Simulation Data for "Model Predictive Control of Anaerobic Digestion Processes Using a Long Short-Term Memory Network Predictor"

This dataset contains the synthetic simulation data supporting the findings of the article "Model Predictive Control of Anaerobic Digestion Processes Using a Long Short-Term Memory Network Predictor". The data were generated to develop, validate, and test a Model Predictive Control (MPC) framework that uses a Long Short-Term Memory (LSTM) network as its internal predictive model for controlling methane flow rate in anaerobic digestion (AD) processes.The dataset is organized to <b>facilitate the replication of our methodology and the verification of our main conclusions</b>. It is structured into four main components, covering both the AM2 and ADM1 process emulators:<b>1. Training, Validation, and Test Datasets:</b> Time-series data generated from the AM2 and ADM1 emulators. These files include the manipulated variables (dilution rate for AM2, influent flow rate for ADM1) and the corresponding output (methane flow rate), with stochastic AR(1) disturbances applied to the influent to simulate realistic process variability.2. Data Quantity Analysis Results:Data used to evaluate the impact of the training set size on the LSTM predictor's accuracy (as shown in Figure 10 of the article). This allows for the analysis of the learning curve and performance plateau.<b>3. Closed-Loop Control Simulation Data:</b> The complete results from the closed-loop LSTM-MPC performance tests. This includes the reference setpoint, the measured plant output (methane flow), and the control action calculated by the MPC at each step for both the AM2 and ADM1 scenarios.4. Alternative Controller Comparison Data:Performance and computational time metrics used for the comparative analysis presented in Section 3.4.2 (Table 1). This includes the results for LSTM-MPC, GRU-MPC, ATTN-MPC, NARX-MPC, ARX-MPC, and a conventional PI controller.