Abstract: For the “long corridor” and “high dynamic” scenarios in underground coal mines, where the topology of the wireless sensor network (WSN) for gas inspection frequently changes, energy consumption and latency are difficult to balance, and abnormal alarms need to be forwarded deterministically, an energy-aware multi-objective dynamic scheduling method suitable for multiple business loads has been constructed. A network model based on the energy consumption of the radio transceiver as the cost was established. The two-segment path losses in free space and multipath, the signal-to-interference-plus-noise ratio (SINR) threshold, the listening energy consumption and queuing delay introduced by carrier sensing/binary exponential backoff in the medium access control (MAC) layer, as well as three types of service constraints (periodic reporting, event-triggering, and abnormal verification) were explicitly taken into account. Based on this, a "coarse-fine" two-level collaborative optimization framework combining differential evolution (DE) and improved particle swarm optimization (IPSO) is proposed. The outer layer uses differential evolution to conduct global search in discrete dimensions (time slots, channels, inter-cluster concurrent combinations), while the inner layer employs improved particle swarm optimization with linearly decreasing inertia weight and contraction factor to achieve fine-grained convergence in continuous dimensions (transmission power, duty cycle, aggregation length). At the same time, a fuzzy weight adaptive mechanism with the average remaining energy and the length of the alarm queue as inputs was introduced, enabling the trade-off among “energy saving-reliability-delay” to switch online according to the network status. In a 100 m × 600 m mining strip area, the number of nodes is 200. The base stations are deployed at the end of the simulation environment and on the 80 m “long corridor” physical platform. The proposed DE-IPSO cooperative scheduling (differential evolution-improved particle swarm optimization cooperative dynamic scheduling, DEIPSO-DS) is compared with the quantum-behaved particle swarm optimization with fuzzy logic baseline (QPSOFL) and the wireless sensor network adaptive differential evolution baseline (WSNADE). The evaluation indicators include the first node, half node, and final node death rounds (FND, HND, LND), the average energy consumption per round of the network, the variance of residual energy, the data packet delivery ratio (PDR), and the average end-to-end delay. In a scenario with an abnormality rate of approximately 5%, the FND of DEIPSO-DS is 1 763 rounds and the LND is 2 458 rounds, which are respectively 15.82% and 49.64% higher than those of QPSOFL and WSNADE. The average energy consumption per round is 0.118 J, which is 13.84% and 27.27% lower than the two baselines respectively; the variance of residual energy is 0.041 J², and the PDR reaches 85.73%. The average latency is 168.4 ms, which is 17.15% and 30.36% shorter than the two baselines. Under approximately 10% and approximately 15% abnormal loads, the relative improvement of FND remains at 17.37% and 16.61% (compared to QPSOFL) and 41.17% and 43.15% (compared to WSNADE), energy consumption is reduced by approximately 15% and 27%, PDR is increased by approximately 12% and 30%, and latency is shortened by approximately 16% and 28%. The real measurement trend of 80 m physical platform is consistent with the simulation trend. The measured energy consumption is approximately 5% to 9% higher than the simulation, PDR is approximately 2% to 3% lower, and latency is approximately 7 to 11 ms higher. The dynamic scheduling framework based on energy consumption fine modeling and the two-level collaboration of DE-IPSO can achieve online reconfiguration of “nodes-channels-time slots-power” under the circumstances of topological migration and abnormal load fluctuations caused by tunneling advancement. It can jointly improve network lifetime, reliability and delay, and maintain stable gains under different abnormal densities. Subsequently, engineering expansion can be carried out in the directions of multi-convergence collaboration, online parameter self-adaptation and more fine-grained MAC/physical layer joint optimization.