New energy and load prediction error cause mining method based on knowledge graph

Abstract

The invention discloses a new energy and load prediction error cause mining method based on a knowledge graph, and relates to the related technical field of new energy management, and the method comprises the steps: building an energy-load prediction knowledge graph based on new energy and load data; new energy power generation data, load data and environment data are collected and updated to the energy-load prediction knowledge graph; executing mode recognition on the data nodes, and determining prediction error data nodes; identifying, analyzing and predicting error data nodes, and determining relevance; quantitatively evaluating a plurality of prediction error influence factors; carrying out weight distribution to obtain an error source distribution result; and adjusting and optimizing the new energy load prediction model. The technical problems of load prediction errors and inaccurate error cause determination caused by complex influence factors and imperfect prediction models in existing new energy load prediction are solved, and the technical effects of improving the accuracy of load prediction and enabling power grid energy distribution to be balanced are achieved.

Description

Technical Field

[0001] The present application relates to fields related to new energy management technology, and specifically to a method for mining the causes of new energy and load forecasting errors based on a knowledge graph. Background Art

[0002] With the transformation of the global energy structure and the accelerated development of smart grids, the status of new energy and load forecasting in energy management and power dispatching has become increasingly prominent. Although new energy such as wind energy and solar energy are clean and renewable, the volatility and uncertainty of their power generation have brought challenges to the stable operation of the power system. At the same time, the power load is affected by many factors, including weather changes, economic conditions, social activities, etc., and the difficulty of its prediction has increased accordingly. Accurate new energy and load forecasting are crucial for the balanced dispatch of the power system, the optimal allocation of energy, and the effective operation of the market. However, in actual applications, due to the imperfection of the prediction model, the instability of data quality, and the complexity of influencing factors, there are often certain errors in the new energy and load forecasting. These errors may not only lead to the waste of power resources, but also affect the safety and stability of the power system.

[0003] Therefore, in the current technologies related to new energy load forecasting, there are technical problems such as errors in load forecasting and inaccurate determination of the causes of the errors due to complex influencing factors and imperfect prediction models. Summary of the invention

[0004] This application provides a method for mining the causes of new energy and load forecasting errors based on a knowledge graph, and adopts technical means such as establishing a knowledge graph, constructing a forecasting model, and calculating the similarity of feature vectors to solve the technical problems existing in the existing new energy load forecasting, such as errors in load forecasting and inaccurate determination of the causes of errors due to complex influencing factors and imperfect forecasting models. This achieves the technical effect of improving the accuracy of load forecasting and balancing the energy distribution of the power grid.

[0005] The present application provides a method for mining the causes of new energy and load forecasting errors based on a knowledge graph, the method comprising: establishing an energy-load forecasting knowledge graph based on new energy and load data, the energy-load forecasting knowledge graph comprising power generation data nodes, load data nodes, and environmental factor nodes; using sensor equipment to collect new energy power generation data, load data, and environmental data, and updating them to the energy-load forecasting knowledge graph; performing pattern recognition on data nodes in the energy-load forecasting knowledge graph, determining prediction error data nodes, and adding them to the energy-load forecasting knowledge graph; identifying and analyzing prediction error data nodes, determining the correlation between prediction error data nodes and the power generation data nodes, the load data nodes, and the environmental factor nodes, and obtaining a set of prediction error influencing factors; quantitatively evaluating multiple prediction error influencing factors in the prediction error influencing factor set, and determining a first error source, a second error source, and up to an Nth error source based on the quantitative evaluation results, wherein N is a positive integer greater than or equal to 1; performing weight allocation according to the first error source, the second error source, ... the Nth error source, to obtain an error source allocation result; and adjusting and optimizing a new energy load forecasting model through the error source allocation result.

[0006] In a possible implementation, the energy-load forecasting knowledge graph is established based on new energy and load data, and the following processing is performed: obtaining historical power generation record data, historical power load data and environmental data of multiple time series of target new energy power generation equipment; preprocessing the historical power generation record data, the historical power load data and the environmental data; determining multiple data nodes of the knowledge graph based on the preprocessing results; connecting the multiple data nodes through relationship edges to form the energy-load forecasting knowledge graph.

[0007] In a possible implementation, the multiple data nodes are connected by relationship edges to form the energy-load prediction knowledge graph, and the following processing is performed: determining the relationship between the multiple data nodes, including the power generation-load relationship, the power generation-environment relationship and the environment-load relationship; connecting the multiple data nodes as core nodes to each other through the relationship edges to form a knowledge graph network; combining the power generation-load relationship, the power generation-environment relationship and the environment-load relationship to establish the energy-load prediction knowledge graph.

[0008] In a possible implementation, pattern recognition is performed on the data nodes in the energy-load forecasting knowledge graph to determine the prediction error data node, and the following processing is also performed: the attribute characteristics of each node in the energy-load forecasting knowledge graph and the relationship characteristics between each node are extracted; based on the attribute characteristics and the relationship characteristics, each node is clustered and grouped to determine the prediction error between each node; the prediction error of each node is comprehensively evaluated to determine the prediction error data node.

[0009] In a possible implementation, the nodes are clustered and grouped based on the attribute features and the relationship features to determine the prediction errors between the nodes, and the following processing is performed: the attribute features and the relationship features are integrated to determine the comprehensive feature vector of each node; the similarity of the comprehensive feature vectors between the nodes is calculated to obtain multiple node groups; the average prediction errors of the nodes in each node group are compared to identify nodes with high prediction errors and nodes with low prediction errors; and the prediction errors between the nodes are obtained based on the clustering grouping results.

[0010] In a possible implementation, the identification and analysis of the prediction error data node, the determination of the correlation between the prediction error data node and the power generation data node, the load data node, and the environmental factor node, and the acquisition of the prediction error influencing factor set, also performs the following processing: according to the attribute characteristics and the relationship characteristics, the data nodes associated with the prediction error data node are extracted to obtain associated nodes, including power generation nodes, load nodes, temperature nodes, humidity nodes, and wind speed nodes; the correlation coefficient between the associated nodes and the prediction error is calculated to construct an association network between the prediction error data node and the associated nodes; multiple influencing factors are determined based on the association network to determine the prediction error influencing factor set.

[0011] In a possible implementation, the quantitative evaluation comprises performing a plurality of prediction error influencing factors within the prediction error influencing factor set, determining a first error source, a second error source, and up to the Nth error source based on the quantitative evaluation result, and further performing the following processing: quantifying a plurality of error influencing factors; constructing an evaluation model with the quantified error influencing factors as input and the prediction error as output; calculating the contribution of each prediction error influencing factor to the prediction error through the evaluation model; and determining the first error source, up to the Nth error source, based on the evaluation result.

[0012] In a possible implementation, the new energy load prediction model is optimized through the error source allocation result, and the following processing is also performed: an initial new energy load prediction model is established based on a neural network; a new energy load prediction model is constructed in combination with the initial new energy prediction model using the historical power generation record data as input, the historical power load data as output and the environmental data as control parameters; the proportion of the model's control parameters is adjusted according to the error source allocation result, the model architecture is optimized, and the new energy load prediction optimization model is determined; and load prediction is performed using the new energy load prediction optimization model.

[0013] It is intended to establish an energy-load forecasting knowledge graph based on new energy and load data through the knowledge graph-based new energy and load forecasting error cause mining method proposed in this application; use sensor equipment to collect new energy power generation data, load data and environmental data, and update them to the energy-load forecasting knowledge graph; perform pattern recognition on the data nodes in the energy-load forecasting knowledge graph, determine the prediction error data nodes, and add them to the energy-load forecasting knowledge graph; identify and analyze the prediction error data nodes, determine the correlation between the prediction error data nodes and the power generation data nodes, the load data nodes, and the environmental factor nodes, and obtain a set of prediction error influencing factors; quantitatively evaluate multiple prediction error influencing factors in the prediction error influencing factor set, and determine the first error source, the second error source, and the Nth error source based on the quantitative evaluation results, where N is a positive integer greater than or equal to 1; perform weight allocation based on the first error source, the second error source...the Nth error source to obtain an error source allocation result; and adjust and optimize the new energy load forecasting model through the error source allocation result. The technical problems existing in the existing new energy load forecasting, such as complex influencing factors and imperfect forecasting models, which lead to errors in load forecasting and inaccurate determination of the causes of the errors, are solved, achieving the technical effect of improving the accuracy of load forecasting and balancing the energy distribution in the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solution of the embodiment of the present disclosure, the accompanying drawings of the embodiment of the present disclosure will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0015] Figure 1 A schematic diagram of the process of the method for mining the causes of new energy and load forecasting errors based on the knowledge graph provided in the embodiment of the present application;

[0016] Figure 2 A schematic diagram of the process of determining the prediction error between nodes in the knowledge graph-based new energy and load prediction error cause mining method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0018] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0019] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0020] The present application embodiment provides a method for mining the causes of new energy and load forecasting errors based on a knowledge graph, such as Figure 1 As shown, the method includes:

[0021] Knowledge graph technology has achieved remarkable results in many fields with its powerful knowledge representation and reasoning capabilities. By representing information such as entities, attributes and relationships in the form of graphs, knowledge graphs can intuitively display the internal connections and patterns of data, providing strong support for data analysis. In the field of new energy and load forecasting, the application of knowledge graphs helps to systematically sort out and integrate various factors that affect forecasting errors, and then reveal the deep-seated causes of errors. The new energy and load forecasting error cause mining method based on knowledge graphs aims to build a comprehensive and systematic error cause knowledge base. By mining and analyzing the correlation between data, the key factors affecting forecasting accuracy are discovered, which not only helps to improve the accuracy and stability of the forecasting model, but also provides a scientific basis for the optimal scheduling and decision-making of the power system.

[0022] Step S100, establish an energy-load forecasting knowledge graph based on new energy and load data, and the energy-load forecasting knowledge graph includes power generation data nodes, load data nodes, and environmental factor nodes. Establish an energy-load forecasting knowledge graph based on new energy and load data. In the energy-load forecasting knowledge graph established based on new energy and load data, the power generation data nodes, load data nodes, and environmental factor nodes each represent different entities and information. These nodes are interconnected by edges (relationships), and together constitute a complex relationship network that describes new energy power generation, power load, and their influence by environmental factors. Specifically, the power generation data node mainly represents the relevant data of new energy power generation, including key indicators such as the power generation, power generation efficiency, and power generation cost of renewable energy such as wind power, solar power, and hydropower. It stores real-time or historical data of new energy power generation, and provides necessary input information for the prediction model; the load data node The nodes mainly record relevant information of power load, including power demand, load peak, load distribution, etc., reflecting the real-time and historical demand for electricity by users in the power system. By analyzing the load data nodes, the changing trend of future power demand can be predicted, providing an important basis for the supply and demand balance and scheduling decision-making of the power system; the environmental factor nodes represent various environmental factors affecting renewable energy power generation and power load, such as weather conditions (temperature, humidity, wind speed, wind direction, light intensity, etc.), seasonal changes, holidays, etc., which are not only related to the power generation data nodes and the load data nodes, but also have complex interactions with each other. Environmental factors have a significant impact on the volatility of renewable energy power generation and the changes in power load.

[0023] In a possible implementation, step S100 further includes step S110, obtaining historical power generation record data, historical power load data and environmental data of multiple time series of the target new energy power generation equipment. When obtaining historical power generation record data, historical power load data and environmental data of multiple time series of the target new energy power generation equipment, it is actually collecting a series of historical data related to new energy power generation and power system operation. It also includes step S120, pre-processing the historical power generation record data, the historical power load data and the environmental data. Before data analysis or modeling, a series of processing operations are performed on these raw data to improve the quality and applicability of the data. Specifically, data cleaning is performed on the historical power generation record data, the historical power load data and the environmental data. Considering that the new energy power generation, power load and environmental data may come from different data sources and acquisition systems, data integration is also required in the pre-processing stage to ensure that the timestamps between different data sets are aligned for time series analysis. It also includes step S130, determining multiple data nodes of the knowledge graph based on the pre-processing results. According to the data after pre-processing operations such as cleaning, integration, and alignment, key information entities are extracted and used as nodes in the knowledge graph, representing the basic concepts or objects in the knowledge graph. They are interconnected through relationships and together constitute the framework of the entire knowledge graph, which may include power generation equipment nodes, power load nodes, environmental factor nodes, and event nodes. Specifically, the power generation equipment nodes represent various new energy power generation equipment, such as wind turbines, solar photovoltaic panels, etc., and contain attribute information such as equipment type, location, and installed capacity; the power load nodes reflect the power load conditions in different time periods or regions, including attributes such as load quantity and load change trend; the environmental factor nodes describe various environmental factors that affect new energy power generation and power load, such as wind speed, light intensity, temperature, etc., and contain real-time or historical environmental data; the event nodes represent important events related to new energy power generation and power load, such as equipment failure, maintenance records, policy adjustments, etc., which are helpful for analyzing the impact of events on new energy power generation and power load. It also includes step S140, connecting the multiple data nodes through relationship edges to form the energy-load prediction knowledge graph. The preprocessed data entities are organized and represented through the association relationships between them to form a structured knowledge network. In the knowledge graph, nodes represent entities (such as power generation equipment, power loads, environmental factors, etc.), and edges represent various relationships between these entities. Specifically, it is necessary to clarify the various relationships that may exist between different data nodes. For example, there may be a power supply relationship between the power generation equipment node and the power load node, and there may be an influence relationship between the environmental factor node and the power generation equipment node. The extracted relationships are used as edges to connect the corresponding data nodes. In this way, a knowledge graph network composed of nodes and edges is formed, which constitutes the energy-load prediction knowledge graph.

[0024] In a possible implementation, step S140 further includes step S141, determining the relationship between the multiple data nodes, including the power generation-load relationship, the power generation-environment relationship, and the environment-load relationship. The power generation-load relationship is reflected in the balance and stability of the power system. The load refers to the total power consumption of various electrical equipment on the power grid. It changes in real time and changes with the opening and closing of electrical equipment. The power generation power refers to the rate at which the power station or power generation unit transmits electrical energy to the power grid. In a stable power grid, the load and power generation power need to be balanced. If the load is greater than the power generation power, the grid voltage will drop, which may cause equipment failure and power outages; conversely, if the power generation power is greater than the load, the grid voltage will rise, which may also cause equipment failure; the power generation-environment relationship is mainly reflected in the impact of the environment on the selection of power generation methods and the power generation process, as well as the impact of power generation activities on the environment. In terms of power generation, environmental factors such as climate and resource distribution will affect the choice of power generation mode. For example, the use of renewable energy such as wind power and solar energy is greatly affected by regional and climatic conditions. The environment-load relationship can be understood from the impact of the environment on electricity demand and the impact of electricity load on the environment. Environmental factors such as temperature and seasonal changes will affect people's electricity demand, thereby affecting the load of the power grid. For example, in high temperatures in summer, the power consumption of refrigeration equipment such as air conditioners will increase significantly, resulting in an increase in the load of the power grid. At the same time, changes in electricity load may also have an impact on the environment. For example, large-scale power consumption may lead to increased energy consumption, thereby exacerbating environmental pollution and climate change. It also includes step S142, connecting the multiple data nodes as core nodes through the relationship edges to form a knowledge graph network. In the process of constructing the knowledge graph, different data nodes are regarded as core entities in the knowledge graph, and the relationship between them is used to create connections, thereby forming a complex network structure. It also includes step S143, combining the power generation-load relationship, the power generation-environment relationship and the environment-load relationship to establish the energy-load prediction knowledge graph. Taking the power generation-load relationship, power generation-environment relationship and environment-load relationship as the core components, a knowledge graph network for predicting energy production and load demand is constructed. Specifically, this information is organized and represented in the form of a knowledge graph. Each entity (such as power station, load center, environmental factor, etc.) is regarded as a node, and the relationships between them (such as the relationship between power generation and load demand, the impact of power generation activities on the environment, etc.) are connected through relationship edges. A complex network structure containing multiple nodes and edges is constructed to describe and display the interaction and influence between power generation, load and environment, which is the energy-load prediction knowledge graph.

[0025] Step S200, using sensor equipment to collect new energy power generation data, load data and environmental data, and update them to the energy-load prediction knowledge graph. Use sensors to collect new energy power generation data, load data and environmental data in real time, for example, the operating status of new energy power generation equipment, power generation, real-time changes in power load, ambient temperature, humidity, wind speed, etc., and use the standardized data to update the energy-load prediction knowledge graph, including adding new nodes and edges to the knowledge graph to reflect new data and relationships; at the same time, it may also be necessary to update or adjust existing nodes and edges to reflect changes in data or new analysis results. By updating the knowledge graph, the knowledge graph can continuously absorb new data and information to maintain its timeliness and accuracy.

[0026] Step S300, perform pattern recognition on the data nodes in the energy-load forecasting knowledge graph, determine the prediction error data nodes, and add them to the energy-load forecasting knowledge graph. Performing pattern recognition on the data nodes in the energy-load forecasting knowledge graph refers to analyzing the relationships and trends between the data nodes to reveal potential patterns and rules. In the energy-load forecasting knowledge graph, each data node may contain a prediction value related to power generation, load or environment. By comparing with the actual observed value, the prediction error can be calculated. The prediction error data nodes refer to those nodes where there is a significant deviation between the predicted value and the actual value. Once the prediction error data nodes are determined, they will be added to the energy-load forecasting knowledge graph.

[0027] In a possible implementation, step S300 further includes step S310, extracting the attribute characteristics of each node in the energy-load forecasting knowledge graph and the relationship characteristics between each node. In the process of constructing and analyzing the knowledge graph, each node in the graph and the connection relationship between the nodes are deeply mined to obtain its inherent attributes and relationship characteristics. Specifically, the attribute characteristics of the node usually refer to the information describing the nature of the node itself. Each node may represent different entities, such as power stations, load centers, environmental factors, etc. The attribute characteristics of the node may include power generation, load, environmental parameters (such as temperature, humidity, wind speed, etc.), geographical location, equipment type, operating status, etc.; the relationship characteristics between nodes refer to the information describing the interaction or association between nodes. The relationship between nodes may include the supply and demand relationship between the power station and the load center, the influence of environmental factors on power generation or load, and the collaborative relationship between different power stations. It is represented by the type, weight or direction of the edge, thereby revealing the complex association and mutual influence between the nodes. It also includes step S320, based on the attribute characteristics and the relationship characteristics, clustering and grouping the nodes to determine the prediction error between the nodes. According to the similarity or distance between the attribute characteristics and relationship characteristics of the nodes, the nodes are divided into different groups or categories. Specifically, clustering grouping can identify nodes with similar behaviors, attributes or influencing factors. For example, power plant nodes with similar power generation change patterns are classified into one category, or load center nodes affected by the same environmental factors are classified into one category. On the basis of clustering grouping, the prediction errors between nodes in different groups or categories can be further analyzed. The prediction error can be calculated by comparing the actual observed value of the node with the predicted value. For example, the error size can be measured using indicators such as mean square error or mean absolute error. It also includes step S330, comprehensively evaluating the prediction error of each node and determining the prediction error data node. After obtaining the prediction error of each node, we can further perform statistics and analysis on these errors. Based on the statistical and analysis results, the prediction error of each node can be comprehensively evaluated. By setting a certain threshold or standard, the node whose prediction error exceeds the acceptable range can be identified, that is, the prediction error data node.

[0028] In one possible implementation, Figure 2As shown, step S320 further includes step S321, integrating the attribute features and the relationship features to determine the comprehensive feature vector of each node. In the analysis process of the energy-load forecasting knowledge graph, the attribute features and relationship features of the node are fused to form a comprehensive feature representation, that is, a comprehensive feature vector. Specifically, the process of integrating attribute features and relationship features usually involves two steps: feature selection and feature fusion. Feature selection refers to selecting the most representative and predictive features from a large number of features to simplify the model and improve the prediction accuracy; feature fusion is to combine these selected features to form a comprehensive feature vector. The purpose of determining the comprehensive feature vector of each node is to provide a comprehensive and comprehensive node representation, so as to more accurately describe the state and behavior of the node. It also includes step S322, calculating the similarity of the comprehensive feature vectors between the nodes, and obtaining multiple node groups. In the energy-load forecasting knowledge graph analysis, by comparing the similarity of the comprehensive feature vectors between different nodes, similar nodes are grouped together to form multiple node groups. Specifically, a similarity calculation method (such as cosine similarity, Euclidean distance, etc.) is used to compare the comprehensive feature vectors between different nodes. For example, cosine similarity is often used to measure the similarity of two vectors in direction, while Euclidean distance measures the distance between two vectors in space. By calculating the similarity between each pair of nodes, a similarity matrix can be constructed, in which each element represents the similarity value between the corresponding node pairs. Based on the similarity matrix, a clustering algorithm (such as K-means, hierarchical clustering, etc.) or other grouping methods are used to group similar nodes together. Finally, through this process, multiple node groups can be obtained, and the nodes in each node group have high similarity in comprehensive features. It also includes step S323, comparing the average prediction errors of the nodes in each node group, and identifying high prediction error nodes and low prediction error nodes. It also includes step S324, based on the clustering grouping results, obtaining the prediction errors between each node. In the energy-load forecasting knowledge graph analysis, after clustering and grouping the nodes according to their comprehensive feature vectors, the prediction errors between the nodes in each group are further calculated and analyzed. Specifically, the prediction errors between the nodes in each group are calculated using appropriate error evaluation indicators.

[0029] Step S400, identify and analyze the prediction error data nodes, determine the correlation between the prediction error data nodes and the power generation data nodes, the load data nodes, and the environmental factor nodes, and obtain a prediction error influencing factor set. The prediction error data nodes are identified and analyzed to determine the correlation between these prediction error data nodes and the power generation data nodes, the load data nodes, and the environmental factor nodes. For example, it can be checked whether there is a certain specific correlation pattern between the prediction error data nodes and the power generation data nodes, such as certain power generation type nodes may cause the prediction error to increase under certain conditions. Similarly, the relationship between the prediction error data nodes and the load data nodes and the environmental factor nodes can also be analyzed. Through the above analysis, a prediction error influencing factor set is obtained, which includes the power generation data nodes, load data nodes, and environmental factor nodes that have significant correlations with the prediction error data nodes. These influencing factors may be the key reasons for the increase in prediction errors.

[0030] In a possible implementation, step S400 further includes step S410, extracting data nodes associated with the prediction error data node according to the attribute characteristics and the relationship characteristics, and obtaining associated nodes, including power generation nodes, load nodes, temperature nodes, humidity nodes, and wind speed nodes. By analyzing the attribute characteristics and relationship characteristics of the nodes, other data nodes that are significantly associated with the prediction error data node are found, including power generation nodes, load nodes, temperature nodes, humidity nodes, and wind speed nodes. It also includes step S420, calculating the correlation coefficient between the associated node and the prediction error, and constructing an associated network between the prediction error data node and the associated node. The correlation coefficient between the associated node and the prediction error is calculated using the Pearson correlation coefficient or the Spearman rank correlation coefficient to understand whether there is a relationship between the numerical change of the associated node and the prediction error, and to give the strength and direction of this relationship, and to obtain a series of numerical results, reflecting the degree of association between different associated nodes and the prediction error. For example, a certain power generation node may have a high positive correlation with the prediction error, which means that the change in power generation of the node will directly lead to an increase in the prediction error; while another humidity node may have a negative correlation with the prediction error, indicating that the change in humidity may help to reduce the prediction error. Based on the correlation coefficient, an association network of the prediction error data node and the associated node is constructed, wherein the node represents the prediction error data node and the associated node (such as the power generation node, the load node, the temperature node, etc.), and the edge represents the correlation relationship between these nodes; the weight of the edge can be determined by the value of the correlation coefficient, and the larger the weight, the closer the association between the nodes. It also includes step S430, determining multiple influencing factors based on the association network, and determining the prediction error influencing factor set. Extract multiple key influencing factors from the association network to form a prediction error influencing factor set, which includes multiple variables that are highly correlated with the prediction error.

[0031] Step S500, quantitatively evaluate multiple prediction error influencing factors in the prediction error influencing factor set, and determine the first error source, the second error source, and the Nth error source based on the quantitative evaluation result, wherein N is a positive integer greater than or equal to 1. Quantitatively evaluate each influencing factor in the prediction error influencing factor set to measure the specific relationship between each factor and the prediction error. For example, the correlation coefficient between each influencing factor and the prediction error can be calculated, or the contribution of each factor to the prediction error can be estimated by building a model; based on the results of the quantitative evaluation, the ranking or priority of different error sources is determined, and the contribution of each factor to the prediction error, the stability of the influence, and the adjustability are comprehensively considered. The first error source is usually the factor with the greatest and most significant influence on the prediction error, while the influence of the second, third, and other error sources decreases in turn.

[0032] In a possible implementation, step S500 further includes step S510, quantifying multiple error influencing factors. Specific numerical measurements are performed on multiple factors affecting prediction accuracy to more accurately understand the influence of each factor on the prediction error, so that the prediction model can be optimized in a targeted manner. It also includes step S520, taking the quantified error influencing factors as input and the prediction error as output to construct an evaluation model. The error influencing factors that have been numerically measured are used as input features of the model, and the prediction error is used as the output target of the model. Through a certain algorithm, a model that can evaluate the prediction error is constructed. It also includes step S530, calculating the contribution of each prediction error influencing factor to the prediction error through the evaluation model. Based on the feature importance evaluation results provided by the model, we can calculate the contribution of each prediction error influencing factor to the prediction error, which is usually achieved by normalizing the feature importance score to a certain range (such as 0 to 1) to ensure that the sum of the contribution of all features is 1 or 100%, and the higher the contribution of the factor, the greater the impact on the prediction error. It also includes step S540, determining the first error source to the Nth error source based on the evaluation result. The first error source to the Nth error source are determined according to the evaluation result of the model, and the contribution of the error source to the prediction error decreases successively.

[0033] Step S600, weights are assigned according to the first error source, the second error source, ... the Nth error source, to obtain an error source assignment result. After identifying multiple prediction error sources, a corresponding weight is assigned to each error source according to its contribution or importance to the prediction error, aiming to quantify the relative impact of different error sources, so as to more accurately understand which factors have a greater impact on the prediction error, thereby being able to optimize the prediction model in a targeted manner. Specifically, after obtaining the error source assignment result, the comprehensive impact of different error sources on the prediction error can be further analyzed to identify the main error sources and secondary error sources.

[0034] Step S700, adjust and optimize the new energy load forecasting model through the error source allocation result. According to the weight and contribution of each error source, improve the structure, parameters or data processing method of the forecasting model in a targeted manner to improve the forecasting accuracy and reduce the error. Specifically, according to the error source allocation result, if it is found that one or some error sources contribute more to the forecasting error, consider adjusting the structure of the model to better capture the influence of these factors. For example, if the wind speed and wind direction have a significant impact on the wind power load forecast, you can add features or components related to wind speed and wind direction to the model; for each error source, you can try to adjust the parameters of the model, such as learning rate, regularization coefficient, etc., to optimize the performance of the model on a specific error source; according to the error source, you can optimize the data preprocessing step; if multiple error sources have a significant impact on the forecasting error, you can consider using ensemble learning or hybrid model methods to combine multiple single models to comprehensively utilize the advantages of different models and improve the forecasting accuracy. After the model is adjusted and optimized, it is necessary to monitor the forecasting performance of the model in real time, and provide feedback and adjustment based on the new error source allocation results to ensure that the model can continue to adapt to changes in the new energy load.

[0035] In a possible implementation, step S700 further includes step S710, establishing an initial new energy load forecasting model based on a neural network. A new energy load forecasting model is constructed using the principle and structure of a neural network for load forecasting. It also includes step S720, using the historical power generation record data as input, the historical power load data as output and the environmental data as control parameters, combined with the initial new energy forecasting model, to construct a new energy load forecasting model. It also includes step S730, adjusting the proportion of the model's control parameters according to the error source allocation result, optimizing the model architecture, and determining a new energy load forecasting optimization model. Adjust the proportion of the model's control parameters based on the distribution results of the error sources, focus on adjusting the control parameters related to the main error sources, increase their proportion, and make the model pay more attention to these error sources during training. For secondary error sources, the proportion of related control parameters can be appropriately reduced to maintain the balance and stability of the model; make targeted improvements to the model architecture, for example, increase or decrease the number of model layers, adjust the number of neurons in each layer, change the type of activation function, etc.; you can also consider introducing new model components or structures, such as attention mechanisms, residual connections, etc., to further improve the performance of the model. Through multiple iterations and adjustments, determine the optimization model for new energy load forecasting and improve the accuracy and reliability of the forecasting model.

[0036] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A method for mining the causes of new energy and load forecasting errors based on knowledge graphs, characterized in that: The method comprises: Establishing an energy-load forecasting knowledge graph based on new energy and load data, wherein the energy-load forecasting knowledge graph includes power generation data nodes, load data nodes, and environmental factor nodes; Using sensor equipment to collect new energy generation data, load data and environmental data, and update them to the energy-load prediction knowledge graph; Performing pattern recognition on data nodes in the energy-load forecasting knowledge graph, determining prediction error data nodes, and adding them to the energy-load forecasting knowledge graph; Identify and analyze the prediction error data node, determine the correlation between the prediction error data node and the power generation data node, the load data node, and the environmental factor node, and obtain a prediction error influencing factor set; Quantitatively evaluating multiple prediction error influencing factors in the prediction error influencing factor set, and determining a first error source, a second error source, and finally an Nth error source based on the quantitative evaluation result, wherein N is a positive integer greater than or equal to 1; Performing weight allocation according to the first error source, the second error source, ... the Nth error source to obtain an error source allocation result; The new energy load forecasting model is adjusted and optimized through the error source allocation result.

2. The method for mining causes of new energy and load forecasting errors based on knowledge graph according to claim 1, characterized in that: The method of establishing an energy-load forecasting knowledge graph based on new energy and load data includes: Obtain historical power generation record data, historical power load data and environmental data of multiple time series of target new energy power generation equipment; Preprocessing the historical power generation record data, the historical power load data and the environmental data; Determine multiple data nodes of the knowledge graph based on the preprocessing results; The multiple data nodes are connected through relationship edges to form the energy-load forecasting knowledge graph.

3. The method for mining causes of new energy and load forecasting errors based on knowledge graph according to claim 2, characterized in that: The method of connecting multiple data nodes through relationship edges to form the energy-load forecasting knowledge graph includes: Determining the relationships between the plurality of data nodes, including a power generation-load relationship, a power generation-environment relationship, and an environment-load relationship; The multiple data nodes are connected to each other as core nodes through the relationship edges to form a knowledge graph network; The energy-load prediction knowledge graph is established by combining the power generation-load relationship, the power generation-environment relationship and the environment-load relationship.

4. The method for mining causes of new energy and load forecasting errors based on knowledge graph according to claim 1, characterized in that: The method of performing pattern recognition on the data nodes in the energy-load forecasting knowledge graph to determine the forecast error data nodes includes: Extracting attribute features of each node and relationship features between nodes in the energy-load forecasting knowledge graph; Based on the attribute features and the relationship features, clustering and grouping the nodes to determine the prediction errors between the nodes; Comprehensively evaluate the prediction error of each node and determine the prediction error data node.

5. The method for mining causes of new energy and load forecasting errors based on knowledge graph according to claim 4, characterized in that: The method of clustering and grouping the nodes based on the attribute features and the relationship features and determining the prediction errors between the nodes includes: Integrate the attribute features and the relationship features to determine a comprehensive feature vector for each node; Calculate the similarity of the comprehensive feature vectors between the nodes and obtain multiple node groups; Compare the average prediction errors of nodes within each node group and identify nodes with high prediction errors and nodes with low prediction errors; Based on the clustering grouping results, the prediction error between each node is obtained.

6. The method for mining causes of new energy and load forecasting errors based on knowledge graph according to claim 4, characterized in that: The method of identifying and analyzing the prediction error data node, determining the correlation between the prediction error data node and the power generation data node, the load data node, and the environmental factor node, and obtaining a prediction error influencing factor set includes: According to the attribute features and the relationship features, extract the data nodes associated with the prediction error data nodes to obtain associated nodes, including power generation nodes, load nodes, temperature nodes, humidity nodes, and wind speed nodes; Calculating the correlation coefficient between the associated node and the prediction error, and constructing an associated network between the prediction error data node and the associated node; A plurality of influencing factors are determined based on the association network, and the prediction error influencing factor set is determined.

7. The method for mining causes of new energy and load forecasting errors based on knowledge graph according to claim 6, characterized in that: The method of quantitatively evaluating the multiple prediction error influencing factors in the prediction error influencing factor set and determining the first error source, the second error source, and finally the Nth error source based on the quantitative evaluation result includes: Quantify multiple error contributing factors; The evaluation model is constructed with the quantified error influencing factors as input and the predicted error as output; Calculating the contribution of each prediction error influencing factor to the prediction error through the evaluation model; The first to the Nth error sources are determined based on the evaluation result.

8. The method for mining causes of new energy and load forecasting errors based on knowledge graph according to claim 2, characterized in that: The method for optimizing the new energy load forecasting model by using the error source allocation result includes: Establish an initial new energy load forecasting model based on neural network; Using the historical power generation record data as input, the historical power load data as output and the environmental data as control parameters, combined with the initial new energy prediction model, to construct a new energy load prediction model; According to the error source allocation result, the proportion of the control parameters of the model is adjusted, the model architecture is optimized, and the new energy load prediction optimization model is determined; Use the new energy load forecasting optimization model to carry out load forecasting.

Images