研究目的
To introduce modern computing techniques as a potential decision-making tool in the field of renewable energy supply and management, specifically to predict the conversion of solar energy by a photovoltaic unit using neural networks.
研究成果
The paper addresses the challenges in forecasting photovoltaic power generation using machine learning over an IoT network. A system architecture was proposed, confirming that open loop configuration provides reliable forecasting. Improvements are listed to enhance the system for a low cost, scalable, and pervasive solution.
研究不足
The model and system could be improved by using better power measuring instruments, changing the luminosity sensor for one with reduced noise, improving the Electro Magnetic Compatibility (EMC) of the smart meter, improving the powering circuit, using a real photovoltaic system, using more data to train the algorithm, using a mixture of exogenous inputs, and building a private data logging server and interface.
1：Experimental Design and Method Selection:
The study involves building a smart meter connected to a low power photovoltaic panel to capture data, which is then sent over a LoRa IoT network to a remote server for processing. Non Linear Autoregressive Neural Networks (NARX) with Matlab and Thingspeak IoT data capture are used for predictions.
2：Sample Selection and Data Sources:
Data is captured by the smart meter from the photovoltaic panel and environmental sensors (TSL2561 for luminosity and BME680 for temperature, humidity, etc.).
3：List of Experimental Equipment and Materials:
Smart meter, photovoltaic panel (SP-P 12W), TSL2561 luminosity sensor, BME680 environmental sensor, LoRa shield, microcontroller, LoRaWAN gateway.
4：Experimental Procedures and Operational Workflow:
The smart meter measures power output and environmental parameters, sends data via LoRa network to a gateway, which then processes the data for predictions using NARX neural networks.
5：Data Analysis Methods:
The NARX training model adopted was Levenberg-Marquardt model, with data partitions for training (75%), testing (15%), and validating (15%). RMSE values were calculated to assess performance.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
