Jingneng Cloud Solar-Storage IoT Data Pipeline Architecture: From Modbus Protocol Normalization to AI Scheduling

In solar-storage projects, Jingneng Cloud uses edge gateways to unify protocols such as Modbus TCP/RTU and DL/T645 into a JSON Schema reported via MQTT; in the cloud, EMQX, Kafka, and Flink clean the data before writing it to TimescaleDB, keeping the P99 latency of 30-day data queries within 200ms. Electricity price forecasting uses LightGBM, charging and discharging strategies use a rule engine plus an optimization solver, and the ONNX inference service, with a single replica of 2 cores and 4G, supports 2,000 power stations and QPS 200+. The team recommends putting the forecasting model into production only after the data pipeline has run stably for 3 months.