Seasonal energy storage effectively addresses seasonal electricity supply–demand imbalances. This study proposes a cross-seasonal compressed air energy storage system using a depleted gas reservoir as the storage reservoir. A six-stage compression–expansion system with intercooling and reheating is designed. Also, the heat of compression is recovered for district heating, and the expansion of cold energy is for cooling supply, thereby avoiding cross-seasonal heat storage costs. For the YD1 depleted gas reservoir case, the system achieves a round-trip efficiency of 57.85% and an exergy efficiency of 70.9%. The total energy utilization ratio, which represents the combined utilization of the electricity, heating, and cooling outputs relative to the corresponding energy input, reaches 166.6%. Incorporating revenue from the heating and cooling sales, the dynamic payback period is 2.38 years, and the investment recovery ratio reaches 4.40. Parametric analysis indicates that thermodynamic performance improves with the increase in discharge pressure and decrease in discharge power. Economic performance improves with longer daily operating hours and plant lifetime. This study demonstrates that depleted gas reservoirs combined with well-designed surface combined cooling, heating, and power systems offer a new research and development direction for large-scale, long-duration seasonal storage, facilitating renewable energy integration and grid stability.

Unconventional gas-well production is jointly controlled by fracture conductivity, stimulated-region supply, matrix replenishment, boundary propagation, and low-pressure fluid-property changes. In practice, RTA diagnostic curves are often affected by variable operating schedules, pressure-measurement errors, and production disturbances, making flow-stage boundaries difficult to define consistently. To reduce the subjectivity of manual interpretation and to capture stage evolution rather than whole-well classes, an ordered flow-state identification method based on a five-region analytical model and RTA sliding-window features is developed. A fully random, large-sample production-response library is generated with the five-region model. Each well production curve is divided into local time windows, from which dynamic features, including RNP, material-balance time, local slopes, pseudopressure derivatives, and normalized cumulative gas production are extracted. K-means clustering is then used to identify local states, which are reordered by material-balance time to form an ordered S1–S5 sequence. Results from 10,000 synthetic wells yielded 689,394 RTA windows, an inter-cluster separation of 1.8924, a stage-regression rate of 0.0238, and an average of 4.24 states per well. S1–S5 represent early fracture–stimulated-region response, stimulated-region supply development, matrix composite supply transition, enhanced boundary/control-volume effects, and late low-pressure property response, respectively. Application to Well M1 shows that S4 contributes the most gas (37.83%), followed by S5 (23.47%), indicating dominant mid-to-late effective supply and low-pressure long-tail production. The method converts empirical flow-regime division into reproducible and comparable window-state identification results, supporting stage diagnosis and production-strategy adjustment for unconventional gas wells.