Columbia U. Develops PSO-SR Model for Precise PV Inverter Temp. Prediction in 10 Mins.

A group of scientists from Colombia’s Pontifical Bolivarian University has developed a new method for predicting the temperature of photovoltaic inverters using symbolic regression (SR) based on particle swarm optimization (PSO). SR is a machine learning technique used to identify mathematical expressions that describe the relationship between input variables and output data, and PSO is a biomimetic optimization algorithm.

Source: Research Team Report

The research team stated, "Proper temperature control of solar inverters is crucial for maintaining the efficiency and lifespan of these systems. Incorrect temperature predictions may lead to suboptimal thermal management strategies, resulting in energy loss and reduced efficiency of solar inverters.”

Comparison of switching power loss in inverter power modules at high temperature (red) and low temperature (blue)

To train and test the requirement model, the team created a database from a photovoltaic system located on the roof of a building in Monteria, Colombia. Over the course of more than a year, the temperature, active power, and DC bus voltage of the inverter were recorded, with 70% of the data points used for training the new method and 30% for testing. The results indicate that SR PSO performs the best in both training and testing. The research team summarized, "Regarding computation time, this algorithm completed 25 iterations within 0.175 hours of each execution. Compared to techniques such as neural networks or traditional multiple regression, it is a competitive method suitable for problems with nonlinear relationships.”