Abstract: The control of electric heavy-duty truck charging stations faced challenges from the high chaos of charging load time series， which were influenced by multiple factors including the randomness of user behaviors， Markov chain characteristics of traffic flow， and harmonic interference from the power grid. Consequently， the conventional methods yielded unsatisfactory performance in controlling the peak-to-valley load difference rate. To address the issue， a collaborative optimization control technology for load-capacity-power of charging stations based on a multi‑strategy improved Northern Goshawk Optimization （NGO） algorithm was proposed. The load data exported from charging pile terminals were cleaned， and a standard matrix of charging load was constructed at 15‑minute intervals. On this basis， a multi‑strategy prediction module was designed to improve the NGO algorithm. Specifically， a dynamic time warping strategy was adopted to update the positions in the first phase of the algorithm， thereby enhancing the ability to capture temporal correlations of loads and compensating for the insufficient population diversity and the tendency to miss the optimal solution in the original algorithm. An undirected weight update mechanism was introduced to optimize the individual evolution process， alleviating the sensitivity to initial values and the problem of local optimum traps. Moreover， an attention mechanism was employed to update the final positions of individuals， increasing the focus on key data features and overcoming the slow convergence in high‑dimensional problems， thus achieving superior global search capability and convergence stability. The improved NGO algorithm was utilized to optimize the capacity configuration of charging stations and to reduce the chaos degree of the charging load time series. Based on the matching relationship between charging load and capacity， the minimum limit of the peak‑to‑valley load difference rate was adjusted， and the load balance of the power grid as well as the collaborative optimization control of charging stations were achieved. Experimental results demonstrated that after applying the proposed method， the peak‑to‑valley load difference rates for random load， price‑guided load， range‑anxiety load， and comprehensive charging load were 53%， 25%， 50%， and 60%， respectively. The control effect on the peak‑to‑valley difference rate was remarkable， which possessed the important engineering value for ensuring the safe and stable operation of the power grid.