Electricity income prediction is a critical component of fund management within the power grid system. To enhance the precision of electricity income prediction, this study proposes employing a particle swarm optimization (PSO) algorithm in conjunction with a self-organizing feature map (SOM) and bidirectional gated recurrent unit (BiGRU) to construct a SOM-PSO-BiGRU cascaded neural network deep learning model for predicting electricity revenue. The SOM model is utilized for clustering data features related to electricity revenue, while the BiGRU algorithm is employed for predicting future daily electricity revenue data. Additionally, PSO is applied to optimize connection weights and node thresholds within the cascaded neural network. Simulation experiments were conducted using MATLAB to compare the proposed model with models of the same type. The experimental results demonstrate that the proposed model has higher prediction accuracy and practical value.