China’s National Energy Administration has published a shortlist of 93 projects for its “AI Plus” energy pilot programme, covering proposed applications of artificial intelligence across the energy sector. The notice says the list followed expert review and verification and includes projects such as intelligent decision support for maintenance of major substation equipment. The public notice period runs from November 5 to November 11, during which organisations and individuals may submit substantiated objections.
The shortlist is part of a broader effort to apply AI to high-value energy scenarios rather than treating it only as a general-purpose software tool. Power systems generate large volumes of operational data from substations, transmission equipment, generation facilities and control systems. AI techniques can help identify abnormal patterns, prioritise maintenance and support decisions where equipment condition changes over time. The value depends on whether the models can work with reliable sensor data and whether operators can validate their recommendations before they affect physical infrastructure.
The example cited in the notice, intelligent decision-making for maintenance of primary substation equipment, points to predictive and condition-based maintenance. Instead of relying only on fixed inspection intervals, operators can combine equipment measurements, historical faults, inspection records and environmental conditions to identify assets that may require attention. Such systems could help allocate maintenance teams and reduce unplanned outages. But a model’s recommendation must be checked against engineering procedures, safety requirements and the operating context of the equipment.
The 93 projects remain at the public-review stage
The pilot list is not the same as proof that all 93 projects are already deployed or provideing measurable results. The National Energy Administration describes them as proposed entries in a publicised shortlist after evaluation. The notice invites objections before the list is finalised. That distinction matters because pilot selection is an administrative milestone, while commercial deployment requires integration, testing, cybersecurity review, operational training and evidence that a system improves performance without creating new failure modes.
AI applications in energy can span several different layers. At the asset level, models may help detect equipment degradation or classify inspection images. At the grid level, analytics may support load forecasting, dispatch planning and fault localisation. At the customer level, software may forecast demand or help manage distributed resources. Each use case has different latency, reliability and explainability requirements. A model that is useful for planning a maintenance schedule may not be appropriate for direct control of a protection system or a safety-critical switching operation.
Cybersecurity is a central consideration because energy infrastructure is operational technology. Data sources can be incomplete, delayed or manipulated, and AI systems may introduce additional interfaces between corporate networks and industrial control environments. Project teams need clear rules for model access, software updates, data retention and human approval. They also need fallback procedures for when a model is unavailable or produces an implausible recommendation. The pilot programme’s long-term value will depend as much on governance and integration as on algorithmic accuracy.
Grid operators will need measurable results and safety controls
The shortlist also creates an opportunity to compare approaches across the energy sector. If the projects publish consistent measures, operators could assess reductions in equipment downtime, false alarms, inspection costs or response times. Without common baselines, however, results can be difficult to compare. A pilot that reports improved prediction accuracy may still have limited operational value if it does not change maintenance decisions or reduce outages. Transparent evaluation should include the cost of sensors, data preparation, computing and staff training.
The next step is the close of the objection period and subsequent confirmation of the project list. Further details about participating organisations, technical scope and implementation timelines will determine how broad the programme becomes. For energy companies, the announcement signals that AI adoption is moving into infrastructure operations, where safety and reliability matter more than novelty. The most useful outcomes will be projects that show measurable improvements while preserving human accountability and established engineering controls.