Modeling and Optimizing State-Based Electricity Consumption Distributions in Dodoma Region, Tanzania Using Hidden Markov Models

The aim of this study is to model and optimize state-based Electricity Consumption Distributions in Dodoma Region, Tanzania using Hidden Markov Models. The research is specifically aimed at the identification of hidden consumption states, and the determination of the best number of hidden states to enhance the state distributions identification. A quantitative approach was utilized based on monthly TANESCO electricity demand time series data for the years 2010-2025. The estimation involved the calculation of state transition probabilities, model diagnostics testing, and evaluation of forecasting performance based on different hidden state configurations. Model selection was based on extensively documented statistical criteria, namely the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). Empirical results revealed that a three-state HMM achieved the best performance with the lowest AIC (5492.480) and BIC (5707.104) values relative to models of two to ten states. The diagnostic tests concluded that segmenting the series into latent states improved statistical attributes such as normality, homoscedasticity, and stationarity that otherwise failed in the original unsegmented data. For instance, State 1 residuals were homoscedastic (Breusch-Pagan p = 0.22), and State 3 demonstrated stationarity (ADF p ≈ 0.01), enhancing interpretability and model fit. These findings show that three states bring optimal fit for prediction, and each state fulfils the assumption of homoscedasticity, as all states follow a normal distribution. The study recommends that energy policymakers and utility providers integrate HMM-based forecasting approaches to improve decision-making in regions with dynamic and complex electricity usage patterns.