In the domain of cloud manufacturing, diverse services and the escalating demand for efficient and cost-effective service composition selection in cloud manufacturing have spurred day by day. Traditional methods face several challenges such as the risk of convergence issues and suboptimal solutions, high dimensionality of decision space, increased computational complexity, and scalability issues. To overcome these complexities, this paper proposed a novel method named the Deep Neural Network (DNN) based Hybrid Many Objective Leopard Seal (HMOLS) algorithm. In this study, DNN is employed to predict future Quality of Service (QoS) for each service based on historical data and integrated features. These predictions are fed back into the optimization model, influencing the dynamic weight calculation for each objective function. In this work, the Leopard Seal Optimization is implemented for hyperparameter optimization to enhance the performance of the HMOLS algorithm. Various evaluation metrics namely accuracy, throughput, reliability, resource utilization, and scalability are utilized to evaluate the performance of the HMOLS method and compare its performance with existing methods. The experimental outcomes depict the effectiveness of the HMOLS approach for the effective selection of services and fusion for cloud manufacturing. DNN-based predictions utilize deep neural networks to predict resource demand or utilization patterns and enable more accurate allocation decisions. These predictions optimize resource allocation by reallocating resources accordingly and address the challenges of resource utilization. Meanwhile, this method is required to overcome the computational overhead for applicability in cloud manufacturing.