Optimization strategy retrieval method for energy supply and consumption system in agricultural parks based on knowledge graph

Multi-type energy consumption and multi-source heterogeneous data have shown the explosive growth in agricultural parks, due to the digitalization and intelligence level of the energy supply and consumption. The high dimensions of the complex data have also increased significantly to seriously restricting the reasoning, judgment, and decision-making during conventional modeling and optimization. The short strategy generation time is often required to rapidly respond to the dispatch needs under dynamic operation scenarios. In this study, an optimal strategy retrieval was proposed for energy supply and consumption in agricultural parks was proposed using a knowledge graph. Knowledge engineering and deep learning were also integrated to process the multi-source data for the highly efficient strategy matching and intelligent regulation. Firstly, the system of feature extraction was constructed for structured data, such as photovoltaic and load prediction. There were the descriptive features (such as meteorology and seasons, which accurately characterized the differences in the energy consumption scenarios under different environments) and digital features (such as temporal change rate, mean value, peak-valley difference, and daily load rate, which captured the dynamic data variation during operations). The interval processing was carried out to convert the digital features into the standard label library of the data feature. As such, the redundant information was effectively reduced in the knowledge graph. Secondly, the knowledge extraction was carried out using deep learning. The Bi-LSTM-CRF model was used to accurately identify 21 types of named entities, including energy consumption scenarios, energy equipment, energy networks, and dispatch strategies. The distinct boundaries and accurate classification were also achieved in domain entity recognition. The Bi-GRU-Attention model was introduced into the terminology library and relationship priority coefficients for the agricultural energy supply and consumption. The higher weights were assigned to the correlation relationships for high accuracy. Furthermore, the high-quality support was also provided to construct the knowledge graph. A knowledge graph of the energy supply and consumption was constructed in agricultural parks using the Neo4j graph database, including a schema and a data layer. Among them, the schema layer was used to clarify 6 types of core information, such as energy consumption scenarios, energy equipment, operating data, and dispatch strategies, as well as corresponding relationships. A knowledge structure was formed after construction. The data layer was effectively integrated to convert the structured data, semi-structured text, and unstructured images/videos after operations, such as structured processing, feature extraction, and labeling. There were structured storage, visual display, and interactive query of the historical strategy information and multi-dimensional data features. Furthermore, a hierarchical retrieval strategy was also proposed after three processes: information parsing, strategy feature matching, and dispatch strategy retrieval. Feature extraction and label matching were first performed on the prediction data of the scenario. Then, a knowledge search engine was used to match the knowledge paths in the graph. Finally, the optimal historical strategy was located after optimization. Once no suitable strategy was matched, the optimization was activated to solve the new strategy, which was then updated to the graph and strategy library in a timely manner, in order to realize the dynamic enrichment of the knowledge graph. An example experiment was selected as the winter day data from a northern agricultural park. Simulation verification shows that the strategy generation time was shortened from 22.500 to 2.162s, with an efficiency improvement of about 90%, compared with the conventional modeling and optimization. The primary indicators of the optimization, such as the operating cost of 15 338.07 Yuan and the useful energy conversion efficiency of 54.19%, shared a deviation of less than 1 percentage point from before. At the same time, there were the multi-dimensional features of the operating scenario and the strategy matching path in the form of a subgraph, in order to enhance the readability and interpretability of the dispatch. Efficient retrieval of optimization strategies was achieved for the energy supply and consumption in the agricultural parks. The reliability of optimization significantly improved the dispatch response speed. A practical and feasible solution was also provided for the application of the knowledge graphs in the field of integrated energy. It is of great significance to promote the intelligent upgrading of the energy management in agricultural parks