To effectively enhance the accuracy of short-term power load forecasting, a model based on Variational Mode Decomposition (VMD) and a multi-strategy improved Pelican Optimization Algorithm (SPOA) optimized Bidirectional Long Short-Term Memory (BiLSTM) neural network for short-term power load forecasting is proposed. Firstly, the original power load data is decomposed using VMD to reduce the complexity of the data. Secondly, the POA is improved by introducing Logistic chaos mapping, nonlinear weight factors, a Cauchy mutation strategy, and a sparrow vigilance mechanism. Then, the SPOA is used to optimize the hyperparameters of the BiLSTM, constructing the SPOA-BiLSTM forecasting model. The decomposed modal components are input into the forecasting model, and the final forecasting results are obtained by superposing the corresponding model forecasts. Experimental results show that the proposed model better fits short-term power load data than other models, demonstrating higher forecasting accuracy.