Comparison event network-driven action analysis and prediction method and system

Abstract

The invention discloses a comparison event network-driven action analysis prediction method and system, and relates to the field related to action prediction, and the method comprises the steps: carrying out the event collection based on a plurality of data sources according to a time sequence, obtaining a multi-dimensional event data set, constructing a time sequence knowledge graph, carrying out the capturing, obtaining a plurality of time sequence characteristics, and obtaining a plurality of time sequence characteristics; calling a historical event data set based on the multi-dimensional event data set, performing comparative analysis on the historical event data set according to a plurality of time sequence characteristics through a comparative event network model, and generating a comparative event relation mode to perform time sequence modeling analysis, and obtaining a potential evolution analysis result, performing data fusion on the multi-dimensional event data set, performing action prediction on the target object, obtaining an action trend prediction result, and making an action decision. The technical problem that the future action trend and probability of the target object cannot be predicted due to the fact that time sequence dependency and evolution relation between events are difficult to capture is solved, and the technical effects of improving prediction precision and providing intelligent support for action decision making are achieved.

Description

Technical Field

[0001] The present application relates to the field of action prediction, and in particular to a method and system for action analysis and prediction driven by a comparative event network. Background Art

[0002] With the continuous improvement of automation and intelligence levels, action prediction, optimization and behavior analysis have been widely used in the military field. Traditionally, action monitoring mainly relies on manual inspection and predetermined monitoring plans. Although this method can solve some problems, it often fails to detect potential behavior trends in a timely manner and is difficult to capture the temporal dependencies and evolution relationships between events, resulting in the technical problem of being unable to predict the future action trends and probabilities of the target object. Summary of the invention

[0003] This application provides a comparative event network-driven action analysis and prediction method and system to solve the technical problem that it is difficult to capture the temporal dependencies and evolution relationships between events, resulting in the inability to predict the future action trends and probabilities of the target objects, thereby achieving the technical effect of improving prediction accuracy to provide intelligent support for action decisions.

[0004] The present application provides a comparative event network-driven action analysis and prediction method, which is applied to a comparative event network-driven action analysis and prediction system, including: collecting events according to time series based on multiple data sources to obtain a multidimensional event data set, and constructing a time series knowledge graph based on the multidimensional event data set; capturing according to the time series knowledge graph to obtain multiple time series characteristics, and retrieving a historical event data set based on the multidimensional event data set; constructing a comparative event network model, and comparatively analyzing the historical event data set according to the multiple time series characteristics through the comparative event network model to generate a comparative event relationship pattern; performing time series modeling analysis based on the comparative event relationship pattern to obtain a potential evolution analysis result, and the potential evolution analysis result includes multidimensional relationship data; performing data fusion on the multidimensional event data set based on the multidimensional relationship data, performing action prediction on the target object according to the fusion result to obtain an action trend prediction result, and making an action decision based on the action trend prediction result.

[0005] In a possible implementation, the time series knowledge graph is constructed based on the multidimensional event data set, and the following processing is performed: the multidimensional event data set is aligned according to the timestamps of the multiple data sources to obtain a multidimensional event aligned data set; the multidimensional event aligned data set is clustered according to the time series to obtain a plurality of nodes; the relationship between the multiple nodes is modeled using graph theory to construct a plurality of relationship edges; the multiple nodes are connected through the multiple relationship edges to obtain the time series knowledge graph.

[0006] In a possible implementation, the historical event data set is compared and analyzed according to multiple time series characteristics through the comparative event network model to generate a comparative event relationship pattern, and the following processing is performed: the temporal knowledge graph is traversed to extract quadruple data, and the quadruple data includes head entity data, relationship data, tail entity data, and timestamp; the historical event data set is synchronized to the comparative event network model and analyzed with the quadruple according to multiple time series characteristics to construct a relationship function; the historical relationship pattern is calculated through the relationship function, and the non-historical relationship pattern is determined based on the historical relationship pattern; the historical relationship pattern is integrated with the non-historical relationship pattern to obtain the comparative event relationship pattern.

[0007] In a possible implementation, a time series modeling analysis is performed based on the comparative event relationship pattern to obtain a potential evolution analysis result, and the following processing is performed: a first relationship distribution function is constructed, and the historical relationship frequency is obtained by calculation through the first relationship distribution function; the first relationship distribution function is converted to obtain a second relationship distribution function, and the non-historical relationship frequency is obtained by calculation through the second relationship distribution function; a time series modeling analysis is performed based on the historical relationship frequency combined with real-time scene parameters to obtain a first evolution path; a time series modeling analysis is performed based on the non-historical relationship frequency combined with real-time scene parameters to obtain a second evolution path; an event relationship analysis is performed based on the first evolution path, the second evolution path and the comparative event relationship pattern to obtain multi-dimensional relationship data, and the multi-dimensional relationship data is added to the potential evolution analysis result.

[0008] In a possible implementation, the multidimensional event data set is subjected to data fusion based on the multidimensional relationship data, and the following processing is performed: a first correlation function is constructed, and a correlation analysis is performed on the multidimensional event data set according to the multidimensional relationship data through the first correlation function to obtain historical relationship correlation parameters; a second correlation function is constructed, and a correlation analysis is performed on the multidimensional event data set according to the multidimensional relationship data through the second correlation function to obtain non-historical relationship correlation parameters; the multidimensional event data set is scored based on the historical relationship correlation parameters to obtain a first score; the multidimensional event data set is scored based on the non-historical relationship correlation parameters to obtain a second score; the multidimensional event data set is subjected to data fusion according to the first score and the second score to generate the fusion result.

[0009] In a possible implementation, the action of the target object is predicted based on the fusion result to obtain the action trend prediction result, and the following processing is performed: capture the target object, perform supervised learning on the target object according to the fusion result, and obtain multiple prediction data; construct a probability function, calculate the multiple prediction data through the probability function, and obtain the prediction probability; according to the prediction probability and the combination of the multiple prediction data, perform dynamic analysis according to the multi-dimensional relationship data to construct the action trend prediction result.

[0010] The present application also provides an action analysis and prediction system driven by a comparative event network, including: a data acquisition module, which is used to collect events according to a time series based on multiple data sources to obtain a multidimensional event data set, and construct a time series knowledge graph based on the multidimensional event data set; a data capture module, which is used to capture according to the time series knowledge graph, obtain multiple time series characteristics, and retrieve a historical event data set based on the multidimensional event data set; a comparative analysis module, which is used to construct a comparative event network model, and compare and analyze the historical event data set according to the multiple time series characteristics through the comparative event network model to generate a comparative event relationship pattern; a modeling and analysis module, which is used to perform time series modeling analysis based on the comparative event relationship pattern to obtain a potential evolution analysis result, and the potential evolution analysis result includes multidimensional relationship data; an action prediction module, which is used to perform data fusion on the multidimensional event data set based on the multidimensional relationship data, perform action prediction on the target object according to the fusion result, obtain an action trend prediction result, and make an action decision based on the action trend prediction result.

[0011] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0012] The comparative event network-driven action analysis and prediction method and system provided in the present application relate to the field of action prediction technology, and solve the technical problem that it is difficult to capture the temporal dependencies and evolution relationships between events, resulting in the inability to predict the future action trends and probabilities of the target objects, so as to achieve the technical effect of improving the prediction accuracy and providing intelligent support for action decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical scheme of the embodiment of the present application, the accompanying drawings of the embodiment of the present application 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 a certain step or several steps of operations can be removed from these processes.

[0014] Figure 1A flowchart of a comparative event network-driven action analysis and prediction method provided in an embodiment of the present application;

[0015] Figure 2 A schematic diagram of the structure of a comparative event network-driven action analysis and prediction system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] 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.

[0017] 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.

[0018] 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 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 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.

[0019] The present application embodiment provides a method for analyzing and predicting actions driven by a comparative event network, and the method is applied to an action analysis and prediction system driven by a comparative event network, such as Figure 1 As shown, the method includes:

[0020] Step A100, collect events according to time series based on multiple data sources to obtain a multidimensional event data set, and construct a time series knowledge graph based on the multidimensional event data set; in a possible implementation, step A100 further includes step A110, aligning the multidimensional event data set according to the timestamps of the multiple data sources to obtain a multidimensional event alignment data set; executing step A120, clustering the multidimensional event alignment data set according to the time series to obtain multiple nodes; executing step A130, using graph theory to model the relationships between the multiple nodes and construct multiple relationship edges; executing step A140, connecting the multiple nodes through the multiple relationship edges to obtain the time series knowledge graph.

[0021] The multidimensional event data set is aligned according to the timestamps of the multiple data sources. Through time alignment processing, events from different data sources are placed on a unified timeline. Then, the multidimensional event aligned data set is clustered according to the time series. Each event is taken as a node, and the attributes of the node may include event type, timestamp, related data, etc. Cluster analysis is to cluster event type, timestamp, related data and other data, and finally each clustered data set is recorded as a node, and multiple nodes are obtained to represent different event groups in the data set.

[0022] Graph theory is used to model the relationships between nodes obtained through clustering. Graph theory is used to describe the relationships (i.e., edges) between nodes. In a time series knowledge graph, the relationship types may include time series relationships, causal relationships, and association relationships. The time series relationship refers to determining the order of events by timestamps. For example, if event A occurs before event B, then there is a time series relationship between A and B. The causal relationship refers to identifying the causal relationship between events by analyzing data or expert experience. For example, an increase in device temperature (event A) may lead to device failure (event B). The association relationship refers to association based on the attributes of the event. For example, the correlation analysis between temperature fluctuation (event A) and device failure (event B).

[0023] When using graph theory, the relationship between events can be represented by defining edge attributes (such as weight, direction, etc.) to determine multiple relationship edges, which can include directed edges and undirected edges. The directed edges are used to represent the causal relationship or temporal relationship between events, and the arrows point in the direction of the causal relationship. The undirected edges are used to represent the correlation relationship between events.

[0024] Finally, by connecting multiple nodes and the relationship edges between them, a complete time series knowledge graph is formed. The time series knowledge graph can not only describe the time series characteristics of a single event, but also show the complex causal relationship, correlation and time dependency between multiple events.

[0025] Execute step A200, capture according to the time series knowledge graph, obtain multiple time series characteristics, and retrieve the historical event data set based on the multidimensional event data set; the time series knowledge graph can capture the time series characteristics, causal relationships and correlations between events by graphically modeling the events (nodes) and the relationships (edges) between them in the event data set. First, all events are extracted from the multidimensional event data set to form nodes. Each node contains the attributes of the event, such as event type, timestamp, state information, etc., which can be modeled according to the relationship between the events. The relationship between the events can include time series relationships, causal relationships, and association relationships. Further, capture according to the time series knowledge graph is to extract the time series dependency and evolution law of events from the time series knowledge graph to identify potential patterns and obtain multiple time series characteristics. The multiple time series characteristics include, time series dependency, causal relationships, hysteresis effects, periodic changes, and trend changes. Time series dependency refers to the time order relationship between events, which is usually represented by directed edges in the graph. That is, after event A occurs, event B is more likely to occur. Causation means that the occurrence of certain events is the cause of other events, for example, high temperature causes equipment failure. Hysteresis means that the impact of certain events may not appear immediately, but only after a period of time. This hysteresis effect is an important feature of time series data. Periodicity means that some events may have periodic characteristics, for example, the working status of equipment changes within a specific time period. Trend change means that some events may show an upward or downward trend over time, such as the gradual decline of equipment performance.

[0026] Furthermore, the historical event dataset is retrieved based on the multidimensional event dataset, and the historical events related to the current task are screened out from the multidimensional event dataset. The appropriate historical dataset is screened based on the event type, timestamp, and attributes, and a time window (such as the past 30 days, 90 days, etc.) is set to extract events that fit the time range from the historical event data. The choice of time window depends on the nature of the event and the requirements of the prediction model. For the retrieved historical event data, key event features (such as event type, occurrence time, duration, etc.) and attributes related to the target task are extracted to obtain the historical event dataset, which contains all event records that have occurred in the past. These historical events can provide valuable information for predictions and decisions in the current situation, and improve prediction accuracy and decision-making effectiveness.

[0027] Execute step A300 to construct a comparison event network model, and perform comparative analysis on the historical event dataset according to the multiple temporal characteristics through the comparison event network model to generate a comparison event relationship pattern; in a possible implementation, step A300 further includes step A310 of traversing the temporal knowledge graph to extract quadruple data, where the quadruple data includes head entity data, relationship data, tail entity data, and timestamp; in the temporal knowledge graph, the quadruple represents the relationship and temporal dependence between events. The composition of the quadruple can include head entity data, relationship data, tail entity data, and timestamp. The head entity data is used to represent the starting point of the event, which can be a device, user, sensor, etc. The relationship data is used to represent the relationship between the head entity and the tail entity, such as causal relationship, temporal relationship, correlation, etc. The tail entity data is used to represent the termination point or target entity of the event, such as device failure, temperature change, operation behavior, etc. The timestamp is used to represent the time when the event occurs, usually UTC time or system timestamp, to ensure the sequentiality of event data on the time axis. The extraction of quadruple data is to transform the information in the multi-dimensional event dataset into a structured representation form for subsequent graphical modeling and relationship analysis.

[0028] Execute step A320 to synchronize the historical event dataset to the comparison event network model and analyze the quadruples according to multiple temporal characteristics to construct a relationship function; based on the quadruples of the temporal knowledge graph, that is, (head entity, relationship, tail entity, timestamp), and use ξ, R, and T to represent finite sets of entities, relationships, and timestamps respectively. h ∈ ξ is the head entity, o ∈ ξ is the tail entity, r ∈ R is the relationship connecting the head entity and the tail entity, and G t represents a snapshot of the temporal knowledge graph of the quadruple occurring at time t. Define as the embedding vectors of h, r, and o, with dimension d. |ξ| × d represents the embedding vectors of all entities. Similarly, |R| × d represents the embedding vectors of all relationships.

[0029] Define the historical event set as Define the historical relationship set as The moment θ represents a certain moment before time t (θ < t), and the relationship function is:

[0030]

[0031] where, is the historical event dataset, the moment θ represents a certain moment before time t (θ < t), is the historical event relationship set, R is the entity, h ∈ ξ is the head entity, o ∈ ξ is the tail entity, r ∈ R is the relationship connecting the head entity and the tail entity, and G θRepresents a snapshot of the temporal knowledge graph of a quadruple occurring at time θ.

[0032] Execute step A330, obtain the historical relationship model through the relationship function calculation, and determine the non-historical relationship model based on the historical relationship model; execute step A340, integrate the historical relationship model with the non-historical relationship model to obtain the comparison event relationship model.

[0033] By comparing historical event data sets with current event data sets, we can obtain known event relationships based on historical data. These relationship patterns reveal the dependencies and regularities between events over a period of time.

[0034] Relationship is called a historical relationship, not a The relations in are called non-historical relations. Define M as a set of historical events. If the historical event A being queried does not exist in M, it is represented as a new event.

[0035] When processing events in a time-series knowledge graph, newly occurring events may not have historical data to refer to. To address this problem, the CENet model is designed with the frequency of both historical and non-historical relationships in mind. This means that when the system is dealing with new or rare events, it can not only rely on historical data for prediction or analysis, but can also consider those situations that are not supported by historical data. This enhances the flexibility and adaptability of the system, especially when dealing with rapidly changing environments.

[0036] Execute step A400, perform time series modeling analysis based on the comparison event relationship pattern, and obtain potential evolution analysis results, which contain multi-dimensional relationship data; in one possible implementation, step A400 further includes step A410, constructing a first relationship distribution function, and calculating through the first relationship distribution function to obtain the historical relationship frequency; in the environment of time series knowledge graph, although many events occur repeatedly, there may be a lack of reference basis for historical events relative to newly occurring events. Therefore, in the preprocessing stage, the historical relationship frequency of the event is first calculated. That is, the frequency of all relationships associated with the head entity and the tail entity before time t, and the first relationship distribution function is:

[0037]

[0038] in, is the historical relationship frequency, and time θ represents a moment before time t (θ <t), is a historical event relation set, h∈ξ is the head entity, o∈ξ is the tail entity, r∈R is the relationship between the head entity and the tail entity, G θ Represents a snapshot of the temporal knowledge graph of a quadruple occurring at time θ.

[0039] Execute step A420, convert the first relationship distribution function to obtain a second relationship distribution function, calculate the non-historical relationship frequency by using the second relationship distribution function; however, since the frequency of the non-historical relationship cannot be calculated, Convert to The values ​​for each slot are not unlimited. They are limited by the hyperparameter λ. λ is used to limit In order to ensure the stability and effectiveness of the model and avoid the adverse effects of excessively large values ​​on the model, the second relationship distribution function is:

[0040]

[0041] in, The transformation obtained, is the non-historical relationship frequency, λ is a hyperparameter and is an exponential function.

[0042] Φ Ω is an exponential function that returns 1 if Ω is true and 0 if it is false. Then event a is a historical relation if Then event a is a non-historical relationship. The event to be predicted is relationship r. Therefore, the comparative event network model adopts a scoring strategy based on a replication mechanism to score the correlation between the relationship and historical and non-historical events.

[0043] Execute step A430, perform time series modeling analysis based on the historical relationship frequency in combination with real-time scene parameters to obtain a first evolution path; execute step A440, perform time series modeling analysis based on the non-historical relationship frequency in combination with real-time scene parameters to obtain a second evolution path; execute step A450, perform event relationship analysis based on the first evolution path, the second evolution path and the comparative event relationship pattern to obtain multi-dimensional relationship data, and add the multi-dimensional relationship data to the potential evolution analysis result.

[0044] The time series modeling analysis based on the historical relationship frequency combined with the real-time scenario parameters refers to modeling based on the time series data of historical events using deep learning models (such as LSTM) or traditional statistical methods (such as ARIMA). The model will learn the evolution law of historical relationships and make predictions based on real-time data. Through modeling, the evolution path of historical events in the future is obtained, thereby obtaining the first evolution path.

[0045] Performing time-series modeling analysis based on the non-historical relationship frequency and combining real-time scenario parameters means conducting another time-series modeling analysis through the non-historical relationship frequency and real-time scenario data. Combining real-time data (such as temperature, load, market price, etc.), using the same time-series modeling method as the historical analysis for real-time analysis, and based on the non-historical relationship frequency, inferring the future event path in the current situation, thereby obtaining the second evolution path.

[0046] Based on the first evolution path (historical relationship analysis) and the second evolution path (non-historical relationship analysis), a comparative analysis of the event relationship patterns is carried out to identify potential relationships between events and their changing trends. By comparing the similarities and differences between historical and non-historical relationship frequencies, the system can discover which historical patterns are still valid in the current situation and which patterns need to be adjusted. This refers to evaluating their applicability in different situations by analyzing the changes of historical and non-historical relationships on the time axis. For example, some historical relationships may no longer be applicable to new environmental or market conditions and need to be adjusted. Additionally, multi-dimensional relationship data generated through comparative analysis can be included, such as time-series relationships, causal relationships, correlations, and other influencing factors (such as environment, external market, sensor data, etc.).

[0047] Add the multi-dimensional relationship data obtained from the comparative analysis of event relationship patterns to the results of the potential evolution analysis. These multi-dimensional data not only reveal the potential evolution paths of events but also provide detailed prediction data for decision-makers to help make more accurate decisions.

[0048] Next, step A500 is executed. Based on the multi-dimensional relationship data, the multi-dimensional event dataset is fused, and an action prediction for the target object is made according to the fusion result to obtain an action trend prediction result, and an action decision is formulated based on the action trend prediction result.

[0049] In a possible implementation, step A500 further includes step A510. A first correlation function is constructed, and through the first correlation function, a correlation analysis of the multi-dimensional event dataset is carried out according to the multi-dimensional relationship data to obtain historical relationship correlation parameters; the queried information is integrated using the activation function tanh, the output result is multiplied by E to convert the data into a vector, and then the corresponding relationship r in each element table ’ is scored for the similarity with the required speculative event a. Then, a copy item is added The citation score of the historical relationship in is increased, but the gradient update is not affected. Therefore, makes the attention to the historical relationship higher.

[0050] The first correlation function is:

[0051]

[0052] in, is the historical relationship correlation parameter, is the non-historical relationship frequency W his and b his is a trainable parameter (W his ∈R d×2d )(b his ∈R d ), represents the connection operator, tanh is the activation function, o∈ξ is the tail entity, s is the event state, and E is the error function.

[0053] Execute step A520 to construct a second correlation function, and use the second correlation function to perform correlation analysis on the multidimensional event data set according to the multidimensional relationship data to obtain non-historical relationship correlation parameters; based on the above, the comparison event network model not only considers historical relationships, but also non-historical relationships. Similarly, the second correlation function is:

[0054]

[0055] in, is the non-historical relationship correlation parameter, is the non-historical relationship frequency W his and b his is a trainable parameter (W his ∈R d×2d )(b his ∈R d ), represents the connection operator, tanh is the activation function, o∈ξ is the tail entity, s is the event state, E is the error function, and n is the number of event data.

[0056] Execute step A530, score the multidimensional event data set based on the historical relationship correlation parameter to obtain a first score; execute step A540, score the multidimensional event data set based on the non-historical relationship correlation parameter to obtain a second score; execute step A550, fuse the multidimensional event data set according to the first score and the second score to generate the fusion result.

[0057] The use of historical relationship correlation parameters to score multidimensional event datasets is to evaluate the rationality of the current event dataset and possible future event paths based on the correlation between events in historical data, and obtain the first score. The stronger the correlation of historical relationships, the more accurate the prediction of event occurrence, and the higher the score. The first score reflects the support of event relationships based on historical data for the current event dataset. Then, the use of non-historical relationship correlation parameters to score multidimensional event datasets is to evaluate the correlation between events in real-time data, and obtain the second score. The stronger the correlation of real-time events, the more accurate the prediction results, and the higher the score. The second score reflects the support of event relationships based on real-time data and environmental conditions for the current dataset. After obtaining the first score and the second score, the two scores need to be fused. The purpose of data fusion is to combine the correlation of historical data with real-time data to obtain a comprehensive and accurate scoring model. The final fusion result is comprehensive multidimensional data, which combines historical and non-historical information and provides a more comprehensive analysis and prediction of future events.

[0058] In a possible implementation, step A500 further includes step A560, capturing a target object, performing supervised learning on the target object according to the fusion result, and obtaining multiple prediction data; first identifying and selecting relevant target objects from the data source. In this process, the target object can be any entity that needs to be monitored, predicted, and controlled, and the target object is obtained. The target object is usually composed of multiple feature data, and these features include information such as the state, environment, and historical records related to the target object.

[0059] After capturing the target object, supervised learning is performed on the target object based on the fusion result. The goal of supervised learning is to train the model with known data so that the model can learn the relationship between input features and target outputs, and to capture the basic structure and relationship between data points by learning the intrinsic representation of the data. This method achieves effective data classification by maximizing the similarity between similar data and minimizing the similarity between dissimilar data.

[0060] After the supervised learning model is trained, new data is input for prediction to obtain multiple prediction data. These data are based on the characteristic data of the target object and the learned pattern, and the predicted target value (such as future state, behavior, etc.).

[0061] Execute step A570, construct a probability function, calculate the multiple prediction data through the probability function, and obtain the prediction probability; use the softmax method as the result of the prediction probability of all relationships, (C is the number of output nodes, r i is the output value of the i-th node), the probability function is:

[0062]

[0063] in, is the predicted probability, is the historical relationship correlation parameter, is the non-historical relationship correlation parameter, C is the number of output nodes, r i is the output value of the i-th node.

[0064] Execute step A580, perform dynamic analysis according to the multi-dimensional relationship data based on the prediction probability and the multiple prediction data, and construct an action trend prediction result.

[0065] Based on historical data, real-time scenario parameters, and prediction probabilities, combined with the results of time series modeling analysis, the action trend prediction results of the target object can be constructed. This prediction result includes the behavior trend and state changes of the target object in the future. For example, in equipment failure prediction, the action trend prediction results may include the probability of equipment failure in the future time period, the scope of impact, and the possible time of failure. The prediction results are quantified into an actionable prediction trend graph, such as a curve showing the change of action probability over time, the predicted change of system performance, etc.

[0066] The embodiments of the present application solve the technical problem that it is difficult to capture the temporal dependencies and evolution relationships between events, resulting in the inability to predict the future action trends and probabilities of the target objects, thereby achieving the technical effect of improving prediction accuracy to provide intelligent support for action decisions.

[0067] In the above, refer to Figure 1 The action analysis and prediction method driven by the contrast event network according to the embodiment of the present application is described in detail. Figure 2 A comparative event network driven action analysis and prediction system according to an embodiment of the present application is described.

[0068] The action analysis and prediction system driven by the comparative event network according to the embodiment of the present application is used to solve the technical problem that it is difficult to capture the temporal dependencies and evolution relationships between events, resulting in the inability to predict the future action trends and probabilities of the target object, so as to achieve the technical effect of improving the prediction accuracy and providing intelligent support for action decisions. The action analysis and prediction system driven by the comparative event network includes: a data acquisition module 10, a data capture module 20, a comparative analysis module 30, a modeling analysis module 40, and an action prediction module 50.

[0069] A data collection module 10 is used to collect events in a time series based on multiple data sources to obtain a multidimensional event data set, and to construct a time series knowledge graph based on the multidimensional event data set;

[0070] A data capture module 20, configured to capture data according to the time series knowledge graph, obtain multiple time series characteristics, and retrieve a historical event data set based on the multidimensional event data set;

[0071] A comparison and analysis module 30 is used to construct a comparison event network model, and compare and analyze the historical event data set according to the multiple time series characteristics through the comparison event network model to generate a comparison event relationship model;

[0072] A modeling and analysis module 40 is used to perform time series modeling and analysis based on the comparison event relationship pattern to obtain a potential evolution analysis result, wherein the potential evolution analysis result includes multi-dimensional relationship data;

[0073] The action prediction module 50 is used to fuse the multidimensional event data set based on the multidimensional relationship data, predict the action of the target object according to the fusion result, obtain the action trend prediction result, and make an action decision according to the action trend prediction result.

[0074] The specific configuration of the data acquisition module 10 will be described in detail below. As described above, the time series knowledge graph is constructed according to the multidimensional event data set, and the data acquisition module 10 may further include: aligning the multidimensional event data set according to the timestamps of the multiple data sources to obtain a multidimensional event alignment data set; clustering the multidimensional event alignment data set according to the time series to obtain multiple nodes; using graph theory to model the relationship between the multiple nodes and construct multiple relationship edges; connecting the multiple nodes through the multiple relationship edges to obtain the time series knowledge graph.

[0075] The specific configuration of the comparison and analysis module 30 will be described in detail below. As described above, the historical event data set is compared and analyzed according to multiple time series characteristics through the comparison event network model to generate a comparison event relationship pattern. The comparison and analysis module 30 may further include: traversing the time series knowledge graph to extract quadruple data, the quadruple data includes head entity data, relationship data, tail entity data, and timestamp; synchronizing the historical event data set to the comparison event network model and analyzing the quadruple according to multiple time series characteristics to construct a relationship function; obtaining a historical relationship pattern through the relationship function calculation, and determining a non-historical relationship pattern based on the historical relationship pattern; integrating the historical relationship pattern with the non-historical relationship pattern to obtain the comparison event relationship pattern.

[0076] The specific configuration of the modeling and analysis module 40 will be described in detail below. As described above, based on the comparison event relationship pattern, the time series modeling and analysis are performed to obtain the potential evolution analysis results. The modeling and analysis module 40 may further include: constructing a first relationship distribution function, calculating through the first relationship distribution function to obtain the historical relationship frequency; converting the first relationship distribution function to obtain a second relationship distribution function, calculating through the second relationship distribution function to obtain the non-historical relationship frequency; performing time series modeling and analysis based on the historical relationship frequency combined with real-time scene parameters to obtain the first evolution path; performing time series modeling and analysis based on the non-historical relationship frequency combined with real-time scene parameters to obtain the second evolution path; performing event relationship analysis based on the first evolution path, the second evolution path combined with the comparison event relationship pattern to obtain multi-dimensional relationship data, and adding the multi-dimensional relationship data to the potential evolution analysis results.

[0077] The specific configuration of the action prediction module 50 will be described in detail below. As described above, the multidimensional event data set is subjected to data fusion based on the multidimensional relationship data, and the action prediction module 50 may further include: constructing a first correlation function, performing correlation analysis on the multidimensional event data set according to the multidimensional relationship data through the first correlation function, and obtaining a historical relationship correlation parameter;

[0078] 520 constructs a second correlation function, and performs correlation analysis on the multi-dimensional event data set according to the multi-dimensional relationship data by using the second correlation function to obtain a non-historical relationship correlation parameter;

[0079] 530 scoring the multidimensional event data set based on the historical relationship correlation parameter to obtain a first score;

[0080] 540 scoring the multidimensional event data set based on the non-historical relationship correlation parameter to obtain a second score;

[0081] 550 The multi-dimensional event data set is subjected to data fusion according to the first score and the second score to generate the fusion result.

[0082] The specific configuration of the item classification module 50 will be described in detail below. As described above, the action of the target object is predicted according to the fusion result to obtain the action trend prediction result. The item classification module 50 may further include: capturing the target object, performing supervised learning on the target object according to the fusion result, and obtaining multiple prediction data; constructing a probability function, calculating the multiple prediction data through the probability function, and obtaining a prediction probability; performing dynamic analysis according to the multi-dimensional relationship data based on the prediction probability and the multiple prediction data to construct the action trend prediction result.

[0083] The comparative event network-driven action analysis and prediction system provided in the embodiments of the present application can execute the comparative event network-driven action analysis and prediction method provided in any embodiment of the present application, and has the corresponding functional modules and beneficial effects of the execution method.

[0084] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present application.

[0085] 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. Comparative event network driven action analysis and prediction method, characterized by: The method comprises: Collect events in time series based on multiple data sources to obtain a multidimensional event data set, and construct a time series knowledge graph based on the multidimensional event data set; Capturing according to the time series knowledge graph to obtain multiple time series characteristics, and retrieving a historical event dataset based on the multidimensional event dataset; Constructing a comparative event network model, and performing comparative analysis on the historical event data set according to the multiple time series characteristics through the comparative event network model to generate a comparative event relationship pattern; Performing time series modeling analysis based on the comparison event relationship pattern to obtain potential evolution analysis results, wherein the potential evolution analysis results include multi-dimensional relationship data; The multidimensional event data set is subjected to data fusion based on the multidimensional relationship data, an action prediction is performed on the target object according to the fusion result, an action trend prediction result is obtained, and an action decision is made according to the action trend prediction result.

2. The action analysis and prediction method driven by contrast event network as claimed in claim 1, characterized in that: The method of constructing a time series knowledge graph according to the multidimensional event data set includes: Aligning the multidimensional event data set according to the timestamps of the multiple data sources to obtain a multidimensional event aligned data set; Performing cluster analysis on the multidimensional event alignment data set according to time series to obtain multiple nodes; Using graph theory to model the relationship between the multiple nodes and construct multiple relationship edges; The multiple nodes are connected through the multiple relationship edges to obtain the temporal knowledge graph.

3. The action analysis and prediction method driven by contrast event network as claimed in claim 2, characterized in that: The historical event data set is compared and analyzed according to multiple time series characteristics by using the comparison event network model to generate a comparison event relationship pattern, the method comprising: Traversing the temporal knowledge graph to extract quadruple data, wherein the quadruple data includes head entity data, relationship data, tail entity data, and a timestamp; Synchronizing the historical event data set to the comparison event network model and the quadruple to analyze according to multiple time series characteristics and construct a relationship function; Obtain a historical relationship pattern by calculating the relationship function, and determine a non-historical relationship pattern by reasoning based on the historical relationship pattern; The historical relationship model and the non-historical relationship model are integrated to obtain the contrast event relationship model.

4. The action analysis and prediction method driven by contrast event network as claimed in claim 3, characterized in that: The relationship function includes: in, is a historical event data set, and time θ represents a moment before time t (θ <t), is a historical event relationship set, R is an entity, h∈ξ is the head entity, o∈ξ is the tail entity, r∈R is the relationship between the head entity and the tail entity, G θ Represents a snapshot of the temporal knowledge graph of a quadruple occurring at time θ.

5. The action analysis and prediction method driven by contrast event network as claimed in claim 1, characterized in that: Based on the comparison event relationship pattern, a time series modeling analysis is performed to obtain a potential evolution analysis result, and the method includes: Constructing a first relationship distribution function, and performing calculations using the first relationship distribution function to obtain a historical relationship frequency; The first relationship distribution function is converted to obtain a second relationship distribution function, and the non-historical relationship frequency is obtained by performing calculation using the second relationship distribution function; Perform time series modeling analysis based on the historical relationship frequency and real-time scenario parameters to obtain a first evolution path; Perform time series modeling analysis based on the non-historical relationship frequency combined with real-time scenario parameters to obtain a second evolution path; An event relationship analysis is performed based on the first evolution path, the second evolution path and the comparison event relationship pattern to obtain multi-dimensional relationship data, and the multi-dimensional relationship data is added to the potential evolution analysis result.

6. The action analysis and prediction method driven by contrast event network as claimed in claim 1, characterized in that: The method for fusing the multidimensional event data set based on the multidimensional relationship data includes: Constructing a first correlation function, and performing a correlation analysis on the multidimensional event data set according to the multidimensional relationship data by using the first correlation function to obtain a historical relationship correlation parameter; Constructing a second correlation function, and performing a correlation analysis on the multidimensional event data set according to the multidimensional relationship data by using the second correlation function to obtain a non-historical relationship correlation parameter; Scoring the multidimensional event data set based on the historical relationship correlation parameter to obtain a first score; Scoring the multidimensional event dataset based on the non-historical relationship correlation parameter to obtain a second score; The multidimensional event data set is subjected to data fusion according to the first score and the second score to generate the fusion result.

7. The action analysis and prediction method driven by contrast event network as claimed in claim 1, characterized in that: According to the fusion result, the action of the target object is predicted to obtain the action trend prediction result, and the method includes: Capturing the target object, performing supervised learning on the target object according to the fusion result, and obtaining multiple prediction data; Constructing a probability function, and calculating the plurality of prediction data by using the probability function to obtain a prediction probability; According to the prediction probability and the plurality of prediction data, dynamic analysis is performed according to the multi-dimensional relationship data to construct an action trend prediction result.

8. Comparative event network driven action analysis and prediction system, characterized by: The system is used to implement the comparative event network-driven action analysis and prediction method according to any one of claims 1 to 7, and the system comprises: A data collection module is used to collect events in time series based on multiple data sources to obtain a multidimensional event data set, and to construct a time series knowledge graph based on the multidimensional event data set; A data capture module, used to capture according to the time series knowledge graph, obtain multiple time series characteristics, and retrieve a historical event data set based on the multidimensional event data set; A comparative analysis module, used for constructing a comparative event network model, performing comparative analysis on the historical event data set according to the multiple time series characteristics through the comparative event network model, and generating a comparative event relationship pattern; A modeling and analysis module, used to perform time series modeling and analysis based on the comparison event relationship pattern to obtain a potential evolution analysis result, wherein the potential evolution analysis result includes multi-dimensional relationship data; The action prediction module is used to fuse the multidimensional event data set based on the multidimensional relationship data, predict the action of the target object according to the fusion result, obtain the action trend prediction result, and make an action decision according to the action trend prediction result.

Images