New energy equipment price prediction method based on incremental knowledge graph

Abstract

The invention provides a new energy equipment price prediction method based on an incremental knowledge graph. The new energy equipment price prediction method comprises the steps of constructing a four-dimensional dynamic knowledge graph, fusing multi-source data, determining nodes to be updated of the four-dimensional dynamic knowledge graph through a quantum optimization algorithm, and extracting spatio-temporal features in combination with time sequence modeling to update the four-dimensional dynamic knowledge graph. The method comprises the following steps: hierarchically configuring an update strategy based on event change frequency, constructing a key event influence model, predicting an event occurrence rule in combination with a long-short-term memory network model to drive a four-dimensional dynamic knowledge graph to update, and finally, inputting four-dimensional dynamic knowledge graph features into a machine learning prediction model to predict the price of new energy equipment. The problems that a traditional prediction model is low in updating efficiency and poor in timeliness are solved, and the price prediction accuracy and efficiency are improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and energy economics, and in particular to a method for predicting the price of new energy equipment based on incremental knowledge graphs. Background Technology

[0002] As a core area of ​​global energy transition, the price fluctuations of the new energy equipment market are influenced by a combination of factors, including supply chain dynamics, policies and regulations, technological iterations, raw material costs, and market sentiment. With the rapid expansion of new energy industries such as photovoltaics and wind power, accurate prediction of equipment prices is crucial for the production planning, investment decisions, and supply chain management of upstream and downstream enterprises in the industry chain.

[0003] Knowledge graphs, as a technology that integrates multi-source data and models the relationships between entities, are widely used in the analysis and prediction of complex systems. By integrating structured, semi-structured, and unstructured data, they provide a multi-dimensional perspective for predicting the prices of new energy equipment.

[0004] In existing technologies, knowledge graph-based methods for predicting new energy equipment prices mainly achieve basic predictions by constructing static entity relationships, but they have limitations in terms of dynamism and real-time performance. This results in insufficient integration of dynamic factors into the prediction model, as it only models static entity relationships and lacks multi-dimensional integration of time decay effects, policy semantic quantification, and market sentiment propagation, leading to limited explanatory power for complex market environments. Summary of the Invention

[0005] This invention provides a new energy equipment price prediction method based on incremental knowledge graphs to solve the problem of low prediction model update efficiency and poor prediction timeliness in the prior art.

[0006] On the one hand, this invention provides a method for predicting the price of new energy equipment based on incremental knowledge graphs, including: Collect multi-source data and parse the multi-source data to form a four-dimensional dynamic knowledge graph; The set of nodes to be updated in the four-dimensional dynamic knowledge graph is determined based on the quantum optimization algorithm, and the spatiotemporal features of the four-dimensional dynamic knowledge graph are extracted through temporal modeling to update the four-dimensional dynamic knowledge graph. Collect the historical change logs of the four-dimensional dynamic knowledge graph, count the frequency of event changes within a preset time period, use the K-means clustering algorithm to classify the events, and configure the update strategy corresponding to the classification results based on the classification results. Construct a key event impact assessment model and calculate the event criticality value by combining the node association relationship of the four-dimensional dynamic knowledge graph; Input the criticality value into the long short-term memory network model, output the probability and time of the event occurrence, integrate the update strategy, and control the update of the four-dimensional dynamic knowledge graph. Extract the node features and their associations from the four-dimensional dynamic knowledge graph, input them into a machine learning prediction model, and output the price prediction results for new energy equipment.

[0007] Optionally, the step of determining the set of nodes to be updated in the four-dimensional dynamic knowledge graph based on the quantum optimization algorithm, and using temporal modeling to extract the spatiotemporal features of the four-dimensional dynamic knowledge graph to update the four-dimensional dynamic knowledge graph, includes: The update requirements of the four-dimensional dynamic knowledge graph are input into a quadratic unconstrained binary optimization model, solved using quantum annealing, and the set of nodes to be updated is output to update the four-dimensional dynamic knowledge graph. The four-dimensional dynamic knowledge graph is used to generate knowledge graph snapshots according to time series, and the snapshot spatiotemporal features of the knowledge graph snapshots are extracted using a temporal graph convolutional network to update the four-dimensional dynamic knowledge graph.

[0008] Optionally, the step of inputting the update requirements of the four-dimensional dynamic knowledge graph into a quadratic unconstrained binary optimization model, solving it using quantum annealing, outputting the set of nodes to be updated, and updating the four-dimensional dynamic knowledge graph includes: The update requirements of the four-dimensional dynamic knowledge graph are transformed into a four-tuple structure including update nodes, relationships, timestamps, and influence dimensions. Each of the quadruplet structures is assigned a qubit; the qubit is used to characterize whether the corresponding quadruplet needs to be updated. Calculate the update value parameter and coupling relationship parameter of the quadruple structure; Based on the updated value parameter, the coupling relationship parameter, and the qubit, a quadratic unconstrained binary optimization model is constructed. The quantum annealing method is used to solve the quadratic unconstrained binary optimization model to obtain the state corresponding to the qubit, and the set of nodes to be updated is determined based on the state. The nodes in the set of nodes to be updated and the node associations corresponding to the nodes are updated in order to update the four-dimensional dynamic knowledge graph.

[0009] Optionally, the four-dimensional dynamic knowledge graph is used to generate knowledge graph snapshots according to time series, and a temporal graph convolutional network is used to extract the snapshot spatiotemporal features of the knowledge graph snapshots to update the four-dimensional dynamic knowledge graph, including: The four-dimensional dynamic knowledge graph is sliced ​​at preset time intervals to generate knowledge graph snapshots corresponding to multiple timestamps. Set a time window and construct a time-series subgraph that includes the changed nodes and their relationships within the time window; The temporal subgraph is extracted using a temporal graph convolutional network to extract snapshot spatiotemporal features. Based on the spatiotemporal characteristics of the snapshot, the update level of the changed node is calculated; Update the change node according to the update level.

[0010] Optionally, the step of using a temporal graph convolutional network to extract snapshot spatiotemporal features of the temporal subgraph includes: Using graph convolutional networks to aggregate the neighborhood information of nodes, capture the spatial relationship structure between upstream and downstream of the industry chain, and generate spatial features; Gated recurrent units are used to process the temporal changes of node features, capture the time lag effect and transmission path of price fluctuations, and generate time features. By fusing the spatial features and the temporal features, a snapshot spatiotemporal feature containing spatiotemporal dependencies is generated.

[0011] Optionally, the step of collecting historical change logs of the four-dimensional dynamic knowledge graph, statistically analyzing the frequency of event changes within a preset time period, classifying the events using a K-means clustering algorithm, and configuring an update strategy corresponding to the classification results, includes: Collect the historical change log of the four-dimensional dynamic knowledge graph. The historical change log includes entity attribute change records, entity relationship change records and corresponding timestamp information. Using a first preset time window as the statistical period, the number of changes of each event within the preset statistical period is counted, and the frequency of change within the preset time period is calculated. The frequency of change is used as a feature vector, and the K-means clustering algorithm is used to perform clustering, and the hierarchical results are output. If the classification result is a high-frequency event, then a first update strategy is set for the high-frequency event; If the classification result is a low-frequency event, then a second update strategy is set for the low-frequency event; If the classification result is a mid-frequency event, then a third update strategy is set for the mid-frequency event.

[0012] Optionally, the construction of the key event impact assessment model, which calculates the event's criticality value by combining the node relationships of the four-dimensional dynamic knowledge graph, includes: Extract the topological attributes of event-related nodes and calculate the node degree value, which represents the degree of association between nodes in the knowledge graph; Calculate the longest propagation path length of the node, where the longest propagation path length characterizes the propagation depth of the event's impact; The semantic and popularity features of the four-dimensional dynamic knowledge graph are invoked to construct a formula for calculating the criticality value and output the criticality value of the event.

[0013] Optionally, the step of inputting the criticality value into a long short-term memory network model, outputting the event occurrence probability and occurrence time, and integrating the update strategy to control the update of the four-dimensional dynamic knowledge graph includes: Collect the event criticality values ​​within the second preset time window to form a time sequence; The time series is input into a long short-term memory network model, and the predicted value of the event criticality within a first preset time period is output. The predicted value includes the probability of the event occurring and the expected time of occurrence. If the probability of the event occurring exceeds a preset threshold, a priority update mechanism is triggered, and the update priority and timing are determined by combining the update strategy corresponding to the grading result. Based on the expected occurrence time, incremental update preparations for associated nodes are initiated in advance, and the four-dimensional dynamic knowledge graph update is completed when the event occurs.

[0014] Optionally, the node features and relationships of the four-dimensional dynamic knowledge graph are extracted, input into a machine learning prediction model, and the output of new energy equipment price prediction results includes: Extract the four-dimensional feature vector of each node in the four-dimensional dynamic knowledge graph; The graph convolutional network is used to aggregate the neighborhood relationships of nodes and generate node embedding vectors with fused topology. The node embedding vector is concatenated with the time series features to form a prediction feature matrix; The predicted feature matrix is ​​input into a machine learning prediction model, which outputs the predicted price of new energy equipment within a second preset time period. Based on the feedback of the deviation between the predicted value and the actual price, the parameters of the machine learning prediction model and the update frequency of the four-dimensional dynamic knowledge graph are dynamically adjusted.

[0015] Optionally, the step of collecting multi-source data and parsing the multi-source data to form a four-dimensional dynamic knowledge graph includes: Collect supply chain data, market dynamics data, policy text data, industry news data, and market sentiment data for new energy equipment, and integrate them into multi-source data; Extract supplier names, supplier geographical locations, neighboring node propagation time, and cost elasticity coefficients from the multi-source data to form spatial dimension features; Extract the node addition time, node update time, and current time from the multi-source data to form time-dimensional features; Extract unstructured data from the multi-source data to form semantic dimension features; Extract social media discussion popularity, industry forum attention, and investor sentiment index from the multi-source data to form popularity dimension features; A four-dimensional dynamic knowledge graph is constructed based on the spatial dimension features, the temporal dimension features, the semantic dimension features, and the popularity dimension features.

[0016] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the new energy equipment price prediction method based on incremental knowledge graph as described above.

[0017] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the new energy equipment price prediction method based on incremental knowledge graph as described above.

[0018] On the other hand, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the new energy equipment price prediction method based on incremental knowledge graph as described above.

[0019] This invention provides a method for predicting the price of new energy equipment based on an incremental knowledge graph. It constructs a four-dimensional dynamic knowledge graph that integrates multi-source data for predicting new energy equipment prices. A quantum optimization algorithm is used to determine the nodes to be updated in the four-dimensional dynamic knowledge graph, and spatiotemporal features are extracted using temporal modeling to update the graph. An update strategy is configured hierarchically based on the frequency of event changes, and a key event influence model is constructed. This model, combined with a long short-term memory network model, predicts the occurrence patterns of events to drive the update of the four-dimensional dynamic knowledge graph. Finally, the features of the four-dimensional dynamic knowledge graph are input into a machine learning prediction model to predict the price of new energy equipment. This method solves the problems of low update efficiency and poor timeliness of traditional prediction models, improving the accuracy and efficiency of price prediction. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the new energy equipment price prediction method based on incremental knowledge graph provided in an embodiment of the present invention. Figure 2This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Figure 1 This is a flowchart illustrating the new energy equipment price prediction method based on incremental knowledge graph provided in an embodiment of the present invention.

[0024] like Figure 1 As shown, the new energy equipment price prediction method based on incremental knowledge graph provided in this embodiment of the invention mainly includes the following steps: 101. Collect multi-source data and analyze the multi-source data to form a four-dimensional dynamic knowledge graph.

[0025] This involves collecting multi-source data, including but not limited to information on new energy equipment, market dynamics, policies and regulations, and technological advancements. This multi-source data is then deeply analyzed and processed to construct a four-dimensional dynamic knowledge graph encompassing spatial, temporal, semantic, and popularity dimensions. This four-dimensional dynamic graph can intuitively display the knowledge structure and changing trends in the new energy equipment field, providing a data foundation for price prediction.

[0026] Specifically, it involves collecting data from multiple sources and parsing this data to form a four-dimensional dynamic knowledge graph, including: 1011. Collect supply chain data, market dynamic data, policy text data, industry news data, and market sentiment data for new energy equipment, and integrate them into multi-source data.

[0027] 1012. Extract the supplier name, supplier geographical location, neighbor node propagation time, and cost elasticity coefficient from the multi-source data to form spatial dimension features.

[0028] This process involves filtering supplier-related information from multi-source data, including supplier names and geographical locations. Supplier names identify different supply entities, while geographical locations reflect the spatial distribution of suppliers. Geographic coordinates of suppliers can be further obtained through Geographic Information Systems (GIS) or map APIs.

[0029] Secondly, neighborhood node propagation time refers to the speed at which information or influence spreads within the supplier network. This neighborhood node propagation time is extracted by analyzing transaction records, cooperation history, and other information in the supply chain data. By calculating the time required for information to spread from one supplier to another, the connectivity and information flow efficiency within the supplier network can be assessed.

[0030] Finally, the cost elasticity coefficient reflects the degree to which supplier costs fluctuate with changes in market conditions. The cost elasticity coefficient is derived through statistical analysis of cost data within the supply chain. Suppliers with higher cost elasticity coefficients are more susceptible to market fluctuations and face greater price risk.

[0031] Integrating the above features forms spatial dimension features.

[0032] For example, in the spatial dimension, each supplier can be abstracted into different nodes, and each node includes the supplier name. Supplier geographical location Supplier risk coefficient Three metadata items. Additionally, each node includes the propagation time of neighboring nodes. and cost elasticity coefficient Two attributes; for the first Each node has the following spatial dimension characteristics: ; Among them, supplier risk coefficient This will be affected by the supplier's supply volume; the third option can be used. Calculate the supplier risk coefficient based on the proportion of each supplier's supply volume. Furthermore, since suppliers of similar products are concentrated in a specific area, customers are more likely to compare prices and make purchases in concentrated supply zones. Therefore, the supplier risk coefficient can be calculated by measuring the average distance between the supplier and all other suppliers. Among them, supplier risk coefficient The calculation formula is: ; in, and Here are the normalization coefficients, and , and Determined by historical data; Indicates the impact of logistics distance; Supplier supply volume percentage; This represents the average geographic location of all suppliers. Indicates the first The geographical location of each supplier.

[0033] 1013. Extract the node addition time, node update time, and current time from the multi-source data to form time-dimensional features.

[0034] In terms of time dimension, each node is described by adding time information to the node. Node update time and current time To represent the temporal state of a node, the time dimension feature is: ; For example, taking the wind turbine blade supplier node of new energy wind power equipment as an example, when a certain wind turbine blade manufacturer A is first entered into the knowledge graph, the initial entry time is recorded, such as January 15, 2024, 09:30:00. This time is the node addition time. , is used to mark the starting point for node creation.

[0035] When manufacturer A's supply volume, geographical location, or other attributes change, such as when production capacity data is adjusted on May 20, 2024 at 14:10:00, the time of this attribute change is recorded as the node update time. This reflects the most recent dynamic adjustment of the node.

[0036] When performing incremental updates or price prediction calculations on the knowledge graph, record the current system processing time, such as 16:45:00 on July 10, 2024, as the current time. It is used to calculate the time decay effect or update interval of a node.

[0037] 1014. Extract unstructured data from multi-source data to form semantic dimension features.

[0038] Among them, policy text keywords, industry news sentiment analysis, and technical terms are unstructured data. Natural language models can be used to structure the multi-source data and obtain the corresponding embedded representations. Furthermore, the impact of the semantic dimension is quantified, whereby the impact of the semantic dimension is as follows: ; in, Used to compute embedding representation With the Nodes The distance between them The date since the event was announced.

[0039] After understanding the influence of the semantic dimension, semantic dimension features can be constructed, where the semantic dimension features are: .

[0040] 1015. Extract social media discussion popularity, industry forum attention, and investor sentiment index from multi-source data to form popularity dimension features.

[0041] To quantify the impact of an event's popularity, a sentiment index can be constructed. The sentiment index is as follows: ; in, For positive word frequency, For negative word frequency, and , respectively, are the weighting factors for positive and negative word frequencies, and , are modifiable normalization hyperparameters. and The specific values ​​can be adjusted based on expert knowledge; positive word frequencies have a smaller impact than negative word frequencies. For example, taking... =0.4, =0.6.

[0042] In obtaining the sentiment index Then, the popularity dimension features can be constructed, where the popularity dimension features are: ; 1016. Construct a four-dimensional dynamic knowledge graph based on spatial dimension features, temporal dimension features, semantic dimension features, and popularity dimension features.

[0043] The expression for the four-dimensional dynamic knowledge graph is: ; in, Used to describe the spatial dimensional characteristics of nodes. Used to describe the temporal dimension characteristics of nodes. Used to describe the semantic dimension features of nodes. Used to describe the popularity dimension of a node.

[0044] In the four-dimensional dynamic knowledge graph, spatial, temporal, semantic, and popularity features are independent of each other, collectively describing the relevant information and trends in the prices of new energy equipment. Spatial features reflect the distribution and differences of equipment in different geographical locations; temporal features reveal the patterns of price changes over time; semantic features, through keyword and topic analysis, delve into textual information related to equipment prices; and popularity features, by quantifying social media discussion volume, industry forum attention, and investor sentiment index, add real-time and dynamic aspects to the four-dimensional dynamic knowledge graph.

[0045] When constructing the four-dimensional dynamic knowledge graph, an incremental approach is adopted, that is, the knowledge graph is continuously updated and improved as new data continues to flow in, which improves the timeliness and accuracy of the four-dimensional dynamic knowledge graph and reduces the construction cost and maintenance difficulty.

[0046] 102. Based on the quantum optimization algorithm, determine the set of nodes to be updated in the four-dimensional dynamic knowledge graph, and extract the spatiotemporal features of the four-dimensional dynamic knowledge graph through temporal modeling to update the four-dimensional dynamic knowledge graph.

[0047] To further improve the update efficiency of the four-dimensional dynamic knowledge graph, quantum optimization algorithm and temporal modeling are used to update the four-dimensional dynamic knowledge graph.

[0048] Specifically, the set of nodes to be updated in the four-dimensional dynamic knowledge graph is determined based on the quantum optimization algorithm, and the spatiotemporal features of the four-dimensional dynamic knowledge graph are extracted through temporal modeling to update the four-dimensional dynamic knowledge graph, including: The update requirements of the four-dimensional dynamic knowledge graph are input into a quadratic unconstrained binary optimization model, solved using the quantum annealing method, and the output set of nodes to be updated is used to update the four-dimensional dynamic knowledge graph.

[0049] In the quadratic unconstrained binary optimization model, the update requirements of the four-dimensional dynamic knowledge graph are transformed into an optimization problem by setting the objective function and constraints. Quantum annealing, as an efficient quantum optimization algorithm, can perform a global search in the solution space to find the optimal or near-optimal solution. Solving using quantum annealing yields a set of nodes to be updated, representing the portion of the four-dimensional dynamic knowledge graph that needs to be updated.

[0050] During the update process, the corresponding nodes in the four-dimensional dynamic knowledge graph are first located based on the set of nodes to be updated. Then, the spatiotemporal features of these nodes are extracted, including geographical location, timestamps, semantic information, and popularity index. Next, temporal modeling methods are used to analyze and process these spatiotemporal features to extract the feature information required for the update. Finally, the updated feature information is integrated into the four-dimensional dynamic knowledge graph to complete the knowledge graph update.

[0051] Furthermore, the update requirements of the four-dimensional dynamic knowledge graph are input into a quadratic unconstrained binary optimization model, solved using quantum annealing, and the output set of nodes to be updated is used to update the four-dimensional dynamic knowledge graph, including: 1021. Transform the update requirements of the four-dimensional dynamic knowledge graph into a four-tuple structure that includes update nodes, relationships, timestamps, and impact dimensions.

[0052] 1022. Allocate qubits to each quadruple structure; qubits are used to characterize whether the corresponding quadruple needs to be updated.

[0053] 1023 Calculate the update value parameters and coupling relationship parameters of the quadruple structure.

[0054] 1024. Based on the updated value parameters, coupling relationship parameters, and qubits, construct a quadratic unconstrained binary optimization model.

[0055] 1025. Solve the quadratic unconstrained binary optimization model using quantum annealing to obtain the state corresponding to the qubit, and determine the set of nodes to be updated based on the state.

[0056] 1026. Update the nodes in the set of nodes to be updated and the node relationships corresponding to the nodes, so as to update the four-dimensional dynamic knowledge graph.

[0057] Specifically, the update requirements of the four-dimensional dynamic knowledge graph are structured and transformed into a four-tuple structure including update nodes, relationships, timestamps, and impact dimensions. By parsing the information in the update requirements, they are organized into four-tuples in a unified format for easier subsequent processing.

[0058] Next, a qubit is assigned to each quadruplet structure. The qubit is used to characterize whether the corresponding quadruplet needs to be updated. Specifically, a qubit state of 1 indicates that an update is needed, and a qubit state of 0 indicates that an update is not needed.

[0059] Then, the update value parameter and coupling relationship parameter are calculated for each quadruple structure. The update value parameter is used to evaluate the importance or priority of the quadruple update, while the coupling relationship parameter describes the potential mutual influence or dependency between quadruples. The calculation of the update value parameter and coupling relationship parameter is based on the current state of the four-dimensional dynamic knowledge graph and the specific content of the update requirement.

[0060] Based on the calculated update value parameters, coupling parameters, and qubits, a quadratic unconstrained binary optimization model is constructed. The objective of the quadratic unconstrained binary optimization model is to maximize or minimize an objective function under certain constraints. The objective function is related to the update value and coupling relationship of the quadruples.

[0061] The minimization problem of the quadratic unconstrained binary optimization model is as follows: ; in, This is an index for a quadruple structure. Indicates the first Does the quadruplet structure require updated qubits? Used to describe the The updating value of a quadruple structure, among which... The calculation formula is: ; in, It is a factor adjustment term. It is the time decay rate. Representing event A on the graph of the first... The degree of influence of each node It is the length of the interval between two events of this node.

[0062] Used to describe the The and the first The coupling relationship between the four tuples, where... The calculation formula is: ; in, It is an adjustable factor term. This indicates the degree of similarity between the word vectors of two relations, A and B.

[0063] Finally, based on the results obtained by quantum annealing, the selected quadruplets are updated, including updating node information and adjusting relationships, to ensure the accuracy and timeliness of the four-dimensional dynamic knowledge graph.

[0064] Among them, for one who has The knowledge graph of nodes, based on a quadratic unconstrained binary optimization model, uses quantum annealing to transform each variable... Mapping this to a qubit, the optimal update method can be solved by sliding from a high-energy state to a low-energy state. , represents the set of nodes that need to be updated, where Indicates the first Each node needs to be updated.

[0065] Furthermore, the four-dimensional dynamic knowledge graph is used to generate knowledge graph snapshots according to time series, and the snapshot spatiotemporal features of the knowledge graph snapshots are extracted using a temporal graph convolutional network to update the four-dimensional dynamic knowledge graph.

[0066] Temporal graph convolutional networks can capture the changes in knowledge graph snapshots at different points in time, as well as the relationships between these changes, thereby extracting snapshot spatiotemporal features. These snapshot spatiotemporal features reflect the evolution of nodes and relationships in the knowledge graph over time, providing an important basis for subsequent predictions. After extracting the snapshot spatiotemporal features, the four-dimensional dynamic knowledge graph is updated using these features to ensure that it accurately reflects the changing trends of new energy equipment prices, thus improving the accuracy and stability of new energy equipment price predictions.

[0067] Specifically, the four-dimensional dynamic knowledge graph is generated into snapshots according to time series, and a temporal graph convolutional network is used to extract the spatiotemporal features of the snapshots. Updating the four-dimensional dynamic knowledge graph includes: The four-dimensional dynamic knowledge graph is sliced ​​at preset time intervals to generate knowledge graph snapshots corresponding to multiple timestamps.

[0068] Set a time window and construct a time series subgraph that includes the changed nodes and their relationships within the time window.

[0069] Temporal graph convolutional networks are used to extract snapshot spatiotemporal features of temporal subgraphs.

[0070] Based on the spatiotemporal characteristics of snapshots, the update level of changed nodes is calculated.

[0071] Update the changed nodes according to the update level.

[0072] The preset time interval is determined based on the update frequency of new energy equipment price data to ensure that the generated knowledge graph snapshot can accurately reflect changes in equipment prices.

[0073] For example, when raw material prices fluctuate, the effects will gradually spread along the industrial chain from mining to refining to batteries to automobiles. There is a time lag effect between different nodes. Based on the price fluctuation frequency of the lithium battery industrial chain, a preset time interval of 1 hour is set, that is, the four-dimensional dynamic knowledge graph is sliced ​​once every hour to generate a knowledge graph snapshot.

[0074] A knowledge graph snapshot includes a timestamp. In this embodiment of the invention, the knowledge graph snapshot may also include an adjacency matrix and a feature matrix, thus the knowledge graph snapshot is: ; in, For timestamps; It is an adjacency matrix. express The relationships between time points; For the characteristic matrix, express Four-dimensional features of each node at any given time.

[0075] For example, when lithium ore prices rise by 15%, Generate snapshot ， in, Record the supply relationship between lithium ore suppliers and refineries at this time. Record the real-time risk coefficient, policy impact, and market sentiment index for each node.

[0076] The size of the time window is determined based on actual needs, aiming to include sufficient information to capture price change trends while avoiding excessive computational complexity due to its large size. When constructing the time series subgraph, only nodes that change within the time window and their relationships are considered to reduce computational load and improve update efficiency.

[0077] For example, the time window size can be set to 12 hours, including the current moment and a snapshot of the knowledge graph from the past 12 hours. At that time, the time window is The time-series subgraph includes lithium ore suppliers. to risk coefficient From 0.2 to 0.7, the refinery Update the purchase price.

[0078] The extraction of snapshot spatiotemporal features is achieved through a temporal graph convolutional network. This network captures changes in nodes and relationships at different points in time, as well as the correlations between these changes, thereby extracting useful features for price prediction. Based on the extracted snapshot spatiotemporal features, the update level of each changed node is calculated. This update level reflects the importance and urgency of the node's price change.

[0079] Finally, the change nodes are updated according to the update level to ensure that the four-dimensional dynamic knowledge graph can accurately reflect the changing trend of new energy equipment prices.

[0080] 103. Collect historical change logs of the four-dimensional dynamic knowledge graph, count the frequency of event changes within a preset time period, use the K-means clustering algorithm to classify the events, and configure the update strategy corresponding to the classification results.

[0081] For example, the frequency of event changes can be categorized into three levels: high-frequency, medium-frequency, and low-frequency. High-frequency events may represent drastic changes in market supply and demand, requiring immediate updates to relevant nodes and relationships in the knowledge graph. Medium-frequency events may involve policy adjustments or technological advancements, requiring updates within a certain timeframe. Low-frequency events may represent routine market fluctuations, which can be updated periodically. Different update frequencies, priorities, and methods are configured for different event levels to ensure the timeliness and accuracy of the four-dimensional dynamic knowledge graph.

[0082] Specifically, historical change logs of the four-dimensional dynamic knowledge graph are collected, the frequency of event changes within a preset time period is counted, the K-means clustering algorithm is used to classify the events, and the update strategy corresponding to the classification results is configured based on the classification results, including: 1031. Collect historical change logs of the four-dimensional dynamic knowledge graph. The historical change logs include records of changes to entity attributes, records of changes to relationships between entities, and corresponding timestamp information.

[0083] 1032. Using the first preset time window as the statistical period, count the number of changes for each event within the preset statistical period, and calculate the frequency of change of the event within the preset time period.

[0084] 1033. Using the frequency of change as a feature vector, input it into the K-means clustering algorithm for clustering and output the hierarchical results.

[0085] Specifically, if the classification result indicates a high-frequency event, a first update strategy is set for high-frequency events. If the classification result indicates a low-frequency event, a second update strategy is set for low-frequency events. If the classification result indicates a medium-frequency event, a third update strategy is set for medium-frequency events.

[0086] By periodically scanning and recording the four-dimensional dynamic knowledge graph, it automatically captures and records changes in the attributes of each entity, increases or decreases in relationships between entities, and the exact time when changes occur, forming a detailed historical change log.

[0087] Then, based on the first preset time window, the number of changes for each event within the period is counted, and the frequency of change of the event per unit of time, such as per hour or per day, is calculated accordingly. The frequency of change reflects the activity level of the event.

[0088] The frequency of change is used as a feature vector and input into a pre-trained K-means clustering algorithm. The K-means clustering algorithm groups events into different groups based on the similarity of the feature vectors, with each group representing a hierarchical result. The number of hierarchical results and the specific criteria for division can be adjusted according to actual needs.

[0089] For the first update strategy, since high-frequency events represent urgent situations such as drastic changes in market supply and demand, the relevant nodes and relationships in the four-dimensional dynamic knowledge graph need to be updated immediately to ensure the timeliness and accuracy of the information. For the second update strategy, since low-frequency events change more gradually, a longer update cycle can be set, such as updating weekly or monthly. For the third update strategy, a balance needs to be found between high-frequency and low-frequency events, such as updating every two or three days.

[0090] 104. Construct a critical event impact assessment model and calculate the event criticality value by combining the node association relationship of the four-dimensional dynamic knowledge graph.

[0091] The critical event impact assessment model comprehensively considers factors such as the nature of the event, its frequency of occurrence, duration, and the strength of its correlation with other nodes to calculate the criticality value of each event. The criticality value reflects the importance and influence of the event within the four-dimensional dynamic knowledge graph. Events with higher criticality values ​​have a more significant impact on the price of new energy equipment, and therefore should be given more attention in price prediction. Furthermore, the critical event impact assessment model can be dynamically adjusted and optimized according to the needs of actual application scenarios to adapt to different market environments and business requirements.

[0092] Specifically, a key event impact assessment model is constructed, which calculates the event's criticality value by combining the node relationships of a four-dimensional dynamic knowledge graph, including: 1041. Extract the topological attributes of the nodes associated with the event and calculate the node degree value. The node degree value represents the degree of association between the nodes in the knowledge graph.

[0093] 1042. Calculate the longest propagation path length of the node. The longest propagation path length represents the propagation depth of the event's impact.

[0094] 1043. Call the semantic dimension features and popularity dimension features of the four-dimensional dynamic knowledge graph to construct the keyness value calculation formula and output the keyness value of the event.

[0095] Extracting the topological attributes of nodes associated with an event involves traversing a four-dimensional dynamic knowledge graph to find all nodes directly or indirectly related to the event and collecting their topological attribute information. These topological attributes include the number of connections and connection types, used to calculate the node's degree value. The node's degree value characterizes the strength of its associations within the knowledge graph; a higher degree value indicates a stronger association with other nodes and a potentially greater influence on the event.

[0096] Next, the longest propagation path length of a node can be calculated by analyzing the connections between nodes in the four-dimensional dynamic knowledge graph to find the longest path from the event node to other nodes. The longest propagation path length reflects the depth of the event's influence, that is, the distance to the farthest node that the event can affect. The longer the path length, the wider the scope of the event's influence and the deeper its propagation.

[0097] Then, the semantic and popularity dimensions of the four-dimensional dynamic knowledge graph are invoked. Semantic features describe the semantic information and contextual relationships of nodes, used to understand the specific meaning and background of events. Popularity features reflect the activity and attention a node receives within the current time window. Combining semantic and popularity features, a formula for calculating the criticality value can be constructed. This formula considers the node degree, the longest propagation path length, and the weights and interactions of semantic and popularity features, calculating the criticality value of the event through weighted summation or non-linear combination.

[0098] Among them, the criticality value To describe what happens at the node The criticality of event A enables proactive identification of critical events; the criticality value is used to determine its importance. for: ; in, Describes the degree of a node, used to reflect node centrality at the structural level; The longest edge describing a node, used to represent the deepest propagation depth of an event; This reflects the intensity of the event's content by assessing its impact on both semantic and popularity levels.

[0099] Finally, the calculated criticality values ​​are sorted or categorized for subsequent analysis and application. Events with higher criticality values ​​should receive more attention and analysis in the process of predicting new energy equipment prices.

[0100] 105. Input the criticality value into the Long Short-Term Memory network model, output the probability and time of event occurrence, integrate the update strategy, and control the update of the four-dimensional dynamic knowledge graph.

[0101] Among them, the Long Short-Term Memory (LSTM) network model, by learning criticality values, can capture the temporal sequence features of events, thereby predicting the probability and timing of events. The fusion update strategy dynamically adjusts the nodes and edges in the four-dimensional dynamic knowledge graph based on the prediction results and the actual occurrence, ensuring the timeliness and accuracy of the graph and realizing the transformation of the four-dimensional dynamic knowledge graph from passive updating to predictive updating.

[0102] Specifically, the keyness value is input into the Long Short-Term Memory network model, which outputs the probability and timing of event occurrence. An update strategy is then used to control the update of the four-dimensional dynamic knowledge graph, including: 1051. Collect the event criticality values ​​within the second preset time window to form a time sequence; 1052. Input the time series into the Long Short-Term Memory Network Model and output the predicted value of the event's criticality within the first preset time period. The predicted value includes the probability of the event occurring and the expected time of occurrence. 1053. If the probability of an event occurring exceeds a preset threshold, a priority update mechanism is triggered. The update priority and timing are determined by combining the update strategy corresponding to the graded results. 1054. Based on the expected occurrence time, start the incremental update preparation of related nodes in advance, and complete the four-dimensional dynamic knowledge graph update when the event occurs.

[0103] The event criticality values ​​collected within the second preset time window reflect the importance of the event across different dimensions. Arranging these criticality values ​​chronologically to form a time series ensures that the data input into the Long Short-Term Memory (LSTM) network model is ordered and continuous, helping the LTM network model capture the temporal sequence characteristics of events.

[0104] Next, the time series is input into the Long Short-Term Memory (LSTM) network model. The LTM network model is a type of recurrent neural network that learns from historical information to predict future events. During the training process, historical event data is used to train the LTM network model, enabling it to accurately predict the criticality, probability of occurrence, and expected timing of future events. When a new time series is input into the LTM network model, it outputs a predicted criticality value for the event within a first preset time period, based on the learned knowledge. This predicted criticality value includes the probability of occurrence and the expected timing of the event.

[0105] Then, the predicted event probability is used to determine whether a priority update mechanism needs to be triggered. If the predicted event probability exceeds a preset threshold, it indicates that the event is of high importance and needs to be prioritized. At this point, the update priority and timing are determined by combining the update strategy corresponding to the grading results. The grading results are based on the criticality value of the event, and different levels of events correspond to different update strategies and resource allocations.

[0106] Finally, based on the predicted event occurrence time, incremental updates of related nodes are initiated in advance. This step includes loading event-related nodes and edges, calculating node degrees, longest edges, and semantic and popularity impacts. When the event actually occurs, the four-dimensional dynamic knowledge graph is updated immediately to ensure that it reflects the latest state of the event in real time.

[0107] For example, suppose a series of event severity values ​​related to new energy policy adjustments are collected within a second preset time window. The time-series sequence formed by these severity values ​​is input into a Long Short-Term Memory (LSTM) network model. This model predicts that a certain new energy policy adjustment event has a high probability of occurring within the next first preset time period and provides the expected occurrence time. At this point, a priority update mechanism is triggered, determining the update priority and timing based on the event's classification results and corresponding update strategies. Then, based on the predicted occurrence time, incremental updates of nodes and edges related to the event are initiated in advance. When the policy adjustment event actually occurs, the four-dimensional dynamic knowledge graph is immediately updated to ensure that the graph contains the latest policy information.

[0108] 106. Extract the node features and node relationships of the four-dimensional dynamic knowledge graph, input them into the machine learning prediction model, and output the price prediction results of new energy equipment.

[0109] The machine learning prediction model, upon receiving node features and their relationships, processes the data using a pre-trained algorithm. First, it analyzes the node features to identify key factors influencing the price of new energy equipment, such as equipment type, production year, and technology level. Then, it combines these node relationships to assess the interactions between these factors and their potential impact on price. Through a complex calculation process, the machine learning prediction model ultimately outputs a price prediction result for new energy equipment. This prediction result includes not only the price value but also information such as price trends and prediction confidence levels.

[0110] Specifically, the node features and relationships of the four-dimensional dynamic knowledge graph are extracted, input into a machine learning prediction model, and the output is the price prediction result for new energy equipment, including: 1061. Extract the four-dimensional feature vectors of each node in the four-dimensional dynamic knowledge graph.

[0111] 1062. Use graph convolutional networks to aggregate the neighborhood relationships of nodes and generate node embedding vectors with fused topology structures.

[0112] 1063. Concatenate the node embedding vector with the time series features to form a prediction feature matrix.

[0113] 1064. Input the predicted feature matrix into the machine learning prediction model and output the predicted price of new energy equipment within the second preset time period.

[0114] 1065. Based on the feedback of the deviation between the predicted value and the actual price, dynamically adjust the parameters of the machine learning prediction model and the update frequency of the four-dimensional dynamic knowledge graph.

[0115] Specifically, historical data from a four-dimensional dynamic knowledge graph is used to extract a four-dimensional feature vector for each node. This four-dimensional feature vector includes the node's historical average price, price volatility, technical grade score, and market share. The historical average price reflects the node's long-term price level, price volatility reveals the stability of the node's price, the technical grade score reflects the node's technological advancement, and market share shows the node's competitive position in the market. By integrating these four dimensions, the price characteristics of a node can be grasped.

[0116] After extracting node feature vectors, a graph convolutional network is used to aggregate the neighborhood relationships of the nodes. The graph convolutional network iteratively updates the node representation, incorporating information from neighboring nodes into the current node, thereby generating node embedding vectors that fuse the topological structure. This process considers not only the features of the nodes themselves but also the relationships between nodes, making the generated node embedding vectors more representative.

[0117] To capture price trends over time, node embedding vectors are concatenated with time-series features. These features include price change rates and market demand changes over a past period. This concatenation creates a predictive feature matrix that includes both node and time-series features.

[0118] The predicted feature matrix is ​​input into a pre-trained machine learning prediction model. Based on the input feature matrix, the machine learning prediction model uses algorithms to predict the price of new energy equipment. The prediction results include price forecasts for a second preset time period, as well as information on price trends and prediction confidence levels. This information provides strong support for decision-makers.

[0119] To continuously improve forecast accuracy, adjustments are made based on the deviation between predicted values ​​and actual prices. On one hand, the parameters of the machine learning forecasting model are adjusted according to the deviation to optimize its predictive performance; on the other hand, the update frequency of the four-dimensional dynamic knowledge graph is adjusted based on the deviation to ensure that the knowledge graph can reflect the latest market changes in a timely manner. Through this dynamic adjustment process, the accuracy and reliability of forecasts can be continuously improved.

[0120] Figure 2 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0121] like Figure 2As shown, the electronic device may include a processor 210, a communications interface 220, a memory 230, and a communication bus 240. The processor 210, communications interface 220, and memory 230 communicate with each other via the communication bus 240. The processor 210 can call logical instructions from the memory 230 to execute a new energy equipment price prediction method based on incremental knowledge graphs.

[0122] Furthermore, the logical instructions in the aforementioned memory 230 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the new energy equipment price prediction method based on incremental knowledge graph provided by the above methods.

[0124] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the new energy equipment price prediction method based on incremental knowledge graph provided by the above methods.

[0125] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0126] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the price of new energy equipment based on incremental knowledge graphs, characterized in that, include: Collect multi-source data and parse the multi-source data to form a four-dimensional dynamic knowledge graph; The set of nodes to be updated in the four-dimensional dynamic knowledge graph is determined based on the quantum optimization algorithm, and the spatiotemporal features of the four-dimensional dynamic knowledge graph are extracted through temporal modeling to update the four-dimensional dynamic knowledge graph. Collect the historical change logs of the four-dimensional dynamic knowledge graph, count the frequency of event changes within a preset time period, use the K-means clustering algorithm to classify the events, and configure the update strategy corresponding to the classification results based on the classification results. Construct a key event impact assessment model and calculate the event criticality value by combining the node association relationship of the four-dimensional dynamic knowledge graph; Input the criticality value into the long short-term memory network model, output the probability and time of the event occurrence, integrate the update strategy, and control the update of the four-dimensional dynamic knowledge graph. Extract the node features and their associations from the four-dimensional dynamic knowledge graph, input them into a machine learning prediction model, and output the price prediction results for new energy equipment.

2. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 1, characterized in that, The process of determining the set of nodes to be updated in the four-dimensional dynamic knowledge graph based on a quantum optimization algorithm, and using temporal modeling to extract the spatiotemporal features of the four-dimensional dynamic knowledge graph to update it, includes: The update requirements of the four-dimensional dynamic knowledge graph are input into a quadratic unconstrained binary optimization model, solved using quantum annealing, and the set of nodes to be updated is output to update the four-dimensional dynamic knowledge graph. The four-dimensional dynamic knowledge graph is used to generate knowledge graph snapshots according to time series, and the snapshot spatiotemporal features of the knowledge graph snapshots are extracted using a temporal graph convolutional network to update the four-dimensional dynamic knowledge graph.

3. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 2, characterized in that, The process of inputting the update requirements of the four-dimensional dynamic knowledge graph into a quadratic unconstrained binary optimization model, solving it using quantum annealing, outputting the set of nodes to be updated, and updating the four-dimensional dynamic knowledge graph includes: The update requirements of the four-dimensional dynamic knowledge graph are transformed into a four-tuple structure including update nodes, relationships, timestamps, and influence dimensions. Each of the quadruplet structures is assigned a qubit; the qubit is used to characterize whether the corresponding quadruplet needs to be updated. Calculate the update value parameter and coupling relationship parameter of the quadruple structure; Based on the updated value parameter, the coupling relationship parameter, and the qubit, a quadratic unconstrained binary optimization model is constructed. The quantum annealing method is used to solve the quadratic unconstrained binary optimization model to obtain the state corresponding to the qubit, and the set of nodes to be updated is determined based on the state. The nodes in the set of nodes to be updated and the node associations corresponding to the nodes are updated in order to update the four-dimensional dynamic knowledge graph.

4. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 2, characterized in that, The four-dimensional dynamic knowledge graph is used to generate knowledge graph snapshots according to time series, and a temporal graph convolutional network is used to extract the snapshot spatiotemporal features of the knowledge graph snapshots to update the four-dimensional dynamic knowledge graph, including: The four-dimensional dynamic knowledge graph is sliced ​​at preset time intervals to generate knowledge graph snapshots corresponding to multiple timestamps. Set a time window and construct a time-series subgraph that includes the changed nodes and their relationships within the time window; The temporal subgraph is extracted using a temporal graph convolutional network to extract snapshot spatiotemporal features. Based on the spatiotemporal characteristics of the snapshot, the update level of the changed node is calculated; Update the change node according to the update level.

5. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 4, characterized in that, The step of using a temporal graph convolutional network to extract snapshot spatiotemporal features of the temporal subgraph includes: Using graph convolutional networks to aggregate the neighborhood information of nodes, capture the spatial relationship structure between upstream and downstream of the industry chain, and generate spatial features; Gated recurrent units are used to process the temporal changes of node features, capture the time lag effect and transmission path of price fluctuations, and generate time features. By fusing the spatial features and the temporal features, a snapshot spatiotemporal feature containing spatiotemporal dependencies is generated.

6. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 1, characterized in that, The process involves collecting historical change logs of the four-dimensional dynamic knowledge graph, statistically analyzing the frequency of event changes within a preset time period, using the K-means clustering algorithm to classify the events, and configuring an update strategy corresponding to the classification results, including: Collect the historical change log of the four-dimensional dynamic knowledge graph. The historical change log includes entity attribute change records, entity relationship change records and corresponding timestamp information. Using a first preset time window as the statistical period, the number of changes of each event within the preset statistical period is counted, and the frequency of change within the preset time period is calculated. The frequency of change is used as a feature vector, and the K-means clustering algorithm is used to perform clustering, and the hierarchical results are output. If the classification result is a high-frequency event, then a first update strategy is set for the high-frequency event; If the classification result is a low-frequency event, then a second update strategy is set for the low-frequency event; If the classification result is a mid-frequency event, then a third update strategy is set for the mid-frequency event.

7. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 1, characterized in that, The construction of the key event impact assessment model, which combines the node relationships of the four-dimensional dynamic knowledge graph to calculate the event's criticality value, includes: Extract the topological attributes of event-related nodes and calculate the node degree value, which represents the degree of association between nodes in the knowledge graph; Calculate the longest propagation path length of the node, where the longest propagation path length characterizes the propagation depth of the event's impact; The semantic and popularity features of the four-dimensional dynamic knowledge graph are invoked to construct a formula for calculating the criticality value and output the criticality value of the event.

8. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 1, characterized in that, The process of inputting the criticality value into a long short-term memory network model, outputting the event occurrence probability and occurrence time, and incorporating the update strategy to control the update of the four-dimensional dynamic knowledge graph includes: Collect the event criticality values ​​within the second preset time window to form a time sequence; The time series is input into a long short-term memory network model, and the predicted value of the event criticality within a first preset time period is output. The predicted value includes the probability of the event occurring and the expected time of occurrence. If the probability of the event occurring exceeds a preset threshold, a priority update mechanism is triggered, and the update priority and timing are determined by combining the update strategy corresponding to the grading result. Based on the expected occurrence time, incremental update preparations for associated nodes are initiated in advance, and the four-dimensional dynamic knowledge graph update is completed when the event occurs.

9. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 1, characterized in that, Extract the node features and relationships of the four-dimensional dynamic knowledge graph, input them into a machine learning prediction model, and output the price prediction results of new energy equipment, including: Extract the four-dimensional feature vector of each node in the four-dimensional dynamic knowledge graph; The graph convolutional network is used to aggregate the neighborhood relationships of nodes and generate node embedding vectors with fused topology. The node embedding vector is concatenated with the time series features to form a prediction feature matrix; The predicted feature matrix is ​​input into a machine learning prediction model, which outputs the predicted price of new energy equipment within a second preset time period. Based on the feedback of the deviation between the predicted value and the actual price, the parameters of the machine learning prediction model and the update frequency of the four-dimensional dynamic knowledge graph are dynamically adjusted.

10. The method for predicting the price of new energy equipment based on incremental knowledge graphs according to claim 1, characterized in that, The process of collecting multi-source data and parsing the multi-source data to form a four-dimensional dynamic knowledge graph includes: Collect supply chain data, market dynamics data, policy text data, industry news data, and market sentiment data for new energy equipment, and integrate them into multi-source data; Extract supplier names, supplier geographical locations, neighboring node propagation time, and cost elasticity coefficients from the multi-source data to form spatial dimension features; Extract the node addition time, node update time, and current time from the multi-source data to form time-dimensional features; Extract unstructured data from the multi-source data to form semantic dimension features; Extract social media discussion popularity, industry forum attention, and investor sentiment index from the multi-source data to form popularity dimension features; A four-dimensional dynamic knowledge graph is constructed based on the spatial dimension features, the temporal dimension features, the semantic dimension features, and the popularity dimension features.

Images