The wide deployment of phasor measurement units has enabled data-driven transient stability prediction (TSP) of power systems. However, ensuring the reliability of TSP results is still a significant challenge that limits the practical application of data-driven methods. To this end, a convolutional neural network (CNN)-based deep ensemble model with optimized class-specific thresholds is proposed to achieve reliable TSP. Specifically, a CNN is utilized as the backbone predictor, where the time-series variables from multiple generators are transformed into image-like inputs, and a CNN-based deep ensemble model is developed to provide accurate confidence estimation for TSP. Subsequently, considering the asymmetric importance of different classes in TSP, a confidence-based class-specific thresholds rule is adopted, and a multi-objective optimization model for determining the class-specific thresholds is formulated. In this optimization model, the reliability requirement of TSP is imposed as a constraint, requiring that true unstable rate (TUR) equal to 100%, with the objectives of minimizing the rejection rate and maximizing the true stable rate (TSR). The Pareto front of the class-specific thresholds can be obtained by solving the optimization model. Test results on two benchmark power systems show that the proposed method achieves a TUR of 100% and a TSR of at least 99% with approximately 10% of the samples rejected, demonstrating its effectiveness and scalability.

With the increasing scale of centralized photovoltaic-plus-storage power stations, they are required not only to supply local load demands but also to provide power support to the upstream grid. The coupled interactions among renewable energy output, sharp fluctuations in local loads, and limited tie-line support capacity cause conventional dispatching methods to face issues in rare extreme scenarios, such as shortfalls in scheduled power delivery, increased imported power during peak support periods, load shedding, and voltage violations. To address these issues, this paper proposes a Wasserstein distributionally robust dispatch strategy for active distribution networks (ADNs) tailored for Upstream Power Support (UPS) tasks, incorporating tail risk assessment. An ADN operation model is established that integrates PV, energy storage, interruptible loads, and a bidirectional interface with the main grid, where the scheduled power delivery during UPS periods characterizes the support demand to the upper grid. A comprehensive risk loss function, encompassing load shedding, power shortfalls, peak-period power import, PV curtailment, and voltage violations, is constructed, and Conditional Value-at-Risk (CVaR) is employed to capture the tail risk caused by extreme scenarios. A Wasserstein ambiguity set is built around the empirical distribution of finite historical samples, and via dual reformulation, the worst-case distribution conditional risk model is transformed into a tractable mixed-integer second-order cone programming problem. Case studies on modified IEEE 33-bus and 69-bus ADN test systems demonstrate that the proposed method achieves a trade-off between routine operational costs and extreme-scenario security. By optimizing day-ahead charging/discharging schedules of energy storage, it reduces scheduled power shortfalls under low-PV and high-load conditions, as well as peak-period power import dependence, thereby enhancing the power support capability and supply resilience of the active distribution network.