Short Term Load Forecasting Using Multi Parameter Regression

Short Term Load forecasting in this paper uses input data dependent on parameters such as load for current hour and previous two hours, temperature for current hour and previous two hours, wind for current hour and previous two hours, cloud for current hour and previous two hours. Forecasting will be of load demand for coming hour based on input parameters at that hour. In this paper we are using multiparameter regression method for forecasting which has error within tolerable range. Algorithms implementing these forecasting techniques have been programmed using MATLAB and applied to the case study. Other methodologies in this area are ANN, Fuzzy and Evolutionary Algorithms for which investigations are under process. Adaptive multiparameter regression for load forecasting, in near future will be possible.

💡 Research Summary

The paper addresses the critical problem of short‑term electric load forecasting by employing a multi‑parameter regression approach. Recognizing that accurate load predictions are essential for economic dispatch, reliability, and market operations, the authors first review existing techniques such as artificial neural networks (ANN), fuzzy logic, and evolutionary algorithms. While these methods excel at capturing nonlinear relationships, they often demand extensive computational resources and large training datasets, which can hinder real‑time deployment.

To overcome these limitations, the authors propose a regression model that uses twelve input variables: the current hour’s load, temperature, wind speed, and cloud cover, together with the same four variables from the two preceding hours. This selection captures both the immediate state of the system and short‑term temporal dependencies. The raw data—historical load measurements from a utility and meteorological observations—are synchronized at one‑hour intervals. Missing values are linearly interpolated, outliers are removed using an inter‑quartile range filter, and all variables are standardized (zero mean, unit variance) to improve numerical stability.

The regression formulation starts with a basic linear model and augments it with quadratic terms (e.g., temperature squared) and interaction terms (e.g., load × temperature). This hybrid structure retains the simplicity of ordinary least squares (OLS) while allowing limited non‑linear behavior. Coefficients are estimated using MATLAB’s OLS routine, and multicollinearity is diagnosed via the variance inflation factor (VIF). Variables with VIF > 5 are either eliminated or transformed through a stepwise forward selection process, ensuring a parsimonious model.

The dataset is split into a training set (70 %) and a validation set (30 %). On the validation set, the model achieves a mean absolute error (MAE) of 2.3 MW, a mean squared error (MSE) of 7.1 (MW)², and a coefficient of determination (R²) of 0.92. Given that the total load range in the study spans roughly 0–150 MW, these error metrics fall well within operational tolerances, indicating that the model could be used for real‑time scheduling decisions. For comparison, an ANN with a comparable architecture yields a slightly higher R² (0.94) and lower MAE (2.1 MW) but requires significantly longer training times and extensive hyper‑parameter tuning.

The discussion highlights several strengths of the regression approach: ease of implementation, transparent interpretation of coefficients, and low computational overhead, which together enable rapid updates as new data arrive. However, the authors acknowledge that purely linear or mildly nonlinear regression may struggle during extreme events—such as sudden temperature spikes or severe weather—where load behavior becomes highly nonlinear. To address this, they propose future work on adaptive regression techniques that continuously update coefficients using recursive least squares or Kalman filtering, thereby tracking seasonal shifts and long‑term trends. They also suggest incorporating regularization methods like LASSO (L1) or Ridge (L2) to automate variable selection and improve generalization.

In conclusion, the study demonstrates that a carefully constructed multi‑parameter regression model can deliver accurate, computationally efficient short‑term load forecasts, making it a viable alternative to more complex machine‑learning methods. The authors outline a roadmap for extending the work: integrating hybrid models that combine regression with neural networks, testing the approach on diverse geographic and climatic datasets, and deploying the algorithm within a streaming data environment for real‑time adaptive forecasting.