Short term power load forecasting is an important component of power system operation and management, playing a crucial role in optimizing power dispatch and improving power system reliability. A short-term power load forecasting model was proposed to address the problem of insufficient ability of existing forecasting models to extract high randomness power load features. It includes the data decomposition part of variational mode decomposition (VMD) using improved black kite algorithm (IBKA) to optimize parameters, as well as the prediction part consisting of wavelet transform convolution (WTC) and time series lightweight adaptive network (TSLANet). Firstly, VMD was used to decompose the raw data into multiple stationary subsequences. The BKA algorithm, which utilizes Latin hypercube sampling, Gompertz model step size adjustment strategy, and northern goshawk optimization (NGO) random integer factor improvement, was introduced to optimize the decomposition layers and penalty factors, thereby improving the decomposition accuracy. Then, the decomposed components were input into the WTC-TSLANeT combination model along with temperature and humidity data for prediction. WTC performs multi-scale decomposition on the time series through wavelet transform to enhance the model’s ability to represent complex time series. TSLANet further improves the model’s ability to model time-dependent relationships through local feature extraction and frequency domain feature enhancement. Finally, the predicted values of each component were superimposed and reconstructed to obtain the final predicted value. The comparative experimental results show that the proposed model has stronger ability to extract power load characteristics and higher prediction accuracy.