Optimization & Data Analysis

Choosing a Texas Energy Plan with Multi-Objective Genetic Algorithms

A MATLAB optimization framework using multi-objective genetic algorithms over twenty years of ERCOT consumption data, balancing cost, renewable share and cancellation fees to identify the optimal residential electricity plan.

By Musa Ibne Mannan · Erik Jonsson School of Engineering and Computer Science, The University of Texas at Dallas

A deceptively hard consumer problem

Texas residents choose their retail electricity provider from a deregulated market with hundreds of plans. The decision looks like arithmetic and is not. Advertised rates depend on consumption tier, so the cheapest plan on paper may be expensive at your actual usage pattern. Renewable content varies independently of price. Early termination fees mean the correct choice today may be locked in when conditions change. Three objectives, in genuine conflict, over a decision space too large to enumerate by hand.

That structure is what makes it a good candidate for evolutionary optimization. There is no single optimal plan — there is a Pareto frontier of plans, each representing a different position on the trade-off between cost, sustainability and flexibility.

Understanding the load profile first

Before optimizing, the consumption data had to be characterized. Historical ERCOT residential energy usage from 2002 to 2022 was analysed three ways:

• Hourly histograms showing the probability distribution of consumption across 24-hour intervals, establishing when demand actually occurs rather than assuming a flat profile.
• A correlation heatmap across time intervals, which revealed strong positive relationships between adjacent hours — consumption is highly autocorrelated, meaning daily patterns are consistent and therefore predictable enough to optimize against.
• Boxplots of hourly consumption, exposing variability and outliers that a mean-based analysis would hide.

This stage did real work. The autocorrelation finding justified treating historical usage as a reliable predictor of future consumption, which is the assumption the whole optimization rests on. Had consumption proven erratic between adjacent intervals, the appropriate model would have been a stochastic one rather than a deterministic multi-objective search.

The optimization

A multi-objective genetic algorithm was implemented in MATLAB to search the plan space against all three objectives simultaneously. Genetic algorithms suit this problem for a specific reason: they evaluate a population of candidate solutions in parallel and evolve it, so they map the trade-off surface rather than collapsing the objectives into a single weighted score. That distinction matters — weighting the objectives in advance presumes you already know how much renewable content is worth in dollars, which is precisely the judgement the analysis should inform rather than assume.

The model converged after 101 generations, identifying Plan A as the optimal solution under the modelled objectives.

What the result demonstrates

The specific plan identified matters less than the demonstration that evolutionary algorithms handle multi-criteria decision-making in energy management robustly, on real historical data, at a scale where manual comparison is not viable. The same framework transfers directly to industrial procurement, demand response participation, or any decision where cost, sustainability and contractual flexibility have to be weighed against a measured load profile rather than a nominal one.

The broader methodological point is the sequence. The data analysis was not preliminary throat-clearing before the interesting algorithmic work — it was what established that the optimization was well-posed.