Enhancing Aquaculture Water Quality Prediction with PID-RENet

By Ashutosh RoyReviewed by Susha Cheriyedath, M.Sc.Aug 19 2023

In an article published in the journal Computers and Electronics in Agriculture, researchers introduced a novel approach to enhance water quality prediction in aquaculture. Their method combines deep neural networks (DNN) with a PID-RENet (Proportional-Integral-Derivative Residual Elimination Network) for more accurate predictions, offering potential benefits for aquaculture management and risk reduction. Using experiments with real and synthetic datasets, the study highlights PID-RENet's effectiveness in refining time-series predictions of crucial water quality parameters.

The significance of time-series prediction

Time-series prediction has garnered widespread attention across diverse industries due to its ability to forecast future values based on historical data patterns. In the context of aquaculture, accurate predictions of water quality parameters can lead to informed decision-making and risk reduction. For instance, predicting dissolved oxygen levels helps prevent oxygen deficiencies that can harm aquatic organisms, while forecasting water temperature aids in managing temperature-sensitive species.

The complexity of water quality prediction

Predicting water quality poses unique challenges. Water quality variables are often interconnected in complex ways, and their interactions can be nonlinear and intricate. Traditional linear prediction models struggle to capture these complexities, leading to inaccurate forecasts. This limitation has prompted the exploration of more advanced techniques that can capture the nuances of water quality dynamics.

Deep neural networks (DNNs) have emerged as powerful tools for predictive modeling. DNNs are designed to mimic the human brain's ability to process information and recognize patterns. They consist of interconnected layers of artificial neurons that can learn complex relationships in data. In aquaculture, DNNs have shown promise in accurately predicting water quality parameters. However, there is room for improvement, particularly in adapting to changing conditions and correcting prediction errors.

The PID-RENet approach seeks to enhance the accuracy of time-series prediction in aquaculture by integrating DNNs with a proportional-integral-derivative residual elimination network (PID-RENet). This approach marries the power of DNNs with the error correction capabilities of a PID controller—a common control mechanism used in engineering and automation.

Understanding PID control

A PID controller is a feedback control system that continuously calculates an error value as the difference between a desired setpoint and a measured process variable. PID controllers are widely used in industries to maintain desired conditions, such as temperature or pressure.

In the context of water quality prediction, the PID-RENet architecture combines a traditional PID controller with a backpropagation (BP) neural network. The PID controller calculates control signals based on historical deviations in predictions made by the DNN. The BP neural network dynamically adjusts the parameters of the PID controller, enabling it to adapt to changing conditions more effectively. This combination of DNN and PID control principles aims to improve the accuracy of predictions and reduce errors.

To validate the effectiveness of the PID-RENet approach, a series of experiments were conducted using both publicly available datasets and custom-created datasets from pond culture experiments. Two benchmark models—Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU)—were used for comparison. LSTM and GRU are advanced variations of recurrent neural networks (RNNs) that are designed to handle sequence data.

The results of the experiments consistently demonstrated that the PID-RENet-corrected models outperformed the benchmark DNN models in terms of accuracy and error reduction in time-series prediction. This outcome holds significant implications for the aquaculture industry. Accurate water quality prediction can lead to optimized breeding conditions, minimized risks of oxygen deficiencies, and effective resource allocation.

Future directions

While the PID-RENet approach has shown promise, there are avenues for further research and enhancement. Exploring the optimization of network weights and extending the application of the method to non-periodic target variables could offer even more accurate predictions. Additionally, as technology continues to evolve, the PID-RENet approach could contribute to more advanced and automated aquaculture management systems.

Conclusion

In conclusion, the PID-RENet approach represents a novel and innovative way to enhance the accuracy of time-series aquaculture predictions by combining deep neural networks' strengths and PID control principles. Water quality prediction is critical to aquaculture management, and accurate forecasts can lead to improved growth and reduced risks. The application of PID-RENet holds the potential to revolutionize aquaculture operations, contributing to informed decision-making and sustainable practices in the industry.