Addressing the issues of localized node and link service overload and unbalanced risk distribution in power communication network, this paper proposes a PDQN-based joint load and risk balancing routing planning algorithm. First, establish a power communication network model comprising both topology and service models. Then, redefine load and risk balancing indices based on bandwidth utilization and service importance, respectively, and formulate the routing problem as a Multi-Objective Markov Decision Process. Furthermore, we propose a Pareto Deep Q-Network (PDQN) algorithm with independent output layers and an \(\varepsilon \) -greedy strategy to simultaneously optimize load and risk balancing objectives. Finally, the proposed approach determines optimal routing strategies for power communication network by efficiently identifying balance points between competing objectives. Experimental results on power communication backbone networks in Jilin and Guangdong provinces demonstrate that PDQN significantly outperforms existing methods. In the Jilin network under high service loads, PDQN achieved a load balancing degree of 0.19 compared to 0.31 for KSP and 0.27 for LRJB. For risk balancing, PDQN reached 7.60, approximately 44% lower than KSP’s 13.63 and 31% lower than LRJB’s 10.99. In the Guangdong network, PDQN maintained superior performance with a load balancing degree of 0.11 (versus 0.21 for competitors) and risk balancing degree of 3.63 (versus 6.95 for KSP and 6.22 for LRJB), while achieving 76% resource utilization with zero blocking rate. The experiments on both networks show that our method works well in different network setups.