Knowledge graph-driven agricultural park energy supply and consumption system regulation and control strategy generation method

Abstract

The invention provides an agricultural park energy supply and consumption system regulation and control strategy generation method driven by a knowledge graph, and the method comprises the steps: extracting time series data feature indexes of structured data, completing the interval tagging processing, and forming a structured data feature tag library according to the complex and diversified scheduling operation demands of an agricultural park energy supply and consumption system; and combining knowledge extraction, fusion and storage of other semi-structured data and non-structured data to construct an agricultural park energy supply and consumption system knowledge graph. According to the method, the input feature information matrix is formed on the basis of the characterization processing of the to-be-optimized operation scene of the agricultural park energy supply and consumption system and in combination with the basic information, so that the corresponding historical strategy stored in the knowledge graph is retrieved layer by layer from the to-be-optimized operation scene, the system information visualization and the interactive query are realized, and meanwhile, the user experience is improved. Intelligent generation of the regulation and control strategy of the energy supply and consumption system in the agricultural park is completed, and support is provided for intelligent decision making of the energy supply and consumption system in the agricultural park.

Description

Technical Field

[0001] The present invention relates to the field of operation of integrated energy systems in agricultural parks, and specifically to the construction of a knowledge graph and a method for generating a control strategy for energy supply and utilization systems in agricultural parks driven by a knowledge graph. Background Art

[0002] With the digitalization and intelligent development of the energy supply and use system in agricultural parks, the amount of data and knowledge has shown explosive growth, making the information contained in the multi-source heterogeneous data of the energy supply and use system in agricultural parks more complex and changeable, bringing huge challenges to the reasoning and decision-making of dispatchers. In this context, knowledge graph, as a new knowledge management tool, provides an efficient data-driven solution for complex scheduling decisions. Compared with traditional database management methods, knowledge graph realizes knowledge reasoning, storage and retrieval functions through graph data organization structure, can deeply explore the implicit connection between knowledge information, handle complex and diverse association representations while ensuring the integrity of information, and is highly consistent with the current multi-information processing and scheduling decision-making needs of the energy supply and use system in agricultural parks. Therefore, the study of the generation method of control strategy for the energy supply and use system in agricultural parks driven by knowledge graph can combine a large amount of historical strategy information, and under the guidance of data-driven thinking, use knowledge graph tools to realize efficient processing and storage of information, thereby promoting the application of knowledge graph in the field of intelligent scheduling of energy in agricultural parks.

[0003] In terms of the formulation of scheduling strategies for energy supply and consumption systems in agricultural parks, existing research has widely focused on the coordinated optimization of multi-source heterogeneous energy such as electricity, cooling, gas, and heat, aiming to achieve goals such as cost minimization, environmental pollution minimization, and maximum consumption of renewable energy through complex computational models. However, with the increasing richness of the park's energy structure and the complexity of the coupling relationship between energy sources, the formulation methods of traditional scheduling strategies face challenges such as high computational complexity, strong model dependence, and difficulty in flexibly responding to changing environments. Therefore, it is particularly urgent to explore more intelligent and flexible scheduling strategy formulation methods.

[0004] With the diversified development of the power industry and the explosive growth of information, using knowledge graph technology to process the complex data information of the power grid has become the current research focus. Knowledge graphs are mainly used in the power industry in the fields of power grid fault diagnosis, equipment defect retrieval, equipment health management, etc., but have not yet been applied to the field of intelligent energy dispatching strategy formulation in agricultural parks. With its powerful semantic representation ability and structured data storage method, knowledge graphs provide new ideas and means for the formulation of dispatching strategies. Therefore, the knowledge graph, an emerging knowledge engineering technology, is introduced into the dispatching strategy formulation of the energy supply and use system of the agricultural park. The structured data of the energy supply and use system of the agricultural park is used as the characteristic indicator, and a large amount of historical optimization strategy information is used as the retrieval library. According to the data characteristics of the operation scenario to be optimized, the corresponding historical strategies stored in the knowledge graph are retrieved layer by layer, thereby completing the generation of the dispatching strategy of the energy supply and use system of the agricultural park, which has significant significance for providing a practical solution to the intelligent dispatching operation problem of the agricultural park. Summary of the invention

[0005] In view of the shortcomings in the prior art, the purpose of the present invention is to construct a knowledge graph of the energy supply and utilization system of an agricultural park, and to provide a method for generating control strategies for the energy supply and utilization system of an agricultural park driven by a knowledge graph, in response to the needs of intelligent scheduling and operation of agricultural parks with complex and diverse internal structures.

[0006] In order to solve the above technical problems, the technical solution provided by the present invention is:

[0007] The knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method includes the following steps:

[0008] S1. Offline stage: Constructing the knowledge graph of the energy supply and use system of the agricultural park, including:

[0009] S1.1 Extract the time series data feature indicators of structured data and establish a structured data feature extraction element composition system;

[0010] S1.2 completing interval labeling processing on the structured data described in S1.1 to form a structured data feature label library;

[0011] S1.3 builds a knowledge graph of the energy supply and utilization system of the agricultural park based on the structured data feature tag library formed in S1.2 and combined with the knowledge extraction, fusion and storage of other semi-structured data and unstructured data;

[0012] S2. Online operation stage: Knowledge graph-driven retrieval of agricultural park energy supply and utilization system control strategies. According to the data characteristics of the operation scenario to be optimized, the corresponding historical strategies stored in the knowledge graph obtained in S1.3 are retrieved layer by layer, including:

[0013] S2.1 Input the basic information of the operation scenario and extract the characteristic index value based on the basic information data;

[0014] S2.2 matches each feature index value obtained in S2.1 with the structured data feature label library formed in S1.2, so that each feature index corresponds to the feature interval to which it belongs, combines the matching feature interval labels with the input basic information, forms a feature information matrix that fully describes the input basic information, and inputs the knowledge graph obtained in S1.3 for retrieval;

[0015] S2.3 retrieves the associated entities layer by layer according to the characteristic attributes matched in the knowledge graph until the corresponding control strategy entity is retrieved, outputs the strategy number retrieved in the knowledge graph, and queries the corresponding control strategy in the historical strategy library;

[0016] S2.4 outputs the control strategy for the operating scenario to be optimized.

[0017] Preferably,

[0018] The structured data in S1.1 includes photovoltaic prediction data and load prediction data. The characteristics of the photovoltaic prediction data include meteorological, seasonal, and time series characteristics; the characteristics of the load prediction data include meteorological, seasonal, time series characteristics, and curve characteristics;

[0019] The characteristic index extraction method of the structured data is:

[0020] S1.1.1 The structured data feature extraction element system is set to two types of descriptive features, namely meteorological and seasonal, and the features of structured time series data are extracted in combination with time series features and curve features; the descriptive features are represented by a text label under its category, and the peak-to-valley difference, daily load rate, daily load fluctuation rate and other features in the photovoltaic prediction data time series features, load prediction data time series features and load prediction data curve features are digital features, which are jointly described and represented by all feature indicators under this category;

[0021] S1.1.2 Extraction of characteristic indicators of photovoltaic prediction data:

[0022] S1.1.2.1 Extraction of meteorological characteristics and seasonal characteristics of photovoltaic prediction data: The meteorological characteristic classification labels include cloudy, sunny, rainy, and snowy, which are determined by the meteorological forecast information provided by the research site. The meteorological characteristics can be determined by satisfying any characteristic element; the seasonal characteristic classification labels include spring, summer, autumn, and winter, which are determined by the geographical location and time of the research site. The seasonal characteristics can be determined by satisfying any characteristic element;

[0023] S1.1.2.2 Time series feature extraction of photovoltaic prediction data:

[0024] The photovoltaic data predicted a day ago are extracted from the photovoltaic prediction data time series according to the hourly time series, which contains 24 time series in total. The photovoltaic output rate, the photovoltaic output mean value and the photovoltaic output change rate in the time series corresponding to each hour jointly determine the photovoltaic prediction data characteristics of the hour. The time series characteristics of 24 hours jointly determine the photovoltaic data characteristics of the prediction day. The time series characteristic indicators of photovoltaic prediction data are as follows:

[0025] ①PV output change rate A within time series i 1,i

[0026]

[0027] Where: i represents the time sequence, i = t; P t pv , are the photovoltaic output power at time t and time t+1, kW; P pv,max is the maximum output value of the photovoltaic equipment, kW;

[0028] ② The average photovoltaic output value A in time series i 2,i

[0029]

[0030] Where: P i pv,ave is the mean value of the predicted photovoltaic output in time series i, kW;

[0031] ③ The average photovoltaic output rate A in time series i 3,i

[0032]

[0033] S1.1.3 Extraction of characteristic indicators of load forecasting data:

[0034] S1.1.3.1 Extraction of meteorological and seasonal characteristics of load forecast data: The meteorological and seasonal characteristics of load forecast data can be determined by satisfying any characteristic element. Both meteorological and seasonal characteristics are determined by external input environmental information.

[0035] S1.1.3.2 Extraction of time series features of load forecasting data:

[0036] The load data predicted a day ago is extracted from the time series features of the load forecast data according to the hourly time series, which contains 24 time series in total. The three characteristic indicators of the load rate in the time series corresponding to the segment of each hour, the time series load mean, and the time series load change rate jointly determine the load data characteristics of the hour. The time series characteristics of 24 hours jointly determine the load data characteristics of the forecast day. The time series characteristic indicators of the load forecast data are as follows:

[0037] ① Load change rate within time series i

[0038]

[0039] Where: i represents the time sequence, i = t; P t e , are the load power at time t and time t+1, kW; P e,max is the maximum value of the load, kW;

[0040] ② Average load value within time series i

[0041]

[0042] Where: P i e,ave is the average value of the load in time series i, kW;

[0043] ③ Load rate within time sequence i

[0044]

[0045] S1.1.3.3 Curve feature extraction of load forecast data:

[0046] The predicted load data are connected to form a load curve. The data characteristics of the load curve are determined by using four characteristic indicators: peak type, peak-to-valley difference, load rate, and volatility. The peak type is a descriptive feature, including three characteristic elements: single peak, double peak, and flat peak. Any type of load in any load forecast data corresponds to only one peak characteristic indicator element. The curve characteristic indicators of the load forecast data are as follows:

[0047] ① Load peak-to-valley difference

[0048] The load peak-to-valley difference is determined by the occurrence time of the peak value and valley value and the load rate at the corresponding moment. The occurrence time of the peak value and valley value is an accurate value and does not need to be processed in intervals.

[0049]

[0050] Where: are the peak and valley characteristic index values ​​of load data respectively; P e,peak , P e,valley are the peak and valley values ​​of load power, kW; P e,max is the maximum value of load power, kW; t e,peak ,t e,valley are the occurrence times of load power peak and valley respectively;

[0051] ②Daily load rate

[0052]

[0053] Where: P t e is the load power value at time t, kW;

[0054] ③Daily load fluctuation rate

[0055]

[0056] Where: is the power value of the load at time t-1, kW.

[0057] Preferably,

[0058] The S1.2 constructs a structured data feature label library, specifically, after the time series data features and curve data feature data of the structured data are extracted, the digital feature indicator values ​​are processed into intervals to form feature intervals corresponding to each digital feature indicator value, and feature value intervals are searched in the structured data feature label library. Repeated intervals are marked with the same feature interval label, and at the same time, feature attributes corresponding to the structured time series data are created in the knowledge graph database. If the generated data information label cannot match the knowledge graph data information entity constructed based on the historical database, the feature indicator value of the data is processed into intervals and added to the feature label library to update the structured data feature label library.

[0059] The construction of the structured data feature tag library in S1.2 and the interval processing formula are as follows:

[0060] ① The characteristic label of the photovoltaic output change rate in time series i [A 1,i ]

[0061]

[0062] In the formula: i represents the time sequence, i=t; are the photovoltaic output power at time t and time t+1, kW;

[0063] ② The feature label of the average photovoltaic output in time series i [A 2,i ]

[0064]

[0065] Where: P pv,max is the maximum output value of the photovoltaic equipment, kW;

[0066] ③ The characteristic label of the average photovoltaic output rate in time series i [A3,i ]

[0067]

[0068] Where: P i pv,ave is the mean value of the predicted photovoltaic output in time series i, kW;

[0069] ④Characteristic label of load change rate within time series i

[0070]

[0071] Where: i represents the time sequence, i = t; P t e , are the load power at time t and time t+1, kW; P e,max is the maximum value of the load, kW;

[0072] ⑤ Load mean feature label within time series i

[0073]

[0074] Where: P e,max is the maximum value of the load, kW;

[0075] ⑥ Load rate characteristic label within time series i

[0076]

[0077] Where: P i e,ave is the average value of the load in time series i, kW;

[0078] ⑦ Load peak-valley difference characteristic label

[0079]

[0080] Where: are the peak and valley characteristic index values ​​of load data respectively; P e,peak , P e,valley are the peak and valley values ​​of load power, kW; P e,max is the maximum value of load power, kW; t e,peak ,t e,valley are the occurrence times of load power peak and valley respectively;

[0081] ⑧Daily load rate characteristic label

[0082]

[0083] Where: P t e is the load power value at time t, kW;

[0084] ⑨Daily load fluctuation characteristic label

[0085]

[0086] Where: is the power value of the load at time t-1, kW.

[0087] Preferably,

[0088] S1.3 constructs a knowledge graph to identify the entities and relationships of the energy supply and utilization system of the agricultural park through a deep learning method, and constructs the identified system entities and relationships into a graph through the Neo4j graph database; the knowledge graph is divided into a model layer and a data layer, and the construction method is as follows:

[0089] S1.3.1 Model layer construction

[0090] The model layer defines six types of first-level entities and their relationships: energy usage scenarios, energy equipment, energy networks, time series data, scheduling strategies, and operating status;

[0091] S1.3.2 Data layer construction

[0092] The data layer constructs the entity, relationship and attribute knowledge information of the agricultural park energy supply and utilization system through data resources. Based on the historical data, text information and image information of the agricultural park energy supply and utilization system, the historical data and operation data structured data of the agricultural park energy supply and utilization system are extracted, and the architecture and basic information semi-structured and unstructured data of the agricultural park energy supply and utilization system are extracted; according to the above data information, named entity recognition is completed based on the bidirectional long short-term memory neural network combined with the conditional random field model, and relationship recognition is completed based on the bidirectional gated loop structure combined with the attention mechanism model, and the entities, attributes and corresponding relationships of the agricultural park energy supply and utilization system are obtained to complete the knowledge extraction of the knowledge graph; then the information obtained by knowledge extraction is processed by entity disambiguation and coreference resolution; finally, the Neo4j graph database is used to store the knowledge graph, and knowledge reasoning is performed to complete the update of the knowledge graph.

[0093] Preferably,

[0094] The S1.3.1 model layer defines six types of first-level entities, namely, energy usage scenarios, energy equipment, energy networks, time series data, scheduling strategies, and operating status, and their relationships are as follows:

[0095] 1) Energy usage scenario entity

[0096] Energy use scenario entities include 4 secondary entities: photovoltaic greenhouse, agricultural product processing, agricultural product storage, and life and office. The 4 secondary entities correspond to 3 types of tertiary entities: electric load, cooling load, and heat load. The heat load entity corresponding to the photovoltaic greenhouse includes 6 quaternary entities: indoor fill light, space electric field generator, physical insecticide lamp, plasma nitrogen fixation equipment, electric irrigation and drainage, and agricultural electric vehicles. Each quaternary entity includes power consumption, energy Attributes such as value and energy quality coefficient;

[0097] 2) Energy equipment entities

[0098] Energy equipment entities include six secondary entities: gas turbines, gas boilers, air source heat pumps, electric heat pumps, electric refrigerators, energy storage equipment, and photovoltaic equipment; energy storage equipment entities include three tertiary entities: batteries, heat storage tanks, and cold storage tanks; gas turbines and gas boilers entities include power generation power, heat generation power, gas consumption power, and energy Value and other attributes; air source heat pump and electric heat pump entities mainly include power consumption, heating power and energy Value attributes; the electric refrigerator entity mainly includes power consumption, cooling power and energy Value attributes; battery, heat storage tank, cold storage tank entities include charging and discharging efficiency, charging and discharging power and energy Value attributes; photovoltaic equipment entities include power generation, energy Value attributes; entities, attributes and their relationships at all levels are similar to energy usage scenarios;

[0099] 3) Energy network entities

[0100] The energy network entity includes four secondary entities: power network, heat network, cooling network and natural gas network; the power network entity includes voltage level, transmission capacity and transmission efficiency attributes; the heat network entity includes transmission temperature and heat loss rate attributes; the cooling network entity includes transmission temperature and cooling efficiency attributes; the natural gas entity includes pressure stability and gas transmission volume attributes; the relationship between entities at all levels is similar to the energy usage scenario;

[0101] 4) Time series data entity

[0102] The time series data entity includes four secondary entities: season characteristics, meteorological characteristics, curve characteristics, and time series characteristics; the season characteristics entity includes four attributes: spring, summer, autumn, and winter; the meteorological characteristics entity includes four attributes: cloudy, sunny, rainy, and snowy; the curve characteristics entity includes peak-to-valley difference, daily load rate, daily load fluctuation rate, and peak value attributes, among which the peak-to-valley difference and peak value are attribute sets, the peak-to-valley difference attribute set includes the peak power change rate within the time series, the peak occurrence time within the time series, the valley power change rate within the time series, and the valley occurrence time within the time series attributes, and the peak attribute set includes single peak, double peak, and flat peak attributes; the time series characteristics entity includes the power change rate within the time series, the power mean, the power ratio, and other attributes;

[0103] 5) Scheduling Strategy Entity

[0104] The scheduling strategy entity includes the scheduling strategy number attribute, including three secondary entities: scheduling target, output plan and control change; the scheduling target entity includes three entities: the highest crop growth comfort, the highest daily operation efficiency and the lowest operation cost; the output plan entity includes the scheduling time of the output plan on the day before, the power purchase power, the power sale power, the gas purchase power, the gas turbine output power, the gas boiler output power, the electric boiler output power, the air source heat pump output power, the absorption chiller output power, the electric chiller output power, the battery, heat storage tank and cold storage tank charging power, the battery, heat storage tank and cold storage tank discharge power, the space electric field load, the physical insect killer lamp load, etc. Ion nitrogen fixation load, electric irrigation and drainage load, agricultural electric vehicle load, and fill-in light load power attributes; the control change entity includes two third-level entities: the long-time scale control change within the day and the short-time scale control change within the day; the long-time scale control change entity within the day includes the gas purchase power change, the gas turbine output electric power change, the gas boiler output thermal power change, the electric boiler output thermal power change, the air source heat pump output thermal power change, the absorption chiller output cold power change, and the electric chiller output cold power change attributes; the short-time scale control change entity within the day includes the power purchase power change, the power sales power change, and the fill-in light load energy power change attributes;

[0105] 6) Operational Status Entity

[0106] The operation status entity includes three secondary entities: photovoltaic absorption, crop growth environment and energy utilization; the photovoltaic absorption entity includes absorption ratio and abandoned light cost attributes; the crop growth environment entity includes crop growth light comfort, crop growth thermal comfort and crop growth environment comprehensive comfort attributes; the energy utilization entity includes energy utilization efficiency, Efficiency and useful energy conversion efficiency attributes; entities, attributes and their relationships at all levels are similar to scheduling strategy entities.

[0107] Preferably,

[0108] The knowledge graph-driven agricultural park energy supply and utilization system control strategy retrieval method is:

[0109] Input the basic information of the system scenario to be optimized into the obtained knowledge graph, and retrieve the historical optimization strategies that meet the optimization requirements in the knowledge graph through information analysis, strategy feature matching, and scheduling strategy query. If the historical optimal strategy is not retrieved, it is solved through the optimization algorithm. After the strategy optimization process is completed, the system automatically extracts the structured information of the optimization strategy and imports it into the case knowledge base to update the knowledge graph, as follows:

[0110] S2.1 Input basic information of operation scenario data;

[0111] S2.2 Information analysis: Extract features and make logical judgments on the prediction data of the scenario to be optimized based on the basic information of the operation scenario data;

[0112] S2.3 Query the corresponding label of each indicator characteristic value in the characteristic label library, and match the judgment result with the entities and relationships stored in the knowledge graph of the agricultural park energy supply and utilization system, and combine the input basic information to form a characteristic information matrix that fully describes the input information;

[0113] S2.4 Strategy feature matching: Describe and confirm the optimization scenario based on the data feature attributes extracted by information analysis and the data entity matching results, use the knowledge search engine to find the matching knowledge path in the knowledge graph, input the feature information matrix of the predicted data into the optimization strategy search in the knowledge graph history database, and generate the optimization scheduling strategy number search result;

[0114] S2.5 Scheduling strategy query, query the historical strategy closest to the current operation scenario to be optimized in the knowledge graph historical optimization strategy library. If the optimal strategy cannot be found in the knowledge graph, the scenario to be optimized will be solved by the optimization algorithm and updated to the knowledge graph of the agricultural park energy supply and utilization system, and the feature label library will be updated. At the same time, the optimization strategy will be output and a new number will be added, the knowledge graph strategy library will be updated, and the optimization scheduling plan will be output; if the optimal strategy is found in the knowledge graph, the strategy number retrieved from the knowledge graph will be output, and the optimization strategy corresponding to the strategy number will be queried in the historical strategy library, and the optimization scheduling plan will be output.

[0115] A method for extracting characteristic indicators of structured data of an agricultural park energy supply and utilization system based on the above-mentioned knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method.

[0116] A method for constructing a structured data feature label library of an agricultural park energy supply and utilization system based on interval processing based on the above-mentioned knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method.

[0117] A method for constructing a knowledge graph of an agricultural park energy supply and utilization system based on the above-mentioned knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method.

[0118] A knowledge graph driven agricultural park energy supply and utilization system control strategy generation system based on the above-mentioned knowledge graph driven agricultural park energy supply and utilization system control strategy generation method.

[0119] The beneficial effects of the present invention are:

[0120] 1) The method for extracting characteristic indexes of structured data of the energy supply and utilization system of an agricultural park of the present invention can quantitatively describe the key characteristics of the daily operation scenario of the energy supply and utilization system of an agricultural park, laying a data foundation for regulatory decision-making.

[0121] 2) The structured data feature tag library construction method based on interval processing of the present invention can improve the retrieval efficiency of the operation scenarios of the agricultural park energy supply and utilization system in the knowledge graph.

[0122] 3) The method for constructing the knowledge graph of the energy supply and utilization system of an agricultural park of the present invention constructs a comprehensive knowledge system reflecting the energy supply and utilization system of an agricultural park, and promotes information integration and retrieval.

[0123] 4) The knowledge graph-based agricultural park energy supply and utilization system control strategy retrieval method of the present invention visualizes the multi-dimensional characteristics of the operation scenario and its strategy matching path, providing support for the control decision-making of the agricultural park energy supply and utilization system. BRIEF DESCRIPTION OF THE DRAWINGS

[0124] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0125] Figure 1 A framework diagram of the method for generating a control strategy for the energy supply and utilization system of an agricultural park driven by knowledge graph

[0126] Figure 2 Extraction of structural data features of energy supply and utilization system in agricultural park to form a system diagram

[0127] Figure 3 Flowchart for the formation of characteristic labels for energy supply and utilization systems in agricultural parks and the construction of a characteristic label library

[0128] Figure 4 Constructing a block diagram for the knowledge graph of energy supply and utilization system in agricultural parks

[0129] Figure 5 A diagram showing the energy scenario entities, attributes and their relationships for the energy supply and consumption system of an agricultural park

[0130] Figure 6A diagram showing the entities, attributes and relationships of the time series data of the energy supply and utilization system of the agricultural park

[0131] Figure 7 A diagram showing the entities, attributes and relationships of the scheduling strategy for the energy supply and utilization system of an agricultural park

[0132] Figure 8 Bi-LSTM-CRF named entity recognition model architecture diagram for the agricultural park energy supply system

[0133] Fig. 9 Bi-GRU-Attention relationship recognition model architecture diagram for the agricultural park energy supply and utilization system

[0134] Fig.10 Schematic diagram of the search idea for optimizing the dispatching strategy of the energy supply and utilization system in the agricultural park

[0135] Fig.11 This is a diagram showing the associated entities and corresponding relationships of strategy 1 in the knowledge graph of the energy supply and utilization system of the agricultural park

[0136] Fig.12 Display diagram of load information entities and related attributes in the knowledge graph of the energy supply and utilization system of the agricultural park

[0137] Fig.13 Schematic diagram for retrieving optimized scheduling strategies in the knowledge graph of energy supply and utilization system in agricultural parks

[0138] Fig.14 A knowledge graph-driven agricultural park energy supply and utilization system control strategy retrieval flow chart DETAILED DESCRIPTION

[0139] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the following embodiments are provided only for the purpose of illustration and are not intended to limit the scope of the present invention. Those skilled in the art may make various modifications and substitutions to the present invention without departing from the purpose and spirit of the present invention.

[0140] The purpose of the present invention is to construct a knowledge graph of the energy supply and utilization system of an agricultural park in response to the demand for intelligent scheduling and operation of agricultural parks with complex and diverse internal structures, and to provide a method for generating control strategies for the energy supply and utilization system of an agricultural park driven by a knowledge graph.

[0141] like Figure 1 The figure shows the principle diagram of the framework of the method for generating the regulation strategy of the energy supply and utilization system of agricultural parks driven by knowledge graph.

[0142] S1. Offline stage: Constructing the knowledge graph of agricultural park energy supply and utilization system.

[0143] S1.1 Establish a system for extracting elements from structured data features of the energy supply and consumption system of agricultural parks, extract elements including meteorological, seasonal, time series characteristics and curve characteristics for renewable energy and loads, and describe the operating scenarios in historical optimization strategies.

[0144] S1.2 performs interval processing on the time series data characteristics and curve data characteristics of renewable energy and load to form a system structured data feature label library.

[0145] The time series characteristics of renewable energy include output rate, output mean and output change rate. The output rate of renewable energy takes the percentage of power change at adjacent moments to the power at the current moment as the center value to determine the interval label, the output mean of renewable energy takes the percentage of power mean at adjacent moments as the center value to determine the interval label, and the output change rate of renewable energy takes the percentage of output mean to the maximum output of renewable energy as the center value to determine the interval label. The load uses load change rate and load mean to describe the time series data characteristics with load rate, and uses peak-to-valley difference, load rate and volatility to describe the curve data characteristics. The interval processing of indicators is similar to that of renewable energy.

[0146] S1.3 After completing the interval labeling processing of the structured information through the above steps, it is combined with the knowledge extraction, fusion and storage of other semi-structured and unstructured information to construct a knowledge graph of the energy supply and use system of the agricultural park.

[0147] S2. Online operation stage: Knowledge graph-driven agricultural park energy supply and utilization system control strategy retrieval, that is, according to the data characteristics of the operation scenario to be optimized, the corresponding historical strategies stored in the knowledge graph are retrieved layer by layer. The specific steps are as follows:

[0148] S2.1 Extract characteristic index values ​​based on operating scenario data.

[0149] S2.2 matches the characteristic values ​​of each characteristic indicator with the characteristic interval label library, combines the matching characteristic interval labels with the input basic information, forms a characteristic information matrix that fully describes the input information, and inputs it into the knowledge graph for retrieval.

[0150] S2.3 retrieves the associated entities layer by layer according to the characteristic attributes matched in the knowledge graph until the corresponding control strategy entity is retrieved, outputs the strategy number retrieved from the knowledge graph, and queries the corresponding control strategy in the historical strategy library.

[0151] S2.4 outputs the control strategy for the operating scenario to be optimized.

[0152] The knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method specifically includes:

[0153] 1.1 Construction of structured data feature label library of agricultural park energy supply and utilization system based on feature index extraction

[0154] The energy supply and utilization system of an agricultural park contains structured data such as equipment parameters, historical operation data, and optimized scheduling strategies, which have important reference value for the monitoring, operation optimization, and scheduling decision-making of the energy supply and utilization system of an agricultural park. Therefore, this invention focuses on the feature extraction and processing of structured data of the energy supply and utilization system of an agricultural park.

[0155] The present invention adopts a method of extracting feature indicators of structured time series data and constructing interval feature labels, processes the structured time series data information into text label information, reduces the creation of redundant information labels in the knowledge graph through the feature label library, and simultaneously creates feature attributes corresponding to the structured time series data in the knowledge graph database, laying the foundation for the visual display of the knowledge graph of the energy supply and utilization system of the agricultural park.

[0156] 1.1.1 Method for extracting characteristic indicators of structured data of energy supply and utilization system in agricultural parks

[0157] In the energy supply and consumption system of agricultural parks, information such as load forecast, photovoltaic forecast, equipment output plan, optimization scheduling strategy and actual system operation data is usually stored in the form of time series, with clear timestamps and corresponding values. The present invention takes load forecast and photovoltaic forecast as two types of structured time series data as examples, extracts data features according to time series and describes them accurately.

[0158] (1) Structured data feature extraction element composition system

[0159] In the energy supply and utilization system of agricultural parks, environmental factors such as weather and seasons will have a significant impact on photovoltaic forecast data and load forecast data. In order to accurately describe the photovoltaic forecast data and load forecast data under different seasons and different meteorological conditions, two types of descriptive features, meteorological and seasonal, are set to combine with digital features such as time series features and curve features to extract the features of structured time series data. The element system diagram of the structured time series data feature extraction of the energy supply and utilization system of agricultural parks constructed by the present invention is as follows: Figure 2 shown.

[0160] Among them, features such as weather, season and load peak type belong to descriptive features, which are only represented by a text label under their category. Features such as the peak-to-valley difference in the timing characteristics of photovoltaic prediction data, the timing characteristics of load prediction data and the load prediction data curve characteristics, daily load rate, and daily load fluctuation rate belong to digital features, and all feature indicators under this category are required to jointly describe the feature elements.

[0161] (2) Extraction elements of photovoltaic prediction data characteristic indicators

[0162] 1) Meteorological and seasonal characteristics of photovoltaic forecast data

[0163] The characteristics of photovoltaic prediction data include meteorological, seasonal, and time series characteristics. Among them, meteorological conditions will affect photovoltaic prediction data through light intensity, temperature, wind speed, and cloud cover. The meteorological characteristic classification labels include cloudy, sunny, rainy, and snowy, which are determined by the meteorological forecast information provided by the research site. The meteorological characteristics can be determined by satisfying any characteristic element; temperature and light are important factors affecting photovoltaic power generation. The temperature difference between different seasons is large, and changes in solar altitude angle, sunshine hours, and solar incidence angle will affect the characteristics of photovoltaic prediction data. The seasonal characteristic classification labels include spring, summer, autumn, and winter, which are determined by the geographical location and time of the research site. The seasonal characteristics can be determined by satisfying any characteristic element.

[0164] 2) Time series characteristics of photovoltaic prediction data

[0165] The photovoltaic data predicted a day ago are extracted from the photovoltaic prediction data time series according to the hourly time series, which contains 24 time series in total. The three characteristic indicators of photovoltaic output rate, photovoltaic output mean value and photovoltaic output change rate in the time series corresponding to each hour jointly determine the photovoltaic prediction data characteristics of the hour. The time series characteristics of 24 hours jointly determine the photovoltaic data characteristics of the prediction day. The time series characteristic indicators of photovoltaic prediction data are as follows:

[0166] ①PV output change rate A within time series i 1,i

[0167]

[0168] Where: i represents the time sequence, i = t; P t pv , are the photovoltaic output power at time t and time t+1, kW; P pv,max is the maximum output value of the photovoltaic equipment, kW.

[0169] ② The average photovoltaic output value A in time series i 2,i

[0170]

[0171] Where: P i pv,ave is the mean value of the predicted photovoltaic output in time series i, kW.

[0172] ③ The average photovoltaic output rate A in time series i 3,i

[0173]

[0174] (3) Extraction elements of load forecasting data characteristic indicators

[0175] 1) Meteorological and seasonal characteristics of load forecast data

[0176] There are four types of loads in the energy supply and consumption system of the agricultural park: electricity, cooling, gas, and heat. The characteristics of the load forecast data include meteorological, seasonal, time series, and curve characteristics. Among them, meteorology and season will affect the energy consumption of the load in the park from aspects such as temperature, humidity, weather conditions, and changes in sunshine hours. The meteorological and seasonal characteristics of the load forecast data can be determined if any characteristic element is met. The meteorological and seasonal characteristics are determined by external input environmental information.

[0177] 2) Timing characteristics of load forecasting data

[0178] The time series characteristics of load forecast data are used to describe the load changes every hour. The load data predicted a day ago is extracted according to the hourly time series, which contains 24 time series. The three characteristic indicators of the load rate in the time series corresponding to the segment every hour, the time series load mean, and the time series load change rate jointly determine the load data characteristics of the hour. The time series characteristics of 24 hours jointly determine the load data characteristics of the forecast day. The time series characteristic indicators of load forecast data are as follows:

[0179] ① Load change rate within time series i

[0180]

[0181] Where: i represents the time sequence, i = t; P t e , are the load power at time t and time t+1, kW; P e,max is the maximum value of the load, kW.

[0182] ② Average load value within time series i

[0183]

[0184] Where: P i e,ave is the average value of load in time series i, kW.

[0185] ③ Load rate within time sequence i

[0186]

[0187] 3) Curve characteristics of load forecast data

[0188] The curve characteristics of load forecast data are used to describe the energy consumption law of load. The predicted load data are connected to form a load curve. Four types of characteristic indicators, such as peak type, peak-to-valley difference, load rate, and volatility, are used to jointly determine the data characteristics of the load curve. The peak type is a descriptive feature, including three characteristic elements: single peak, double peak, and flat peak. It is usually affected by the user's living habits, production mode, weather and other factors. Any type of load in any load forecast data corresponds to only one peak characteristic indicator element. The curve characteristic indicators of load forecast data are as follows:

[0189] ① Load peak-to-valley difference

[0190] The load peak-to-valley difference is determined by the occurrence time of the peak value and valley value and the load rate at the corresponding moment. Among them, the occurrence time of the peak value and valley value is an accurate value and does not need to be processed in intervals.

[0191]

[0192]

[0193] Where: are the peak and valley characteristic index values ​​of load data respectively; P e,peak , P e,valley are the peak and valley values ​​of load power, kW; P e,max is the maximum value of load power, kW; t e,peak ,t e,valley are the occurrence times of load power peak and valley respectively.

[0194] ②Daily load rate

[0195]

[0196] Where: P t e is the load power value at time t, kW.

[0197] ③Daily load fluctuation rate

[0198]

[0199] Where: is the power value of the load at time t-1, kW.

[0200] 1.1.2 Construction of structured data feature label library of agricultural park energy supply and utilization system based on interval processing

[0201] In order to alleviate the redundancy and retrieval problems caused by precise digital features in the knowledge graph of energy supply and consumption in agricultural parks, the above digital feature index values ​​are processed by interval processing to form feature interval labels that describe data features and store them in the feature label library. This can reduce the number of entities and attributes in the optimization scheduling strategy library of the energy supply and consumption system in agricultural parks to a certain extent, reduce redundant information, and make the retrieval of knowledge graphs more concise and clear. The interval processing formula is as follows:

[0202] ① The characteristic label of the photovoltaic output change rate in time series i [A 1,i ]

[0203]

[0204] Where: i represents the time sequence, i = t; P t pv , are the photovoltaic output power at time t and time t+1, kW.

[0205] ② The feature label of the average photovoltaic output in time series i [A 2,i ]

[0206]

[0207] Where: P pv,max is the maximum output value of the photovoltaic equipment, kW

[0208] ③ The characteristic label of the average photovoltaic output rate in time series i [A 3,i ]

[0209]

[0210] Where: P i pv,ave is the mean value of the predicted photovoltaic output in time series i, kW.

[0211] ④Characteristic label of load change rate within time series i

[0212]

[0213] Where: i represents the time sequence, i = t; P t e , are the load power at time t and time t+1, kW; P e,max is the maximum value of the load, kW.

[0214] ⑤ Load mean feature label within time series i

[0215]

[0216] Where: P e,max is the maximum value of the load, kW.

[0217] ⑥ Load rate characteristic label within time series i

[0218]

[0219] Where: P i e,ave is the average value of load in time series i, kW.

[0220] ⑦ Load peak-valley difference characteristic label

[0221]

[0222] Where: are the peak and valley characteristic index values ​​of load data respectively; P e,peak , P e,valley are the peak and valley values ​​of load power, kW; P e,max is the maximum value of load power, kW; t e,peak ,t e,valley are the occurrence times of load power peak and valley respectively.

[0223] ⑧ Load day load rate characteristic label

[0224]

[0225] Where: P t e is the load power value at time t, kW.

[0226] ⑨ Load daily load fluctuation characteristic label

[0227]

[0228] Where: is the power value of the load at time t-1, kW.

[0229] Taking the intervalization of the three characteristic indicators of photovoltaic output rate, output mean and output change rate in the ninth time series as an example, the interval characteristic labels are shown in the following table, where the maximum output value of the photovoltaic equipment is 500kW.

[0230] Table 1 Photovoltaic structured data interval feature label table

[0231]

[0232] In the process of constructing the knowledge graph retrieval database of the energy supply and use system of the agricultural park, after completing the feature extraction of the data, it is necessary to process the numerical feature index values ​​into intervals and input them into the feature label library for annotation. At the same time, the corresponding data feature attributes are created in the knowledge graph database to lay the foundation for the strategic retrieval of the knowledge graph. The formation of the feature labels of the energy supply and use system of the agricultural park and the construction process of the feature label library are as follows: Figure 3 shown.

[0233] For the newly generated structured time series data of the agricultural park energy supply and use system, its characteristic values ​​need to be extracted according to the data characteristic indicators, and the characteristic value interval matching is performed in the characteristic interval label library. The characteristic value of a certain indicator may match multiple labels of the characteristic interval library of the indicator at the same time. The matching multiple labels can be searched in the knowledge graph at the same time, and finally the unique strategy is retrieved corresponding to the characteristic label that meets all characteristic indicators. If the data information label generated by the system cannot match the data information entity of the knowledge graph of the agricultural park energy supply and use system built based on the historical database, the characteristic indicator value of the data is interval processed and added to the characteristic label library to realize the enrichment and update of the structured data characteristic label library of the agricultural park energy supply and use system.

[0234] 1.2 Construction of knowledge graph of energy supply and utilization system in agricultural parks

[0235] Through deep learning methods, we can accurately identify key entities in the energy supply and use system of agricultural parks, explore the associations between entities in the energy supply and use system, and lay the foundation for the construction of the knowledge graph of the energy supply and use system of agricultural parks. Through the Neo4j graph database, the above-identified entities and relationships of the energy supply and use system of agricultural parks are constructed into a graph, which can realize the visualization and interactive query of system information, provide users with intuitive and efficient system management tools, and promote the intelligent development of the energy supply and use system of agricultural parks.

[0236] The construction of the knowledge graph of the energy supply and utilization system of the agricultural park is divided into the construction of the model layer and the construction of the data layer. The construction principle diagram is as follows: Figure 4 shown.

[0237] (1) Model layer construction

[0238] The model layer is the knowledge organization structure of the knowledge graph of the energy supply and use system of the agricultural park, which describes the knowledge information such as entities, relationships between entities and entity attributes within the energy supply and use system of the agricultural park. With the help of experts in the power field, the present invention conducts a detailed analysis of the text content of the plan and extracts meaningful conceptual entities and related work attributes of the agricultural park. The model layer of the knowledge graph of the energy supply and use system of the agricultural park defines six types of first-level entities and their relationships: energy use scenarios, energy equipment, energy networks, time series data, scheduling strategies, and operating status.

[0239] 1) Energy usage scenario entity

[0240] The energy scenario entity of the energy supply and consumption system of the agricultural park describes the various types of energy demand in different areas of the park, mainly including four secondary entities: photovoltaic greenhouse, agricultural product processing, agricultural product storage, and life and office. The four secondary entities correspond to three types of third-level entities: electric load, cooling load, and thermal load. Among them, the thermal load entity corresponding to the photovoltaic greenhouse includes six fourth-level entities: indoor fill light, space electric field generator, physical insecticide lamp, plasma nitrogen fixation equipment, electric irrigation and drainage, and agricultural electric vehicles. Each fourth-level entity mainly includes power consumption, energy The energy use scenarios of the agricultural park energy supply and use system at all levels, their attributes and their relationships are as follows: Figure 5 shown.

[0241] 2) Energy equipment entities

[0242] The energy equipment entity of the agricultural park energy supply and use system covers the energy production, conversion and storage equipment of the agricultural park, including six secondary entities: gas turbines, gas boilers, air source heat pumps, electric heat pumps, electric refrigerators, energy storage equipment and photovoltaic equipment. The energy storage equipment entity includes three tertiary entities: batteries, heat storage tanks and cold storage tanks; the gas turbine and gas boiler entities mainly include power generation power, heat generation power, gas consumption power and energy Value and other attributes; air source heat pump and electric heat pump entities mainly include power consumption, heating power and energy The electric refrigerator entity mainly includes power consumption, cooling power and energy The battery, heat storage tank, and cold storage tank entities mainly include charging and discharging efficiency, charging and discharging power, and energy Attributes such as power generation, energy, etc.; photovoltaic equipment entities mainly include power generation, energy Entities, attributes and their relationships at all levels are similar to those in energy usage scenarios.

[0243] 3) Energy network entities

[0244] The energy network entity of the agricultural park energy supply and consumption system is the infrastructure for energy transmission and distribution, including four secondary entities: power network, thermal network, cooling network and natural gas network. The power network entity mainly includes attributes such as voltage level, transmission capacity and transmission efficiency; the thermal network entity mainly includes attributes such as transmission temperature and heat loss rate; the cooling network entity mainly includes attributes such as transmission temperature and cooling efficiency; the natural gas entity mainly includes attributes such as pressure stability and gas transmission volume. The relationship between entities at all levels is similar to the energy consumption scenario.

[0245] 4) Time series data entity

[0246] The time series data entity of the energy supply and consumption system of the agricultural park is an important basis for the energy management decision-making of the agricultural park. It includes four secondary entities: seasonal characteristics, meteorological characteristics, curve characteristics and time series characteristics. The seasonal characteristic entity includes four attributes: spring, summer, autumn and winter; the meteorological characteristic entity includes four attributes: cloudy, sunny, rainy and snowy; the curve characteristic entity mainly includes peak-to-valley difference, daily load rate, daily load fluctuation rate, peak value and other attributes, among which the peak-to-valley difference and peak value are attribute sets, and the peak-to-valley difference attribute set includes the peak power change rate within the time series, the peak occurrence time within the time series, the valley power change rate within the time series, the valley occurrence time within the time series and other attributes. The peak attribute set includes single peak, double peak and flat peak and other attributes; the time series characteristic entity mainly includes the power change rate within the time series, the power mean, the power ratio and other attributes. Time series data entities, attributes and their relationships at all levels, such as Figure 6 shown.

[0247] 5) Scheduling Strategy Entity

[0248] The scheduling strategy entity of the energy supply and consumption system of the agricultural park is the core means of energy optimization management. The scheduling strategy entity contains the scheduling strategy number attribute. It should be pointed out that after retrieving the scheduling strategy number in the knowledge graph layer by layer according to the data characteristics of the operation scenario to be optimized, the corresponding control strategy stored in the historical strategy library can be called out through the scheduling strategy number. The scheduling strategy entity contains three secondary entities: scheduling target, output plan, and control change. The scheduling target entity contains three-level entities such as system operation target type, such as the highest crop growth comfort, the highest daily operation efficiency, and the lowest operation cost; the output plan entity mainly includes the scheduling time of the output plan of the day before, the power purchase power, the power sale power, the gas purchase power, the gas turbine output power, the gas boiler output power, the electric boiler output power, the air source heat pump output power, the absorption chiller output power, the electric chiller output power, the battery, the heat storage tank and the cold storage tank charging power, the battery, the heat storage tank and the cold storage tank discharge power, the space electric field load, the physical insecticide lamp load, the plasma nitrogen fixation load, the electric drainage load, the agricultural electric vehicle load The control change entity includes two third-level entities: the long-time scale control change within the day and the short-time scale control change within the day; the long-time scale control change entity within the day mainly includes the gas purchase power change, the gas turbine output electric power change, the gas boiler output thermal power change, the electric boiler output thermal power change, the air source heat pump output thermal power change, the absorption chiller output cold power change, the electric chiller output cold power change, etc. The short-time scale control change entity within the day mainly includes the power purchase power change, the power sale power change, the fill light load energy power change, etc. The entities, attributes and their relationships at all levels of the scheduling strategy, such as Figure 7 shown.

[0249] 6) Operational Status Entity

[0250] The operation status entity of the energy supply and use system of the agricultural park includes three secondary entities: photovoltaic absorption, crop growth environment and energy utilization. The photovoltaic absorption entity mainly includes the absorption ratio, abandoned light cost and other attributes; the crop growth environment entity mainly includes the crop growth light comfort, crop growth heat comfort and crop growth environment comprehensive comfort and other attributes; the energy utilization entity mainly includes energy utilization efficiency, The entities, attributes and their relationships at all levels are similar to those of the scheduling strategy entity.

[0251] Data layer construction

[0252] The data layer is the data information part of the knowledge graph of the energy supply and use system of the agricultural park, and the knowledge information such as the entities, relationships and attributes of the energy supply and use system of the agricultural park is constructed through data resources. The construction of the data layer of the present invention is based on the historical data, text information and image information of the energy supply and use system of the agricultural park, and the structured data such as the historical data and operation data of the energy supply and use system of the agricultural park, and the semi-structured and unstructured data such as the architecture and basic information of the energy supply and use system of the agricultural park are extracted; secondly, according to the above data information, the named entity recognition is completed based on the bidirectional long short-term memory neural network combined with the conditional random field (Bi-LSTM-CRF) model, and the relationship recognition is completed based on the bidirectional gated loop structure combined with the attention mechanism (Bi-GRU-Attention) model, and the entities, attributes and the corresponding relationships between them of the energy supply and use system of the agricultural park are obtained to complete the knowledge extraction of the knowledge graph; then the information obtained by the knowledge extraction is processed by entity disambiguation and coreference resolution, and entities that may have multiple meanings are distinguished and nouns and pronouns with the same meaning and reference are merged; finally, the Neo4j graph database is used to store the knowledge graph, and knowledge reasoning is performed to complete the update of the knowledge graph.

[0253] The goal of named entity recognition is to identify named entities from unstructured text and classify these entities. The Bi-LSTM-CRF model is a popular method to accomplish this task. Its principle is mainly divided into three parts: embedding layer, bidirectional long short-term memory network and conditional random field.

[0254] like Figure 8As shown in the figure, this is the architecture diagram of the Bi-LSTM-CRF named entity recognition model for the energy supply and utilization system of the agricultural park. The architecture diagram of the Bi-LSTM-CRF named entity recognition model for the energy supply and utilization system of the agricultural park first converts each word in the text into a vector rich in semantic information through the embedding layer; then the bidirectional long short-term memory network (Bi-LSTM) is used to capture the forward and backward context features of these words; finally, the conditional random field (CRF) layer, based on the output of the Bi-LSTM, considers the dependencies between labels to ensure that the predicted named entity label sequence is both accurate and grammatical.

[0255] The goal of relation recognition is to identify the relationship between entities from text. The Bi-GRU-Attention model is an effective method to accomplish this task. Its principle can be divided into three parts: input layer, Bi-GRU layer and attention layer. Fig. 9 The figure shows the architecture of the Bi-GRU-Attention relationship recognition model for the energy supply and utilization system of the agricultural park. First, the text is converted into a vector form through the input layer; the bidirectional gated recurrent unit (Bi-GRU) is used to capture the context of each word in the text; on this basis, the attention mechanism is introduced to further highlight the key information related to relationship recognition in the text and ignore unimportant details. This weighted attention method enables the model to focus on the most valuable parts, thereby improving the accuracy of relationship recognition. Finally, by integrating these features, the model output can accurately determine the type of relationship between entities in the text.

[0256] To sum up, through operations such as knowledge extraction, knowledge fusion, and knowledge storage, a knowledge graph of the agricultural park energy supply and utilization system is formed, which takes into account the physical structure and operating status of the agricultural park energy supply and utilization system, and provides knowledge reference for intelligent decision-making of the agricultural park energy supply and utilization system.

[0257] 2 Knowledge graph-driven retrieval method for agricultural park energy supply and utilization system regulation strategy

[0258] The knowledge graph of the agricultural park energy supply and utilization system stores a rich historical database, which can effectively integrate and represent the system scheduling related information based on the optimized operation scenarios of the agricultural park energy supply and utilization system, and provide support for the intelligent decision-making of the agricultural park energy supply and utilization system. The knowledge graph-driven agricultural park energy supply and utilization system control strategy retrieval process, such as Fig.14 shown.

[0259] 2.1 Extract characteristic index values ​​based on operation scenario data

[0260] 2.2 Knowledge graph-driven search ideas for agricultural park energy supply and utilization system regulation strategies

[0261] The retrieval ideas of the knowledge graph optimization strategy of the energy supply and utilization system of the agricultural park, such as Fig.10 As shown in the figure, after inputting the basic information such as photovoltaic prediction data and load prediction data of the system scenario to be optimized into the knowledge graph of the energy supply and consumption system of the agricultural park, the historical optimization strategy that meets the optimization requirements in the knowledge graph can be retrieved through three steps: information analysis, strategy feature matching, and scheduling strategy retrieval. If the historical optimal strategy is not retrieved, it is solved through the optimization algorithm. After the strategy optimization process is completed, the system automatically extracts the structured information of the optimization strategy and imports it into the case knowledge base to enrich the knowledge base of the knowledge graph of the energy supply and consumption system of the agricultural park. Among them, the information analysis module will perform feature extraction and logical judgment on the predicted data of the scenario to be optimized, query the corresponding label of each indicator characteristic value in the feature label library, and match the judgment result with the entities and relationships stored in the knowledge graph of the agricultural park energy supply and use system; the strategy feature matching module describes and confirms the scenario to be optimized in the agricultural park energy supply and use system according to the data feature attributes extracted by the information analysis module and the data entity matching results, and uses the knowledge search engine to find the matching knowledge path in the knowledge graph of the agricultural park energy supply and use system, and generates the optimization scheduling strategy number retrieval result; finally, through the scheduling strategy query module, the historical strategy closest to the current operation scenario to be optimized is queried in the historical optimization strategy library of the agricultural park energy supply and use system. If the optimal strategy cannot be found in the knowledge graph, the scenario to be optimized will be solved by the optimization algorithm and updated to the knowledge graph of the agricultural park energy supply and use system, providing auxiliary decision-making reference for the optimization scheduling of the agricultural park energy supply and use system.

[0262] 2.3 Knowledge graph-driven agricultural park energy supply and utilization system control strategy retrieval process

[0263] (1) The structure of the optimization scheduling strategy of the energy supply and utilization system of the agricultural park in the knowledge graph

[0264] In order to realize the retrieval of the control strategy of the energy supply and use system of the agricultural park driven by the knowledge graph, it is necessary to build a complete knowledge graph of the optimization and scheduling strategy of the energy supply and use system of the agricultural park based on a large amount of historical operation data, historical optimization strategies and scheduling information of the system. Since the knowledge graph can only recognize and store text information, the present invention stores the historical optimization scheduling strategy in the form of strategy number tags in the knowledge graph, and the specific scheduling strategy needs to be obtained by querying the corresponding number in the strategy library.

[0265] Taking the historical optimization strategy 1 stored in the knowledge graph of the energy supply and utilization system of the agricultural park as an example, the storage structure of its related information entities and corresponding relationships in the Neo4j graph database is as follows: Fig.11As shown in the figure. It can be seen from the figure that the only scheduling strategy entity can be determined through the control target entity with the minimum operating cost, the photovoltaic feature 1 data entity, and the load information 1 data entity. Among them, the photovoltaic feature 1 data entity is determined by 26 data feature entities, including sunny weather, summer season and the corresponding photovoltaic data time series characteristics, and each photovoltaic data time series feature entity is described by 3 photovoltaic data time series feature attributes; the load information 1 data entity is determined by 10 entities, including sunny weather, summer season, load time series characteristics and load curve characteristics corresponding to electric cooling heat load, and each load time series characteristic entity is jointly determined by 24 load data time series characteristic entities, and each load data time series data characteristic entity is jointly described by 3 load time series characteristic attributes; each load curve characteristic entity is described by 4 load data time series characteristic entities and their corresponding attributes.

[0266] Taking the electric load data feature entity in strategy 1 as an example, the relationship between the data entities associated with the load information 1 entity and the corresponding attributes is as follows: Fig.12 shown.

[0267] (2) Knowledge graph-driven agricultural park energy supply and utilization system control strategy retrieval process

[0268] The Neo4j graph database can intuitively display the information entities and data attributes related to strategy 1 in the knowledge graph of the energy supply and use system of the agricultural park. Conversely, the historical strategies that are closest to the optimization requirements of the energy supply and use system of the agricultural park can be retrieved layer by layer in the knowledge graph through the predicted data feature attributes extracted from the scenario to be optimized. Taking the time series feature attributes of the electric load of strategy 1 as an example, the optimization scheduling strategy retrieval diagram in the knowledge graph of the energy supply and use system of the agricultural park is as follows: Fig.13 As shown, the specific process is as follows:

[0269] 1) Extract the characteristic index value B2 of the electric load in the 8-9 period of the day according to the optimized scenario data;

[0270] 2) Match the characteristic interval label that the characteristic index value B2 of the electric load in the 8-9 period of the day meets in the electric load time series characteristic label library, and input the label information into the knowledge graph for retrieval;

[0271] 3) Retrieve the 8-9 period electric load time series feature attributes corresponding to the 8-9 period electric load time series feature labels stored in the knowledge graph of the agricultural park energy supply and utilization system;

[0272] 4) According to the time series characteristic attributes of the electric load in the period 8-9 matched in the knowledge graph, the associated entities are retrieved layer by layer until the corresponding control strategy entity is retrieved and the strategy matching path is generated. This completes the retrieval process from the time series characteristic index value to the optimized scheduling strategy.

[0273] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

[0274] Finally, it should be noted that the above description is only a preferred example of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method, characterized in that: The following steps are involved: S1. Offline stage: Constructing the knowledge graph of the energy supply and use system of the agricultural park, including: S1.1 Extract the time series data feature indicators of structured data and establish a structured data feature extraction element composition system; S1.2 completing interval labeling processing on the structured data described in S1.1 to form a structured data feature label library; S1.3 builds a knowledge graph of the energy supply and utilization system of the agricultural park based on the structured data feature tag library formed in S1.2 and combined with the knowledge extraction, fusion and storage of other semi-structured data and unstructured data; S2. Online operation stage: Knowledge graph-driven retrieval of agricultural park energy supply and utilization system control strategies. According to the data characteristics of the operation scenario to be optimized, the corresponding historical strategies stored in the knowledge graph obtained in S1.3 are retrieved layer by layer, including: S2.1 Input the basic information of the operation scenario and extract the characteristic index value based on the basic information data; S2.2 matches each feature index value obtained in S2.1 with the structured data feature label library formed in S1.2, so that each feature index corresponds to the feature interval to which it belongs, combines the matching feature interval labels with the input basic information, forms a feature information matrix that fully describes the input basic information, and inputs the knowledge graph obtained in S1.3 for retrieval; S2.3 retrieves the associated entities layer by layer according to the characteristic attributes matched in the knowledge graph until the corresponding control strategy entity is retrieved, outputs the strategy number retrieved in the knowledge graph, and queries the corresponding control strategy in the historical strategy library; S2.4 outputs the control strategy for the operating scenario to be optimized.

2. According to the knowledge graph driven agricultural park energy supply and utilization system control strategy generation method of claim 1, it is characterized in that: The structured data in S1.1 includes photovoltaic prediction data and load prediction data. The characteristics of the photovoltaic prediction data include meteorological, seasonal, and time series characteristics; the characteristics of the load prediction data include meteorological, seasonal, time series characteristics, and curve characteristics; The characteristic index extraction method of the structured data is: S1.1.1 The structured data feature extraction element system is set to two types of descriptive features, namely meteorological and seasonal, and the features of structured time series data are extracted in combination with time series features and curve features; the descriptive features are represented by a text label under its category, and the peak-to-valley difference, daily load rate, daily load fluctuation rate and other features in the photovoltaic prediction data time series features, load prediction data time series features and load prediction data curve features are digital features, which are jointly described and represented by all feature indicators under this category; S1.1.2 Extraction of photovoltaic prediction data characteristic indicators: S1.1.2.1 Extraction of meteorological characteristics and seasonal characteristics of photovoltaic prediction data: The meteorological characteristic classification labels include cloudy, sunny, rainy, and snowy, which are determined by the meteorological forecast information provided by the research site. The meteorological characteristics can be determined if any characteristic element is met; The seasonal characteristics classification labels include spring, summer, autumn, and winter, which are determined by the geographical location and time of the research site. The seasonal characteristics can be determined if any characteristic element is met; S1.1.2.2 Time series feature extraction of photovoltaic prediction data: The photovoltaic data predicted a day ago are extracted from the photovoltaic prediction data time series according to the hourly time series, which contains 24 time series in total. The photovoltaic output rate, the photovoltaic output mean value and the photovoltaic output change rate in the time series corresponding to each hour jointly determine the photovoltaic prediction data characteristics of the hour. The time series characteristics of 24 hours jointly determine the photovoltaic data characteristics of the prediction day. The time series characteristic indicators of photovoltaic prediction data are as follows: ①PV output change rate A within time series i 1,i Where: i represents the time sequence, i = t; P t pv , are the photovoltaic output power at time t and time t+1, kW; P pv,max is the maximum output value of the photovoltaic equipment, kW; ② The average photovoltaic output value A in time series i 2,i Where: P i pv,ave is the mean value of the predicted photovoltaic output in time series i, kW; ③ The average photovoltaic output rate A in time series i 3,i S1.1.3 Extraction of characteristic indicators of load forecasting data: S1.1.3.1 Extraction of meteorological and seasonal characteristics of load forecast data: The meteorological and seasonal characteristics of load forecast data can be determined by satisfying any characteristic element. Both meteorological and seasonal characteristics are determined by external input environmental information. S1.1.3.2 Extraction of time series features of load forecasting data: The load data predicted a day ago is extracted from the time series features of the load forecast data according to the hourly time series, which contains 24 time series in total. The three characteristic indicators of the load rate in the time series corresponding to the segment of each hour, the time series load mean, and the time series load change rate jointly determine the load data characteristics of the hour. The time series characteristics of 24 hours jointly determine the load data characteristics of the forecast day. The time series characteristic indicators of the load forecast data are as follows: ① Load change rate within time series i Where: i represents the time sequence, i = t; P t e , are the load power at time t and time t+1, kW; P e,max is the maximum value of the load, kW; ② Average load value within time series i Where: P i e,ave is the average value of the load in time series i, kW; ③ Load rate within time sequence i S1.1.3.3 Curve feature extraction of load forecast data: The predicted load data are connected to form a load curve. The data characteristics of the load curve are determined by using four characteristic indicators: peak type, peak-to-valley difference, load rate, and volatility. The peak type is a descriptive feature, including three characteristic elements: single peak, double peak, and flat peak. Any type of load in any load forecast data corresponds to only one peak characteristic indicator element. The curve characteristic indicators of the load forecast data are as follows: ① Load peak-to-valley difference The load peak-to-valley difference is determined by the occurrence time of the peak value and valley value and the load rate at the corresponding moment. The occurrence time of the peak value and valley value is an accurate value and does not need to be processed in intervals. Where: are the peak and valley characteristic index values ​​of load data respectively; P e,peak , P e,valley are the peak and valley values ​​of load power, kW; P e,max is the maximum value of load power, kW; t e,peak ,t e,valley are the occurrence times of load power peak and valley respectively; ②Daily load rate Where: P t e is the load power value at time t, kW; ③Daily load fluctuation rate Where: is the power value of the load at time t-1, kW.

3. The method for generating a control strategy for an agricultural park energy supply and utilization system driven by a knowledge graph according to claim 2 is characterized in that: S1.2 constructs a structured data feature label library, specifically, after extracting the time series data features and curve data features of the structured data, the digital feature index values ​​are processed into intervals to form feature intervals corresponding to each digital feature index value, and feature value intervals are searched in the structured data feature label library. Repeated intervals are marked with the same feature interval label, and feature attributes corresponding to the structured time series data are created in the knowledge graph database. If the generated data information label cannot match the knowledge graph data information entity constructed based on the historical database, the feature index value of the data is processed into intervals and added to the feature label library, and the structured data feature label library is updated, specifically: The construction of the structured data feature tag library in S1.2 and the interval processing formula are as follows: ① The characteristic label of the photovoltaic output change rate in time series i [A 1,i ] Where: i represents the time sequence, i = t; P t pv , are the photovoltaic output power at time t and time t+1, kW; ② The feature label of the average photovoltaic output in time series i [A 2,i ] Where: P pv,max is the maximum output value of the photovoltaic equipment, kW; ③ The characteristic label of the average photovoltaic output rate in time series i [A 3,i ] Where: P i pv,ave is the mean value of the predicted photovoltaic output in time series i, kW; ④Characteristic label of load change rate within time series i Where: i represents the time sequence, i = t; P t e , are the load power at time t and time t+1, kW; P e,max is the maximum value of the load, kW; ⑤ Load mean feature label within time series i Where: P e,max is the maximum value of the load, kW; ⑥ Load rate characteristic label within time series i Where: P i e,ave is the average value of the load in time series i, kW; ⑦ Load peak-valley difference characteristic label Where: are the peak and valley characteristic index values ​​of load data respectively; P e,peak , P e,valley are the peak and valley values ​​of load power, kW; P e,max is the maximum value of load power, kW; t e,peak ,t e,valley are the occurrence times of load power peak and valley respectively; ⑧Daily load rate feature label Where: P t e is the load power value at time t, kW; ⑨Daily load fluctuation characteristic label Where: is the power value of the load at time t-1, kW.

4. The method for generating a control strategy for an agricultural park energy supply and utilization system driven by a knowledge graph according to claim 1 is characterized in that: S1.3 constructs a knowledge graph to identify the entities and relationships of the energy supply and utilization system of the agricultural park through a deep learning method, and constructs the identified system entities and relationships into a graph through a Neo4j graph database; The knowledge graph is divided into a model layer and a data layer, and the construction method is as follows: S1.3.1 Model layer construction The model layer defines six types of first-level entities and their relationships: energy usage scenarios, energy equipment, energy networks, time series data, scheduling strategies, and operating status; S1.3.2 Data layer construction The data layer constructs the entity, relationship and attribute knowledge information of the agricultural park energy supply and utilization system through data resources. Based on the historical data, text information and image information of the agricultural park energy supply and utilization system, the historical data and operation data structured data of the agricultural park energy supply and utilization system are extracted, and the architecture and basic information semi-structured and unstructured data of the agricultural park energy supply and utilization system are extracted; according to the above data information, named entity recognition is completed based on the bidirectional long short-term memory neural network combined with the conditional random field model, and relationship recognition is completed based on the bidirectional gated loop structure combined with the attention mechanism model, and the entities, attributes and corresponding relationships of the agricultural park energy supply and utilization system are obtained to complete the knowledge extraction of the knowledge graph; then the information obtained by knowledge extraction is processed by entity disambiguation and coreference resolution; finally, the Neo4j graph database is used to store the knowledge graph, and knowledge reasoning is performed to complete the update of the knowledge graph.

5. According to claim 4, the method for generating a control strategy for an agricultural park energy supply and utilization system driven by a knowledge graph is characterized in that: The S1.3.1 model layer defines six types of first-level entities, namely, energy usage scenarios, energy equipment, energy networks, time series data, scheduling strategies, and operating status, and their relationships are as follows: 1) Energy usage scenario entity Energy use scenario entities include 4 secondary entities: photovoltaic greenhouse, agricultural product processing, agricultural product storage, and life and office. The 4 secondary entities correspond to 3 types of tertiary entities: electric load, cooling load, and heat load. The heat load entity corresponding to the photovoltaic greenhouse includes 6 quaternary entities: indoor fill light, space electric field generator, physical insecticide lamp, plasma nitrogen fixation equipment, electric irrigation and drainage, and agricultural electric vehicles. Each quaternary entity includes power consumption, energy Attributes such as value and energy quality coefficient; 2) Energy equipment entities Energy equipment entities include six secondary entities: gas turbines, gas boilers, air source heat pumps, electric heat pumps, electric refrigerators, energy storage equipment, and photovoltaic equipment; energy storage equipment entities include three tertiary entities: batteries, heat storage tanks, and cold storage tanks; gas turbines and gas boilers entities include power generation power, heat generation power, gas consumption power, and energy Value and other attributes; air source heat pump and electric heat pump entities mainly include power consumption, heating power and energy Value attributes; the electric refrigerator entity mainly includes power consumption, cooling power and energy Value attributes; battery, heat storage tank, cold storage tank entities include charging and discharging efficiency, charging and discharging power and energy Value attributes; photovoltaic equipment entities include power generation, energy Value attributes; entities, attributes and their relationships at all levels are similar to energy usage scenarios; 3) Energy network entities The energy network entity includes four secondary entities: power network, heat network, cooling network and natural gas network; the power network entity includes voltage level, transmission capacity and transmission efficiency attributes; the heat network entity includes transmission temperature and heat loss rate attributes; the cooling network entity includes transmission temperature and cooling efficiency attributes; the natural gas entity includes pressure stability and gas transmission volume attributes; the relationship between entities at all levels is similar to the energy usage scenario; 4) Time series data entity The time series data entity includes four secondary entities: seasonal characteristics, meteorological characteristics, curve characteristics, and time series characteristics; The seasonal feature entity contains four attributes: spring, summer, autumn, and winter; the meteorological feature entity contains four attributes: cloudy, sunny, rainy, and snowy; The curve feature entity includes peak-to-valley difference, daily load rate, daily load fluctuation rate, and peak value attributes, among which the peak-to-valley difference and peak value are attribute sets. The peak-to-valley difference attribute set includes the peak power change rate within the time series, the peak occurrence time within the time series, the valley power change rate within the time series, and the valley occurrence time within the time series attributes. The peak attribute set includes single peak, double peak, and flat peak attributes. The time series feature entity includes attributes such as power change rate within the time series, power mean, and power ratio. 5) Scheduling Strategy Entity The scheduling strategy entity includes the scheduling strategy number attribute, including three secondary entities: scheduling target, output plan and control change; the scheduling target entity includes three secondary entities: the highest crop growth comfort, the highest daily operation efficiency and the lowest operation cost; the output plan entity includes the scheduling time of the day's output plan, the power purchase power, the power sale power, the gas purchase power, the gas turbine output power, the gas boiler output power, the electric boiler output power, the air source heat pump output power, the absorption chiller output power, the electric chiller output power, the battery, the heat storage tank and the cold storage tank charging power, the battery, the heat storage tank and the cold storage tank discharging power, the space electric field load, the physical insecticidal lamp load, the plasma nitrogen fixation load, the electric drainage load, the agricultural electric vehicle load, and the fill light load power attributes; The control change entity includes two third-level entities: the long-time scale control change within the day and the short-time scale control change within the day; the long-time scale control change entity within the day includes the gas purchase power change, the gas turbine output electric power change, the gas boiler output thermal power change, the electric boiler output thermal power change, the air source heat pump output thermal power change, the absorption chiller output cold power change, and the electric chiller output cold power change attributes; the short-time scale control change entity within the day includes the power purchase power change, the power sales power change, and the fill light load energy power change attributes; 6) Operational Status Entity The operation status entity includes three secondary entities: photovoltaic absorption, crop growth environment and energy utilization; the photovoltaic absorption entity includes absorption ratio and abandoned light cost attributes; the crop growth environment entity includes crop growth light comfort, crop growth thermal comfort and crop growth environment comprehensive comfort attributes; the energy utilization entity includes energy utilization efficiency, Efficiency and useful energy conversion efficiency attributes; entities, attributes and their relationships at all levels are similar to scheduling strategy entities.

6. The method for generating a control strategy for an agricultural park energy supply and utilization system driven by a knowledge graph according to claim 5 is characterized in that: The knowledge graph-driven agricultural park energy supply and utilization system control strategy retrieval method is: Input the basic information of the system scenario to be optimized into the obtained knowledge graph, and retrieve the historical optimization strategies that meet the optimization requirements in the knowledge graph through information analysis, strategy feature matching, and scheduling strategy query. If the historical optimal strategy is not retrieved, it is solved through the optimization algorithm. After the strategy optimization process is completed, the system automatically extracts the structured information of the optimization strategy and imports it into the case knowledge base to update the knowledge graph, as follows: S2.1 Input basic information of operation scenario data; S2.2 Information analysis: Extract features and make logical judgments on the prediction data of the scenario to be optimized based on the basic information of the operation scenario data; S2.3 Query the corresponding label of each indicator characteristic value in the characteristic label library, and match the judgment result with the entities and relationships stored in the knowledge graph of the agricultural park energy supply and utilization system, and combine the input basic information to form a characteristic information matrix that fully describes the input information; S2.4 Strategy feature matching: Describe and confirm the optimization scenario based on the data feature attributes extracted by information analysis and the data entity matching results, use the knowledge search engine to find the matching knowledge path in the knowledge graph, input the feature information matrix of the predicted data into the optimization strategy search in the knowledge graph history database, and generate the optimization scheduling strategy number search result; S2.5 Scheduling strategy query, query the historical strategy closest to the current operation scenario to be optimized in the knowledge graph historical optimization strategy library. If the optimal strategy cannot be found in the knowledge graph, the scenario to be optimized will be solved by the optimization algorithm and updated to the knowledge graph of the agricultural park energy supply and utilization system, and the feature label library will be updated. At the same time, the optimization strategy will be output and a new number will be added, the knowledge graph strategy library will be updated, and the optimization scheduling plan will be output; if the optimal strategy is found in the knowledge graph, the strategy number retrieved from the knowledge graph will be output, and the optimization strategy corresponding to the strategy number will be queried in the historical strategy library, and the optimization scheduling plan will be output.

7. A method for extracting characteristic indicators of structured data of an agricultural park energy supply and utilization system based on the knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method described in claim 2.

8. A method for constructing a structured data feature label library of an agricultural park energy supply and utilization system based on interval processing based on the knowledge graph-driven agricultural park energy supply and utilization system control strategy generation method described in claim 3.

9. A method for constructing a knowledge graph for an agricultural park energy supply and utilization system based on the method for generating a control strategy for an agricultural park energy supply and utilization system driven by the knowledge graph as described in claim 4.

10. A knowledge graph driven agricultural park energy supply and utilization system control strategy generation system based on the knowledge graph driven agricultural park energy supply and utilization system control strategy generation method described in claim 1.

Images