Deep Boltzmann Wavelet Neural Network to detect changing in dynamic system

Dynamic systems, such as energy demand, exhibit complex time-varying behaviors that are challenging to model accurately due to non-stationarity, noise, and intricate temporal dependencies. Traditional Deep Boltzmann Machines (DBMs) struggle to capture these multiscale patterns effectively, limiting their predictive performance. To address this, we propose the Deep Boltzmann Wavelet Neural Network (DBWNN), This proposed model is a combination of two models: Deep Boltzmann machine and wavelet transformation. Where Deep Boltzmann machine has a superior result at classification, feature learning and prediction and wavelet transformation used for data preprocessing, noise reduction and data decomposition. Our experimental outcomes affirm the superior efficacy of DBWNN compared to other detection and prediction models. When applied to an hourly energy demand dataset for Brazil, DBWNN achieved a test loss of 0.004308. These outcomes illustrate the potential of DBWNN as a robust tool for analyzing and forecasting dynamic systems (Energy demand), particularly in scenarios involving complex temporal dependencies.