基于能源大数据的电力保供综合分析数据

基于本项目提出的水光储柔性资源多目标联合优化调控策略，建立基于强化学习的调度智能体，通过智能体对预测结果与实时运行数据进行持续评估训练，引入不确定性优化方法来回溯和评价实际运行场景下调度决策行为的合理性和最优性，实现调节方案的多次评估；基于持续性评估结果分析最优资源调配方案，以15分钟/次的频率，输出精细化调度策略，在负荷波动较大时可以快给出精准调控方案，实现二次调节；完成对系统安全稳定的精准刻画和预警，快速给出安全运行、经济运行、削峰填谷等精准辅助控制策略。1、数据清洗：对数据集进行清洗，包括去除异常数据、处理缺失值。
2、数据转换: 同一区域内所有场站需要合并计算，数据统计周期至少包含24个月。
3、数据加工: 根据衢州市的用电负荷、发电量、气象数据、可调资源、水雨情、水库等数据，并推算顶峰保供能力、调节准确率、水电预测精度、光伏预测精度、全社会负荷预测精度、网供负荷预测精度、用电指标。
算法处理公式如下：
顶峰保供能力：可调资源总装机容量 - 可调资源实时出力
调节偏差率：（计划调节量 - 实际调节量） / 计划调节量
水电预测精度：水力发电预测准确率
光伏预测精度：光伏发电预测准确率
全社会负荷预测精度：全社会负荷预测准确率
网供负荷预测精度：网供负荷预测准确率
用电指标：（全社会负荷预测值-发电设备总实时出力）* 随机因子

Based on the multi-objective joint optimal regulation strategy for water-solar-storage flexible resources proposed in this project, a reinforcement learning-based scheduling agent is established. The agent continuously conducts evaluation and training on prediction results and real-time operation data, and introduces uncertainty optimization methods to retrospectively evaluate the rationality and optimality of scheduling decision behaviors in actual operation scenarios, realizing multiple evaluations of regulation schemes.
Based on the sustained evaluation results, the optimal resource allocation scheme is analyzed, and refined scheduling strategies are output at a frequency of once every 15 minutes. When the load fluctuates severely, precise regulation schemes can be quickly provided to realize secondary regulation. Accurate characterization and early warning of system security and stability are completed, and precise auxiliary control strategies such as safe operation, economic operation, peak shaving and valley filling are quickly proposed.

1. Data cleaning: Clean the dataset, including removing abnormal data and handling missing values.
2. Data transformation: All stations in the same area need to be combined for calculation, and the data statistics period shall be at least 24 months.
3. Data processing: Calculate and derive indicators including peak load guarantee capacity, regulation accuracy, hydropower prediction accuracy, photovoltaic (PV) prediction accuracy, social-wide load prediction accuracy, grid-supplied load prediction accuracy and power consumption indicators based on the power load, power generation, meteorological data, adjustable resources, water regime and rainfall data, reservoir data and other relevant data of Quzhou City.

The formulas for algorithm processing are as follows:
Peak load guarantee capacity: Total installed capacity of adjustable resources - Real-time output of adjustable resources
Regulation deviation rate: (Planned regulation volume - Actual regulation volume) / Planned regulation volume
Hydropower prediction accuracy: Prediction accuracy of hydropower generation
Photovoltaic (PV) prediction accuracy: Prediction accuracy of photovoltaic power generation
Social-wide load prediction accuracy: Prediction accuracy of social-wide load
Grid-supplied load prediction accuracy: Prediction accuracy of grid-supplied load
Power consumption indicators: (Social-wide load prediction value - Total real-time output of power generation equipment) * Random factor