Electricity Consumption Forecasting for Seafood Processing Plants Using a Hybrid ANN–PSO Model

Abstract

The study develops a hybrid ANN–PSO model to forecast the electrical load of a seafood processing plant in Sóc Trăng, where consumption fluctuates strongly due to seasonal production and refrigeration systems. Data were collected from an MFM-384-C meter (1 hour/sample for 1 month) and processed through an HMI WEINTEK MT8071iE on an IoT platform. The ANN–PSO model uses a 1–5–1 structure, with PSO optimizing weights and biases to improve accuracy. Results show very low errors: hourly forecasts are mostly <0.5% (hour 1: actual 315 kWh, predicted 314.52 kWh – 0.15%, maximum 1.20%); weekly forecasts mostly <1% (hour 1: 2141 kWh vs. 2138.54 kWh – 0.11%, maximum 4.29%); monthly errors remain stable, with day 19 at 7524 kWh vs. 7525.41 kWh – 0.02%, and the highest error 1.45% on day 30. Compared to traditional ANN, ANN–PSO performs better, reducing RMSE from 1.45% to 0.53% (hourly) and from 1.85% to 0.78% (monthly). The model proves feasible for accurate energy consumption forecasting and energy management in processing plants.