Distributed photovoltaic performance evaluation and prediction method and device based on edge intelligence
By collecting and processing multi-source data in real time through edge intelligent architecture, the problem of insufficient prediction accuracy of photovoltaic systems is solved, and efficient and accurate photovoltaic performance evaluation and prediction are achieved.
Patent Information
- Application Number
- CN202510722927.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-30
- Publication Date
- 2025-09-05
AI Technical Summary
The existing photovoltaic system power generation efficiency prediction has problems such as large regional differences, data processing delays and lack of multi-dimensional data integration, resulting in insufficient prediction accuracy.
It adopts edge intelligent architecture to synchronously collect meteorological, spectral and photovoltaic multi-source data in real time, uses pre-trained machine learning models to evaluate distributed photovoltaic performance, integrates meteorological measurement modules, spectral measurement modules and photovoltaic measurement modules, and uses Jetson Orin Nano as the edge computing host for data processing and prediction.
It improves the accuracy and flexibility of photovoltaic power generation prediction, adapts to various environmental conditions, realizes fully automatic data collection and real-time analysis, and improves prediction accuracy and efficiency.
Smart Images
Figure CN120601403A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic performance prediction, and in particular to a distributed photovoltaic performance evaluation and prediction method and device based on edge intelligence. Background Art
[0002] As the global energy structure transforms towards low-carbon and intelligent development, photovoltaic power generation technology has become an important part of renewable energy utilization. However, the power generation efficiency of photovoltaic systems is affected by a variety of environmental factors, including meteorological conditions (temperature, humidity, wind speed, solar irradiance, etc.) and solar spectrum characteristics, resulting in significant deviations between actual power generation performance and theoretical models. Therefore, how to accurately monitor the operating status of photovoltaic systems and use real-time data to optimize power forecasting is an important research direction in the current photovoltaic field. Currently, most studies rely on official meteorological data sets (such as Huld et al., 2013) rather than on-site data collection. This data source has the following problems:
[0003] 1. Errors introduced by regional differences: Official meteorological data usually comes from fixed sites and cannot reflect the specific environmental conditions at the location where the photovoltaic panels are installed, resulting in insufficient data accuracy.
[0004] 2. Data processing delay: Traditional data acquisition systems usually adopt centralized storage and processing methods, which cannot meet the needs of real-time monitoring and analysis.
[0005] In addition, existing systems for photovoltaic power generation efficiency monitoring usually focus on a single data type (such as meteorological data or photovoltaic data) and lack a comprehensive platform that can integrate multi-dimensional data collection. For example:
[0006] Meteorological data acquisition systems: such as integrated weather stations, which can measure parameters such as temperature, humidity, and wind speed, but cannot simultaneously collect spectral and photovoltaic data.
[0007] Photovoltaic data acquisition systems, such as photovoltaic analyzers, can plot the IV characteristic curve of photovoltaic panels, but lack the support of environmental data, making it difficult to fully analyze photovoltaic performance.
[0008] Spectral data acquisition systems, such as spectrometers, can measure solar irradiance and spectral power distribution, but they typically operate independently and cannot be synchronized with other data collection and analysis. Summary of the Invention
[0009] The purpose of the present invention is to provide a distributed photovoltaic performance evaluation and prediction method and device based on edge intelligence to improve the accuracy of photovoltaic prediction.
[0010] The purpose of the present invention can be achieved by the following technical solutions:
[0011] A distributed photovoltaic performance evaluation and prediction method based on edge intelligence includes the following steps:
[0012] Real-time and synchronous collection of multi-source data of distributed photovoltaics, including meteorological measurement parameters, spectral measurement parameters and photovoltaic measurement parameters;
[0013] The meteorological measurement parameters and spectral measurement parameters are input into the pre-trained machine learning model to predict photovoltaic performance, and the distributed photovoltaic performance prediction results are output.
[0014] The multi-source data is used to pre-train the machine learning model, and the prediction method is processed based on the edge architecture.
[0015] Furthermore, the meteorological measurement parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure, light intensity, PM2.5 and PM10.
[0016] Furthermore, the spectrum measurement parameters include spectral power distribution and solar irradiance at visible light wavelengths.
[0017] Furthermore, the photovoltaic measurement parameters include a photovoltaic IV characteristic curve and a maximum power point, wherein the maximum power point is calculated based on the IV characteristic curve.
[0018] Furthermore, the pre-trained machine learning model is one of XGBoost, KRR, MLP, KRR-KRR, and SVM.
[0019] The present invention also provides a device according to the above-mentioned distributed photovoltaic performance evaluation and prediction method based on edge intelligence, comprising an edge computing host and a meteorological measurement module, a spectral measurement module and a photovoltaic measurement module connected thereto.
[0020] Meteorological measurement module: used to collect meteorological measurement parameters in real time;
[0021] Spectral measurement module: used for real-time collection of photovoltaic measurement parameters of distributed photovoltaics;
[0022] Photovoltaic measurement module: used for real-time acquisition of spectral measurement parameters;
[0023] Edge computing host: used to predict photovoltaic performance based on the meteorological measurement parameters, spectral measurement parameters and photovoltaic measurement parameters using a pre-trained machine learning model, and output distributed photovoltaic performance prediction results.
[0024] Furthermore, the spectrum measurement module includes a spectrometer and an external computer, and the spectrum measurement module is used to measure spectral power distribution and solar irradiance.
[0025] Furthermore, the photovoltaic measurement module includes a photovoltaic panel and an electronic load, and the photovoltaic panel measures an IV characteristic curve through the electronic load to calculate the maximum power point.
[0026] Furthermore, the edge computing host is integrated with a time synchronization control module, which is used to control the meteorological measurement module, the spectral measurement module and the photovoltaic measurement module to collect data simultaneously to ensure the alignment of timestamps.
[0027] Furthermore, it also includes an outdoor power supply module, which is respectively connected to the edge computing host, meteorological measurement module, spectral measurement module and photovoltaic measurement module.
[0028] Compared with the prior art, the present invention has the following beneficial effects:
[0029] (1) This invention adopts an edge computing architecture, using Jetson Orin Nano as the edge computing host, and can directly perform data collection, preprocessing, power prediction, and optimization calculations on the photovoltaic end (edge side). Traditional photovoltaic prediction usually transmits data back to a central server for centralized calculation, while this invention completes the analysis locally, fully utilizing the low latency characteristics of edge computing, and is more suitable for the decentralized deployment characteristics of distributed photovoltaic systems, thereby improving the accuracy of photovoltaic prediction.
[0030] (2) Traditional photovoltaic systems only predict power based on environmental parameters such as irradiance, temperature and humidity. The present invention additionally introduces a spectral measurement module, which, combined with a machine learning algorithm, can effectively improve the prediction capability of photovoltaic power generation.
[0031] (3) Traditional photovoltaic monitoring equipment is mostly fixed and lacks flexibility and outdoor adaptability. The present invention integrates the system into a wheeled cart with a waterproof, dustproof, and heat-insulated chassis. The photovoltaic panels are fixed at a 25° south-facing angle. The system can operate under high temperature (>30°C) and light rain (humidity ≤95% RH) conditions.
[0032] (4) Traditional systems rely on manual operation or semi-automatic data collection, which is inefficient and prone to errors. This invention uses Python scripts to achieve fully automatic data collection. The sampling time of each module data can be customized and automatic sampling is performed without manual supervision. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 Schematic diagram of the system structure of the present invention;
[0034] Figure 2 The performance of the spectrum prediction task of the present invention;
[0035] Figure 3 This is a comparison chart of photovoltaic maximum power prediction based on long short-term memory neural network of the present invention. DETAILED DESCRIPTION
[0036] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.
[0037] This embodiment provides a distributed photovoltaic performance evaluation and prediction method based on edge intelligence. The method is implemented based on a prediction system that is used to collect and process meteorological, spectral and photovoltaic performance data in real time, and run a machine learning model to achieve real-time photovoltaic performance prediction. Figure 1 As shown, the dotted line represents the data transmission line and the solid line represents the electric line. The prediction system includes:
[0038] The edge computing host 1 and the meteorological measurement module 2, spectral measurement module 3, photovoltaic measurement module 4 and outdoor power supply module 5 connected to it. The outdoor power supply module 5 is also connected to the meteorological measurement module 2, spectral measurement module 3, and photovoltaic measurement module 4. Each module works together through a wired connection. The specific structure and function are as follows:
[0039] 1) Edge computing host 1
[0040] The Jetson Orin Nano serves as the core processing unit, responsible for data acquisition, storage, and preliminary analysis for meteorological measurement module 2, spectral measurement module 3, and photovoltaic measurement module 4. Equipped with multiple USB and GPIO ports, supporting RS485-to-USB communication, running Python control programs, and equipped with 8GB of memory and a high-efficiency GPU, this host computer meets the demands of edge computing. It can receive and process multi-source data in real time, executing pre-trained machine learning models (such as XGBoost, KRR, MLP, KRR-KRR, and SVM) to predict photovoltaic performance. Furthermore, this edge computing host 1 integrates a time synchronization control module, which coordinates the synchronous measurements of each measurement module through time control signals, ensuring simultaneous data acquisition and time stamp alignment among the three modules.
[0041] During the data storage process, the edge computing host 1 can store multi-source data as CSV files on a daily basis through RS485 and manual synchronization to ensure timestamp alignment and support subsequent analysis.
[0042] 2) Meteorological measurement module 2
[0043] An ultrasonic integrated weather station was used, mounted horizontally in the rear corner of the chassis. As shown in Table 1, the measured meteorological parameters included temperature, humidity, wind speed, wind direction, atmospheric pressure, light intensity, PM2.5, and PM10, providing key inputs for PV performance prediction. Communication was via RS485, using the ModBus-RTU protocol.
[0044] Table 1 Meteorological measurement parameters
[0045]
[0046]
[0047] 3) Spectral measurement module 3
[0048] An intelligent spectrometer was used to measure the spectral distribution and solar irradiance at visible wavelengths, as shown in Table 2. The spectrometer was mounted on an adjustable bracket, with its sensor parallel to the photovoltaic panel to ensure consistent measurement angles. Data was transferred to an external computing device via a USB port. By measuring the solar spectral power distribution and irradiance, the inclusion of spectral data improved prediction accuracy.
[0049] Table 2 Spectral measurement parameters
[0050] parameter unit scope Wavelength range nm 380-780 Relative spectral power distribution <![CDATA[W / m 2 / nm]]> Spectrum relative to energy per nm Solar irradiance <![CDATA[W / m 2 ]]> 0.1-20000
[0051] 4) Photovoltaic measurement module 4
[0052] Using an electronic load, the PV panel's current and voltage were tested 150 times each time. (Edge computing host 1 was subsequently used to plot the IV characteristic curve and calculate the maximum power point (P_max) to evaluate the PV panel's performance, as shown in Table 3. The PV panel was fixed at a 25° south tilt angle, and the communication method was RS485.
[0053] Table 3 Photovoltaic measurement parameters
[0054] Supply voltage DC12V±0.5V Load Mode Constant Current Test voltage 0-60V Constant current step value 0.01A,0.1A,1A Discharge current 0.01-10A Voltage test accuracy ±(0.1%+0.025%FS) Maximum power 150W(25℃) Current readback accuracy ±(0.25%+0.1%FS) AH upper limit 1000AH WH upper limit 10000WH
[0055] 5) Outdoor power module 5
[0056] An outdoor power supply with a capacity of 1050Wh and a rated power of 1200W is used to power the edge computing host 1, the weather station, the electronic load, and the spectrometer. This ensures that the system can collect data outdoors for extended periods of time and guarantees system stability.
[0057] The above modules will be integrated into a protective chassis to ensure long-term outdoor operation. The specific structural layout is as follows:
[0058] Internal layout: The edge computing host 1, electronic load, and power supply are placed in the lower layer of the chassis, and the display device is placed in the upper layer to ensure heat dissipation and convenient operation.
[0059] External deployment: The photovoltaic panel is fixed on the top of the chassis, tilted 25° towards the south; the spectrometer is placed on an adjustable bracket, parallel to the photovoltaic panel; the weather station is installed in the rear corner of the chassis to avoid obstruction.
[0060] The system is equipped with a protective chassis to protect sensitive electronic components and has waterproof and dustproof functions. The entire machine is installed on a wheeled cart for easy deployment and use in outdoor environments.
[0061] Based on the above prediction system, the prediction method performs the following steps based on the edge computing architecture:
[0062] Edge computing host 1 runs a Python script to collect multi-source data from meteorological measurement module 2, spectral measurement module 3, and photovoltaic measurement module 4.
[0063] Pre-training the selected machine learning model based on the multi-source data collected simultaneously to obtain a pre-trained machine learning model;
[0064] Re-collect real-time meteorological and spectral measurement parameters, input them into the pre-trained machine learning model, and output the performance prediction results of distributed photovoltaics in real time.
[0065] Table 4 Comparison of photovoltaic power prediction performance
[0066]
[0067] Table 4 shows the performance comparison of different input data and models in photovoltaic power prediction (error sorted in ascending order). The models include extreme gradient boosting (XGBoost), kernel ridge regression (KRR), multi-layer perceptron (MLP), and support vector machine (SVM). The results show that the prediction accuracy of the model is significantly improved after adding spectral data. For example, when using meteorological and spectral data input, the extreme gradient boosting (XGBoost) model has an MSE of 8.0576 and an R 2 is 0.93, which is better than using only meteorological data (MSE is 11.7734, R 2 0.8977) or spectral data (MSE was 12.1082, R 2 The performance of the spectral data is 0.8948), which shows the importance of spectral data in improving the prediction ability.
[0068] Figure 2 The system's performance in spectral prediction tasks is demonstrated by comparing the measured spectral power (solid line) with the predicted spectral power (dashed line) at different time points (06:36:00, 10:36:00, 13:36:00, and 16:36:00 on November 13, 2024). The results show that the predicted spectral power closely matches the measured value, with the predicted curve closely following the measured curve. This demonstrates the system's excellent performance in meteorological spectral prediction tasks and provides reliable spectral data support for photovoltaic power prediction.
[0069] Figure 3This is a comparison chart of photovoltaic maximum power (Pm) prediction based on long short-term memory neural network (LSTM), showing the prediction effect of LSTM model under different input dimensions (spectral data, meteorological data, spectrum + meteorological data) under three typical meteorological conditions: sunny, cloudy and overcast.
[0070] The solid line in the figure represents the measured maximum power (Pm), while the dashed lines represent the prediction results for three models: inputting only spectral data, inputting only meteorological data, and inputting both spectral and meteorological data. The comparison results show that the LSTM model is generally able to capture the changing trends of Pm well. The model combining spectral and meteorological data achieves the highest prediction accuracy across various weather scenarios, with the fit being closest to the measured values. It is particularly sensitive to power fluctuations under complex weather conditions, such as cloudy and overcast skies, demonstrating the effectiveness of this system in integrating multi-source information for prediction.
[0071] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0072] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.
Claims
1. A distributed photovoltaic performance evaluation and prediction method based on edge intelligence, characterized in that: The following steps are involved: Real-time and synchronous collection of multi-source data of distributed photovoltaics, including meteorological measurement parameters, spectral measurement parameters and photovoltaic measurement parameters; The meteorological measurement parameters and spectral measurement parameters are input into the pre-trained machine learning model to predict photovoltaic performance, and the distributed photovoltaic performance prediction results are output. The multi-source data is used to pre-train the machine learning model, and the prediction method is processed based on the edge architecture.
2. A distributed photovoltaic performance evaluation and prediction method based on edge intelligence according to claim 1, characterized in that: The meteorological measurement parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure, light intensity, PM2.5 and PM10.
3. The distributed photovoltaic performance evaluation and prediction method based on edge intelligence according to claim 1 is characterized in that: The spectrum measurement parameters include spectral power distribution and solar irradiance at visible light wavelengths.
4. The distributed photovoltaic performance evaluation and prediction method based on edge intelligence according to claim 1 is characterized in that: The photovoltaic measurement parameters include a photovoltaic IV characteristic curve and a maximum power point, wherein the maximum power point is calculated based on the IV characteristic curve.
5. The distributed photovoltaic performance evaluation and prediction method based on edge intelligence according to claim 1 is characterized in that: The pre-trained machine learning model is one of XGBoost, KRR, MLP, KRR-KRR, and SVM.
6. A device for the distributed photovoltaic performance evaluation and prediction method based on edge intelligence according to any one of claims 1 to 5, characterized in that: It includes an edge computing host (1) and a meteorological measurement module (2), a spectrum measurement module (3) and a photovoltaic measurement module (4) connected thereto. Meteorological measurement module (2): used for collecting meteorological measurement parameters in real time; Spectral measurement module (3): used for real-time acquisition of photovoltaic measurement parameters of distributed photovoltaics; Photovoltaic measurement module (4): used for real-time acquisition of spectral measurement parameters; An edge computing host (1) is used to predict photovoltaic performance based on the meteorological measurement parameters, spectral measurement parameters and photovoltaic measurement parameters using a pre-trained machine learning model, and output a distributed photovoltaic performance prediction result.
7. The device according to claim 6, characterized in that The spectrum measurement module (3) comprises a spectrometer and an external computer, and is used to measure spectral power distribution and solar irradiance.
8. The device according to claim 6, characterized in that The photovoltaic measurement module (4) comprises a photovoltaic panel and an electronic load. The photovoltaic panel measures an IV characteristic curve through the electronic load to calculate a maximum power point.
9. The device according to claim 6, characterized in that The edge computing host (1) is integrated with a time synchronization control module, and the time synchronization control module is used to control the meteorological measurement module (2), the spectrum measurement module (3) and the photovoltaic measurement module (4) to perform simultaneous acquisition to ensure time stamp alignment.
10. The device according to claim 6, characterized in that: It also includes an outdoor power supply module (5), which is respectively connected to the edge computing host (1), the meteorological measurement module (2), the spectrum measurement module (3) and the photovoltaic measurement module (4).