule is used to fuse the features of different sources and types in the characteristic photovoltaic equipment data set to obtain a fused data set; A weighting submodule, used for assigning corresponding weight coefficients to features of different importance and influence in the fused data set by using a weighted summation method; The normalization submodule is used to normalize the fused data set using the min-max normalization method to obtain a normalized data set; The submodule is obtained to construct a high-dimensional feature vector, each feature vector represents the comprehensive data characteristics of the photovoltaic power station at a specific moment, and a vector photovoltaic equipment data set is obtained.
12. A system for implementing a multi-DNN technology non-sensing shunting method based on the photovoltaic industry as claimed in claim 1, characterized in that: The system includes the following submodules: The first submodule is used to establish a DNN power generation performance prediction model, using the input layer of the DNN power generation performance prediction model to receive photovoltaic panel data and meteorological data related feature vectors, deeply mine the data through multiple hidden layers, and the output layer predicts the power generation curve in the future period; The second establishment submodule is used to establish a DNN device health monitoring model, input the device operation data feature vector into the DNN device health monitoring model, and output the device failure probability and possible failure type; The third submodule is used to establish a system optimization suggestion DNN model, and use the system optimization suggestion DNN model to integrate all types of data feature vectors to output optimization strategy suggestions for the overall operation of the photovoltaic power station; The optimization submodule is used to train each DNN model using historical data of photovoltaic power plants, and optimize the weight and bias parameters of the model based on adaptive learning rate adjustment and regularization technology to prevent overfitting.