基于IHO-Mamba-MHSA的红瓜子斑鱼养殖水温多步预测模型

Multi-step Water Temperature Prediction Model for Cephalopholis sonnerati Based on IHO-Mamba-MHSA

摘要:

为了提高工厂化红瓜子斑鱼养殖水温预测精度，提出了一种基于改进河马优化算法（Improved hippopotamus optimization algorithm, IHO）、Mamba模型和多头自注意力机制（Multi-head self-attention，MHSA）相结合的工厂化红瓜子斑鱼养殖水温多步预测模型（IHO-Mamba-MHSA）。为降低异常值和噪声干扰，分别采用四分位距（Interquartile range，IQR）法识别异常值和线性插值法填补缺失值，通过极端梯度提升（Extreme gradient boosting，XGBoost）进行关键因子特征筛选；为提高河马算法全局和局部搜索性能，提高其收敛速度，提出了差分变异、Levy飞行和柯西变异融合改进IHO优化多目标算法；为增强预测模型捕捉水温非线性关系、处理多步依赖性和全局信息的能力，提出Mamba模型与MHSA结合的预测模型；通过IHO优化并获得Mamba-MHSA模型组合参数，构建了IHO-Mamba-MHSA的工厂化红瓜子斑鱼养殖水温多步预测模型。将该模型对山东省莱州市某工厂化红瓜子斑鱼养殖水温进行验证，本文提出的IHO算法与遗传算法（Genetic algorithm，GA）、粒子群优化算法（Particle swarm optimization algorithm，PSO）和标准河马优化算法（Hippopotamus optimization algorithm，HO）相比，本文算法的MAE、MSE和MAPE分别最高降低33.33%、21.74%和18.37%，R2最高提升4.42%，说明IHO具有较好的多参数优化性能；与LSTM、GRU、BPNN及TCN模型对比，本模型在各预测步长下均表现最佳，当步长为24时R2仍高达0.888，充分表现其在单步与多步预测中的卓越性。各项实验结果表明本模型能够满足实际工厂化红瓜子斑鱼养殖水温精准预测与精细化管理的需求，为工厂化水产养殖水质调控提供参考。

Abstract:

Aiming to improve the accuracy of water temperature prediction in industrial Cephalopholis sonnerati aquaculture, a multi-step prediction model, IHO-Mamba-MHSA, which integrated the improved hippopotamus optimization algorithm (IHO), the Mamba model, and the multi-head self-attention (MHSA) mechanism was proposed. The interquartile range (IQR) method identified outliers, and linear interpolation imputed missing values to reduce noise impact. Key feature selection was performed by using extreme gradient boosting (XGBoost). To enhance the global and local search capabilities of the hippopotamus optimization algorithm (HO) and improve its convergence speed, an IHO was proposed by incorporating differential mutation, Levy flight, and cauchy mutation to optimize a multi-objective algorithm. To further strengthen the model’s ability to capture nonlinear relationships in water temperature, handle multi-step dependencies, and extract global information, a predictive framework combining the Mamba model and MHSA was introduced. The IHO was used to optimize the hyperparameters of the Mamba-MHSA model, forming the IHO-Mamba-MHSA multi-step prediction model for industrial Cephalopholis sonnerati aquaculture water temperature. The proposed model was validated by using water temperature data from an industrial aquaculture facility in Laizhou, Shandong. Compared with genetic algorithm (GA), particle swarm optimization (PSO), and the standard HO, the IHO achieved the highest reductions in MAE, MSE, and MAPE by 33.33%, 21.74%, and 18.37%, respectively, while increasing R2 by up to 4.42%, demonstrating its superior multi-parameter optimization performance. Furthermore, compared with long short-term memory (LSTM), gated recurrent unit (GRU), backpropagation neural network (BPNN), and temporal convolutional network (TCN), the proposed model consistently outperformed across different forecasting horizons, maintaining an R2 as high as 0.888 even at a 24-step horizon. The experimental results indicated that the proposed model met the requirements for precise water temperature prediction and refined management in industrial aquaculture, providing valuable insights for water quality regulation in intensive fish farming systems.