Multivariate spatio-temporal data causal cascade mode mining method and system

Abstract

The invention provides a causal cascade mode mining method and system for multivariate spatio-temporal data, and relates to the technical field of data processing. The method comprises the following steps: collecting multivariate spatio-temporal data; performing time delay sensing segmentation on the multivariate spatio-temporal data to obtain a plurality of subsequences, and defining the plurality of subsequences as a multivariate spatio-temporal event data set; establishing a causal cascade propagation probability between two different events, and selecting a causal cascade mode of which the causal cascade propagation probability is greater than a set threshold value; constructing a dynamic causal cascade network according to all the mined causal cascade modes; and screening out a conventional causal cascade network from the dynamic causal cascade network according to the frequency of the causal cascade mode. According to the invention, a variable-time-delay Granger causality inference model and a multivariate space-time cascade mode mining method are combined, a variable-time-delay Granger causality cascade mode mining algorithm is designed, and a causality cascade mode in multivariate space-time data can be accurately extracted.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for mining causal cascade patterns of multivariate spatiotemporal data. Background Art

[0002] Mining spatiotemporal correlation patterns in multivariate spatiotemporal data is of great significance to real-world issues such as intelligent transportation, environmental monitoring, and public safety. The spatiotemporal cascade pattern is a special form of the spatiotemporal correlation pattern. Based on the spatial proximity relationship, the spatiotemporal cascade pattern focuses on the temporal order of spatiotemporal events between multiple spaces, which helps to better understand the potential propagation process in spatiotemporal data.

[0003] Existing spatiotemporal cascade pattern mining methods are mainly based on time series pattern mining, which extends the correlation relationship of patterns in sequence data to multivariate spatiotemporal data, and extracts the propagation law of patterns over time and space based on the autocorrelation and time lag of multivariate spatiotemporal data. According to the method of considering cascade spatiotemporal relationships in pattern mining, the existing methods can be roughly divided into two categories: one is the spatiotemporal pattern mining method based on statistical tests, such as ST-MST, which mines spatiotemporal patterns by calculating frequency and clustering, and analyzes valuable patterns based on significance tests. The other is the cascade pattern mining method based on cascade propagation probability and transfer tree model, such as the CasInf algorithm for traffic congestion events, which mines possible cascade patterns through the temporal proximity and spatial proximity of events, and mines the spatiotemporal cascade network with the highest probability by constructing a transfer tree.

[0004] However, the above-mentioned spatiotemporal cascade pattern mining algorithm ignores the mutual correlation between multiple variables, and the existing spatiotemporal cascade pattern mining method only constructs the cascade pattern according to the time sequence of spatiotemporal events and the frequency of pattern occurrence, ignoring the causality of multivariate spatiotemporal data, resulting in the low reliability of the mined cascade pattern. In addition, although pg-Causality combines the advantages of frequent pattern mining and Gaussian Bayesian learning network, it first captures the frequent evolution patterns in spatiotemporal data, and then effectively identifies the spatiotemporal causal path by screening out reliable spatiotemporal causal relationships based on Bayesian learning. However, the symbolic processing of multivariate spatiotemporal sequences in the frequent evolution pattern mining process cannot effectively mine the causal relationship between event sequences. Summary of the invention

[0005] In view of the above problems, the purpose of the present invention is to provide a method and system for mining causal cascade patterns in multivariate spatiotemporal data. By combining the variable lag Granger causal inference model with the cascade pattern mining algorithm, a variable lag Granger causal cascade pattern mining algorithm (Variable Lag Granger Causality Cascade Inference algorithm, VLGC-CasInf) is proposed to accurately extract causal cascade patterns in multivariate spatiotemporal data. The technical solution is as follows:

[0006] In one aspect, a method for mining causal cascade patterns of multivariate spatiotemporal data is provided, the method comprising the following steps:

[0007] S1, collect multivariate spatiotemporal data;

[0008] S2, performing time-lag aware segmentation on the multivariate spatiotemporal data to obtain multiple subsequences, and defining the multiple subsequences as a multivariate spatiotemporal event dataset;

[0009] S3, establish the probability of causal cascade propagation between two different events, and select the causal cascade mode whose causal cascade propagation probability is greater than the set threshold;

[0010] S4. Construct a dynamic causal cascade network based on all the mined causal cascade patterns;

[0011] S5. Filter out the normalized causal cascade network from the dynamic causal cascade network according to the frequency of the causal cascade pattern.

[0012] Optionally, in step S1, the multivariate spatiotemporal data is a spatiotemporal data sequence with spatial information collected by sensors, including environmental monitoring data and traffic data.

[0013] Optionally, the step S2 specifically includes:

[0014] Assume that the multivariate spatiotemporal data of P variables observed at N spatial positions within T time points is X; based on the MLVAMTS segmentation algorithm, the multivariate spatiotemporal data X is segmented with time lag awareness and divided into multiple subsequences; the subsequence of the pth variable at the i-th spatial position in the time interval [ts, te] in the segmentation result is defined as the event ;

[0015] Among them, 1≤i≤N, 1≤p≤P, 1≤ts <te≤T。

[0016] Optionally, in step S3, establishing the causal cascade propagation probability between two different events specifically includes:

[0017] For any two events in the multivariate spatiotemporal event data set, based on cascade pattern mining, the variable-delay Granger causality method is used to analyze whether there is a causal relationship between the subsequences corresponding to the two events, and the probability of causal cascade propagation between the two events is calculated;

[0018] The calculation of the causal cascade propagation probability between two events includes cascade probability calculation and causal relationship inference.

[0019] Optionally, in step S3, selecting a causal cascade pattern whose causal cascade propagation probability is greater than a set threshold specifically includes:

[0020] For any spatiotemporal event of variable q in any spatial position j in the time interval [ts',te'] , assuming that the variable p in spatial position i has a spatiotemporal event in the time interval [ts,te] The probability of causal cascade propagation is the largest and greater than the set threshold γ, then define A causal cascade pattern.

[0021] Optionally, the step S4 specifically includes:

[0022] The spatial locations and variables of all mined causal cascade patterns are connected in series to construct a dynamic causal cascade network.

[0023] Optionally, the step S5 specifically includes:

[0024] The frequency threshold min_freq of the causal cascade network connection is set, and the causal cascade patterns with a frequency greater than min_freq in the dynamic causal cascade network in all time periods are calculated as the normalized causal cascade patterns. All normalized causal cascade patterns are connected in series to obtain the normalized causal cascade network.

[0025] On the other hand, a multivariate spatiotemporal data causal cascade pattern mining system is provided for implementing any of the above methods, the system comprising:

[0026] Data collection module, used to collect multivariate spatiotemporal data;

[0027] A data processing module, used for performing time-lag aware segmentation on the multivariate spatiotemporal data to obtain multiple subsequences, and defining the multiple subsequences as a multivariate spatiotemporal event data set;

[0028] The causal cascade pattern mining module is used to establish the probability of causal cascade propagation between two different events, and select the causal cascade patterns whose causal cascade propagation probability is greater than the set threshold; construct a dynamic causal cascade network based on all the mined causal cascade patterns; and screen out a regularized causal cascade network from the dynamic causal cascade network based on the frequency of the causal cascade patterns.

[0029] Optionally, the data acquisition module includes a sensor component, a serial port server, a management workstation, an interface server, an information center storage server, and a platform core switch;

[0030] Among them, the sensor component is connected to the management workstation through the serial port server, the management workstation is connected to the interface server, the interface server is connected to the information center storage server through a unidirectional isolation gateway, the information center storage server is connected to the platform core switch through optical fiber, and the platform core switch is connected to multiple cloud computing nodes and web servers.

[0031] Optionally, the system further includes a monitoring APP, and the monitoring APP includes: a client, a server, and a system management background;

[0032] The client is used for user registration and login, online query, modification and logout; the server is used for registration and login verification as well as data transmission, addition, modification and deletion; the system management background is used to manage the database.

[0033] In another aspect, an electronic device is provided, the electronic device comprising:

[0034] processor;

[0035] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are loaded and executed by the processor, the steps of the above-mentioned multivariate spatiotemporal data causal cascade pattern mining method are implemented.

[0036] On the other hand, a computer-readable storage medium is provided, in which program code is stored. The program code can be called by a processor to execute the steps of the above-mentioned multivariate spatiotemporal data causal cascade pattern mining method.

[0037] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0038] In an embodiment of the present invention, in order to accurately extract causal cascade patterns in multivariate spatiotemporal data, after completing the time-lag-aware segmentation, the existing univariate cascade pattern mining algorithm is first extended to multivariate spatiotemporal data, and then the variable-lag Granger causal inference model is combined with the multivariate spatiotemporal cascade pattern mining method, and the causal cascade propagation probability is designed to mine the dynamic multivariate spatiotemporal causal cascade patterns in the multivariate spatiotemporal data based on it. Finally, the dynamic causal cascade patterns of different time intervals are integrated into a regularized causal cascade network, providing an overall representation of the causal cascade relationship in the multivariate spatiotemporal data.

[0039] The present invention also constructs a multivariate spatiotemporal data causal cascade pattern mining system, which consists of three main parts: data collection, causal cascade pattern mining, and monitoring APP, which are interconnected to form a complete system, providing a useful reference for research and development in related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 is a flow chart of a method for mining causal cascade patterns of multivariate spatiotemporal data provided by an embodiment of the present invention;

[0042] Figure 2 are two cascade network examples in the dynamic cascade network set provided by an embodiment of the present invention;

[0043] Figure 3 is a flow chart of normalized causal cascade network mining provided by an embodiment of the present invention;

[0044] Figure 4 It is a schematic diagram of comparison results of mining patterns under different causal cascade propagation probability thresholds and frequency thresholds provided by an embodiment of the present invention;

[0045] Figure 5 is a schematic diagram of the comparison results of the running time of various algorithms provided by the embodiments of the present invention on three actual data sets;

[0046] Figure 6 It is a visualization diagram of a part of the normalized causal cascade network in the Beijing air pollution data set provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the word "exemplary" is used to present concepts in a specific way.

[0049] In order to accurately extract causal cascade patterns in multivariate spatiotemporal data, an embodiment of the present invention provides a method for mining causal cascade patterns in multivariate spatiotemporal data. After completing the time-lag-aware segmentation of multivariate spatiotemporal data, the existing single-variable cascade pattern mining algorithm is first extended to multivariate spatiotemporal data, and then the variable-lag Granger causal inference model is combined with the multivariate spatiotemporal cascade pattern mining method, and the causal cascade propagation probability is designed to mine the dynamic causal cascade pattern in the multivariate spatiotemporal data based on it. Finally, the dynamic causal cascade patterns of different time intervals are integrated into a normalized causal cascade network, providing an overall representation of the causal cascade relationship in the multivariate spatiotemporal data.

[0050] Specifically, Figure 1 As shown, the multivariate spatiotemporal data causal cascade pattern mining method includes the following steps:

[0051] S1. Collect multivariate spatiotemporal data.

[0052] In the embodiment of the present invention, multivariate spatiotemporal data is a spatiotemporal data sequence with spatial information collected by sensors, including environmental monitoring data, traffic data, etc.

[0053] S2. Perform time-lag-aware segmentation on the multivariate spatiotemporal data to obtain multiple subsequences, and define the multiple subsequences as a multivariate spatiotemporal event dataset.

[0054] Assume that the multivariate spatiotemporal data of P variables observed at N spatial locations within T time is X, ; Based on the MLVAMTS segmentation algorithm, the multivariate spatiotemporal data X is segmented with time lag awareness and divided into multiple subsequences; the subsequence of the pth variable at the i-th spatial position in the segmentation result in the time interval [ts,te] is Defined as an event .

[0055] Among them, 1≤i≤N, 1≤p≤P, 1≤ts <te≤T。

[0056] In the embodiment of the present invention, it is necessary to mine multivariate causal cascade patterns for all time subsequences in multivariate spatiotemporal data and represent the mining results as causal cascade graphs. Table 1 introduces the symbols and meanings involved in the present invention.

[0057] Table 1 Symbols and meanings

[0058]

[0059] Among the massive multivariate spatiotemporal events, the present invention focuses more on the important event subsequences. The selection of important events needs to be determined according to the specific practical application field. For example, in the field of air pollution in environmental monitoring, the focus is on pollution events of multiple pollutants. If the air pollution data is segmented into subsequences, If the average pollution concentration value in exceeds the pollution concentration standard value, then the subsequence is represented as a pollution event subsequence. Based on the above definition of important events, the present invention further defines all multivariate spatiotemporal important events on the location i variable p sequence as the set ,Since the changing trends of different variables in different locations are different, the number of important spatiotemporal events in different sequences is different.

[0060] S3. Establish the probability of causal cascade propagation between two different events, and select the causal cascade mode whose causal cascade propagation probability is greater than the set threshold.

[0061] First, the spatiotemporal cascade pattern mining algorithm and Granger causality model are introduced. Most spatiotemporal cascade pattern mining algorithms are based on propagation probability and transfer tree. For example, the Cascade Inference Algorithm (CasInf) for single variable spatiotemporal cascade pattern mining algorithm mines accurate cascade relationships by designing the propagation probability related to the time and place of the event, and integrates the complex cascade relationship into a cascade network through the transfer tree.

[0062] Given N spatial position nodes A multivariate spatiotemporal event dataset of a specific variable within T days. The algorithm first extracts a set of dynamic cascade networks , where c t represents the cascade network extracted on day t, Figure 2 (a) and (b) show examples of cascade networks on day 1 and day 2. Based on this set of extracted dynamic cascade networks C, an implicit normalized cascade network G can be inferred.

[0063] The algorithm determines the cascade relationship based on the temporal proximity and spatial proximity of multivariate spatiotemporal events, which divides the event into i The starting time of occurrence ts i and in r j The starting time of occurrence ts j The time difference between them is regarded as the propagation time. The longer the propagation time, the faster the event is at node r. i To r jThe smaller the probability of transmission between them. In addition, due to the influence of spatial distance, the direct impact of adjacent spatial locations will be greater than the indirect impact of distant locations. Integrating time distance and spatial distance into the exponential model is used to describe the multivariate spatiotemporal event from location node r i Spread to r j The conditional probability of:

[0064] (1)

[0065] in Represents multivariate spatiotemporal events from spatial nodes r i Spread to r j The time difference, d i,j Represents spatial node r i With r j The spatial distance between i,j represents the transfer rate, α is the setting parameter, and λ is the weight of spatial distance in the propagation probability.

[0066] According to the propagation probability calculated from different spatial multivariate spatiotemporal events within the tth day, the maximum propagation probability is selected as the cascade connection and the cascade network c within the tth day is obtained. t And integrate all cascade networks within T days into the dynamic cascade network set C.

[0067] The algorithm assumes that each set of cascade relationships in the normalized cascade network G is conditionally independent. Therefore, the joint probability of a set of cascades in G can be simply expressed as follows:

[0068] (2)

[0069] Given the observed multivariate spatiotemporal data and a set of cascade networks C obtained from a given time span of each day, the algorithm transforms the problem into a cascade network inference problem based on the possibility of pairwise propagation of spatiotemporal distances. The goal is to find a network , so that:

[0070] (3)

[0071] By optimizing formula (3), the obtained cascade network can be The highest probability encompasses and explains the observed dynamic cascade network.

[0072] The Granger causality model is a method of causal inference in time series data. The causal inference model assumes that each timestamp that affects the time series will be affected by other time series with a fixed time lag. However, in practical applications, a series of timestamps in two time series may have a non-fixed time lag mutual influence. Since the traditional Granger causality inference algorithm can only analyze the causal relationship with a fixed time lag between two series, the traditional Granger causality test cannot correctly infer the variable time lag Granger causality between different time series for multivariate spatiotemporal data with dynamic time lags.

[0073] In order to solve this problem, the present invention uses the optimal warping path of dynamic time warping to infer variable-lag causality to solve the fixed-lag restriction in Granger causality. The definitions involved in the present invention are as follows:

[0074] Definition 1: Time series warp alignment. The warp alignment of time series is to define a warped sequence for two time series x and y ,in Align the timestamps x(t) and y(t') in the two time series.

[0075] Definition 2: Granger causality with variable lag. For time series x and y, assume is the maximum lag (the upper limit of the lag between any two time series that are considered causal). The variable lag Granger causality defines the regression residual at time stamp t for:

[0076] (4)

[0077] in , Δ t >0 is the best alignment sequence P for x and y l The time delay constant in the i 、b i and c i Let the residual set Variance Minimize. If its variance Less than the variance of the sequence y , then define x as time-delay Granger causality leading to y.

[0078] Based on the above theory, an embodiment of the present invention proposes a multivariate spatiotemporal data causal cascade pattern mining algorithm VLGC-CasInf based on variable-delay Granger causality.

[0079] First, establish the probability of causal cascade propagation between two different events. For any two events in the multivariate spatiotemporal event dataset: and ,in Based on cascade pattern mining, the subsequences corresponding to two events are analyzed by using the variable-delay Granger causality method. and Is there a causal relationship between them, and calculate the probability of causal cascade propagation between the two events, as shown in formula (5):

[0080] (5)

[0081] Formula (5) consists of two parts: cascade probability calculation and causal relationship inference.

[0082] 1) Cascade Probability middle Represents two events and The difference between the occurrence time ts and ts'; d i,j It represents the spatial distance between the locations i and j of two event subsequences. The cascade probability formula takes into account the temporal proximity and spatial proximity of two events to determine whether there is a potential cascade propagation relationship between the two events.

[0083] 2) Causal Inference Represents two events and Whether there is a variable-lag Granger causality relationship between them, the calculation formula is as follows:

[0084] (6)

[0085] in For event subsequence The variance of For collection The variance of For events The variable-lag Granger causality regression residual with time stamp ν in is calculated as:

[0086] (7)

[0087] in The maximum time lag range set for variable time lag causal inference. The variable time lag alignment sequence is defined in the causal inference process. , where Δ ν >0 is P * The time delay constant of the timestamp ν is calculated by Δ ν Get alignment timestamp Based on the analysis of the cascade propagation relationship between events, the variable-time-delay Granger causal inference model is used to further screen out events with causal relationships and explore more rigorous and accurate causal cascade relationships.

[0088] Secondly, for any space-time event of variable q in the time interval [ts',te'] at a spatial location j , assuming that the variable p in spatial position i has a spatiotemporal event in the time interval [ts,te] The probability of causal cascade propagation The maximum value is greater than the set threshold γ, then define is a causal cascade pattern. The purpose of selecting the causal cascade propagation probability greater than the set threshold γ is to ensure the accuracy of the cascade; selecting the nodes and variables with the largest causal cascade propagation probability can ensure that the obtained causal cascade pattern has high reliability and avoid cross-level propagation.

[0089] S4. Construct a dynamic causal cascade network based on all the mined causal cascade patterns.

[0090] After mining all the causal cascade patterns in the continuous event subsequence, the spatial locations and variables where all the causal cascade patterns occur are connected in series to construct a dynamic causal cascade network. Since the mining results of causal cascade patterns in different time intervals are not completely consistent, a series of dynamic causal cascade networks are finally obtained. , where K is the number of disconnected causal cascade networks in the mined multivariate spatiotemporal data.

[0091] S5. Filter out the normalized causal cascade network from the dynamic causal cascade network according to the frequency of the causal cascade pattern.

[0092] Through the above steps, the multivariate spatiotemporal causal cascade patterns of different time periods and their corresponding dynamic causal cascade networks have been excavated. Based on the dynamic causal cascade network, the present invention attempts to filter out reliable regularized causal cascade propagation paths according to the frequency of causal cascades and connect them into a regularized causal cascade propagation network G. The process of filtering out regularized causal cascade networks from the dynamic causal cascade network set is as follows: Figure 3 As shown in the figure, there are two parts: the dynamic multivariate spatiotemporal causal cascade network and the conventional multivariate spatiotemporal causal cascade network. The dynamic multivariate spatiotemporal causal cascade network part gives the dynamic causal cascade patterns and network examples of K different time periods obtained by mining. Due to the different time spans and numbers of event subsequences in different time periods, the time periods of the obtained dynamic causal cascade network are not completely consistent, and the causal cascade connections in different time periods are dynamically changing.

[0093] For example, in Figure 3 In the example of the dynamic causal cascade network of Beijing's air pollution data, the causal cascade pattern in period 1 It means that the pollution event of variable 1 (PM2.5) at location 3 (Guanyuan Monitoring Station) led to the pollution event of variable 1 (PM2.5) at location 2 (Dongsi Monitoring Station), and the causal cascade pattern in period 2 and It means that the pollution event of variable 1 (PM2.5) at location 3 (Guanyuan Monitoring Station) caused the pollution of variable 2 (NO 2 ) pollution events and the occurrence of variable 1 (PM2.5) events at location 6 (Daxing monitoring station). In addition, the causal cascade pattern is partitioned according to its length to intuitively display the length of the causal cascade pattern, such as Cascade 2-pattern in period 2 and 3-mode .

[0094] In order to mine the implicit conventional multivariate spatiotemporal causal cascade network from the dynamically changing patterns and networks, the present invention introduces the concept of causal cascade network connection frequency, sets the frequency threshold min_freq of causal cascade network connection, calculates the causal cascade pattern with a frequency greater than min_freq in the dynamic causal cascade network in all K time periods, and takes it as the conventional causal cascade pattern. All conventional causal cascade patterns are connected in series to obtain a conventional causal cascade network. Figure 3 As shown in the example of the normalized multivariate spatiotemporal causal cascade network in , the mined normalized causal cascade network is not a single network, but a network collection composed of multiple groups of unconnected normalized cascade networks.

[0095] The reason why the multivariate causal cascade patterns and networks mined from multivariate spatiotemporal data change dynamically over time is that many factors in space are not constant. However, the sources of many spatiotemporal events are generally fixed, so there are often strong conventional paths implied in the dynamic cascade patterns. Therefore, in the present invention, the conventionalized multivariate causal cascade network is mined from these dynamic causal cascade patterns through the concept of frequency. In the embodiment of the present invention, the causal cascade connection of all events at location i variable p and all events at location j variable q is calculated. The frequency is as follows:

[0096] (8)

[0097] in is the number of causal cascade patterns in which events propagate from variable p in position i to variable q in position j in the dynamic causal cascade network, K is the number of causal cascade networks mined from multivariate spatiotemporal data, if If it is greater than the frequency threshold min_freq, then the cascade relationship from the variable q in position j to the variable p in position i in the causal cascade network is defined as a frequent causal cascade relationship, which is used as a connection in the regularized causal cascade network to screen out the frequent causal cascade connections in the dynamic causal cascade network and construct the regularized causal cascade network G.

[0098] In an embodiment of the present invention, in order to accurately extract causal cascade patterns in multivariate spatiotemporal data, after completing the time-lag-aware segmentation, the existing univariate cascade pattern mining algorithm is first extended to multivariate spatiotemporal data, and then the variable-lag Granger causal inference model is combined with the multivariate spatiotemporal cascade pattern mining method, and the causal cascade propagation probability is designed to mine the dynamic multivariate spatiotemporal causal cascade patterns in the multivariate spatiotemporal data based on it. Finally, the dynamic causal cascade patterns of different time intervals are integrated into a regularized causal cascade network, providing an overall representation of the causal cascade relationship in the multivariate spatiotemporal data.

[0099] The specific implementation steps of the VLGC-CasInf algorithm are shown in Algorithm 1.

[0100]

[0101] Algorithm 1 mainly consists of two parts: first, dynamic multivariate spatiotemporal causal cascade patterns are mined from the time-delay-aware segmentation results of multivariate spatiotemporal data, and then the normalized multivariate spatiotemporal causal cascade patterns and networks are further screened on this basis. In the process of mining dynamic causal cascade patterns, important spatiotemporal events must first be scanned from the segmentation results, and its complexity is O(nPN), where n is the number of subsequences of the time-delay-aware segmentation. Secondly, the causal relationship and cascade propagation probability between important spatiotemporal events are mined on the spatiotemporal event data set. In this process, each variable at each location needs to be mined, and the search complexity is O(mP 2 N 2 ), where m is the average number of important spatiotemporal events in each sequence; at the same time, the complexity of calculating the probability of cascade propagation and causal inference mainly lies in causal inference, and its complexity is O(T / n). Therefore, the time complexity of mining dynamic multivariate spatiotemporal causal cascade patterns is O(mP 2 N 2 T / n). In the second part of mining the normalized multivariate spatiotemporal causal cascade pattern, it is necessary to calculate the frequency of each pattern in the dynamic cascade pattern, and its time complexity is O(mk), where k is the number of cascade network edges. In summary, the time complexity of the entire algorithm 1 is O((mP 2 N 2 T / n)+mk).

[0102] Accordingly, an embodiment of the present invention further provides a method and system for mining causal cascade patterns of multivariate spatiotemporal data, the system comprising:

[0103] Data collection module, used to collect multivariate spatiotemporal data;

[0104] A data processing module, used for performing time-lag aware segmentation on the multivariate spatiotemporal data to obtain multiple subsequences, and defining the multiple subsequences as a multivariate spatiotemporal event data set;

[0105] The causal cascade pattern mining module is used to establish the probability of causal cascade propagation between two different events, and select the causal cascade patterns whose causal cascade propagation probability is greater than the set threshold; construct a dynamic causal cascade network based on all the mined causal cascade patterns; and screen out a regularized causal cascade network from the dynamic causal cascade network based on the frequency of the causal cascade patterns.

[0106] The system of this embodiment can be used to execute Figure 1 The technical solution of the method embodiment shown has similar implementation principles and technical effects, which will not be repeated here.

[0107] Furthermore, the data acquisition module includes a sensor component, a serial port server, a management workstation, an interface server, an information center storage server, and a platform core switch.

[0108] Among them, the sensor component is connected to the management workstation through the serial port server, the management workstation is connected to the interface server, the interface server is connected to the information center storage server through a unidirectional isolation gateway, the information center storage server is connected to the platform core switch through optical fiber, and the platform core switch is connected to multiple cloud computing nodes and web servers.

[0109] The data acquisition module is a supporting system responsible for sensor monitoring data acquisition. It uses C++ as the development language and is embedded with multiple data communication protocols such as 101, 102, 103, 104, Modbus, CDT, DISA, etc. of IEC 60870-5. The modeling complies with the requirements of the interface reference model, common information model (CIM) and component interface specification (CIS) in IEC 61970, complies with international standards, and can be seamlessly integrated with various systems as middleware; it realizes the access of system data such as monitoring system, integrated energy management and control system, metering, fault analysis and alarm push. The system supports the access of multiple devices and has the ability to parse multiple protocols.

[0110] The data acquisition module adopts a two-level architecture, namely a single integrated management and control system and a cloud platform centralized monitoring system. The integrated management and control system is used to collect real-time operation monitoring data, realize local data monitoring, historical data sampling and storage, and upload key real-time data to the cloud platform centralized monitoring system. The cloud platform centralized monitoring system obtains real-time monitoring data. The communication protocol between the two-level systems can adopt the power standard IEC104 protocol or other protocols. The real-time data acquisition frequency supports seconds according to the protocol requirements, and can support change quantity upload, cyclic upload and call modes.

[0111] Furthermore, the system also includes a monitoring APP, and the monitoring APP includes: a client, a server and a system management background.

[0112] The client is used for user registration and login, online query, modification and logout; the server is used for registration and login verification as well as data transmission, addition, modification and deletion; the system management background is used to manage the database.

[0113] The client is developed and designed using the MUI front-end framework, and HTML5, CSS, and JavaScript are used for front-end development, which is used for user registration and login, online query, modification, and logout. The server is developed using the ThinkJS server-side framework, combined with the MySQL database, which is used for registration and login verification as well as data transmission, addition, modification, and deletion. The system management backend is developed using HTML5, CSS, and JavaScript to manage the database.

[0114] The monitoring APP is easy to operate and has a simple and beautiful interface. It is real-time, and registered users can log in to the system through the mobile APP no matter where they are. The system provides automatic query and display functions, as well as user registration information management functions.

[0115] The multivariate spatiotemporal data causal cascade pattern mining system constructed by the present invention is composed of three main parts: data collection, causal cascade pattern mining, and monitoring APP, which are interconnected to form a complete system, providing a useful reference for research and development in related fields.

[0116] In terms of experimental verification, this paper uses three real data sets in the fields of traffic and air pollution to conduct experiments to evaluate the proposed algorithm. All experiments are implemented using the Python 3.8 compiler in a Win10 operating system, a 2.30GHz processor, and a 16GB memory environment.

[0117] The three datasets include:

[0118] (1) Air-BJ: Air-BJ contains the PM2.5, PM10, NO2, and O3 concentration data of 34 air pollution monitoring stations in Beijing in 2018, with a sampling period of 1 hour.

[0119] For air pollution data, the present invention focuses on pollution events in the data. According to the Ambient Air Quality Standard (GB3095-2012), the PM2.5 pollution concentration standard is 75 mg / m 3 , PM10 pollution concentration standard is 100mg / m 3 , NO 2 The pollution concentration standard is 40mg / m 3 , O 3 The pollution concentration standard is 100mg / m 3 When the average pollutant concentration in a subsequence exceeds the pollution concentration standard, the subsequence is defined as a pollution event subsequence.

[0120] (2) PeMSD7: The PeMSD7 dataset contains traffic flow data from May to June 2012 collected by 228 sensors on roads in seven districts in California, with a sampling period of 5 minutes.

[0121] (3) PeMSD8: PeMSD8 contains traffic flow data from July to August 2016 collected by 170 sensors on 8 roads in the San Bernardino area, with a sampling period of 5 minutes.

[0122] For traffic data, if the car speed is less than 20km / h, it is defined as congestion. In the segmented subsequences of the traffic data set, the subsequence with congestion is defined as the congestion event subsequence.

[0123] In order to evaluate the effectiveness of the mining algorithm described in the present invention, a time window γ is given. w , if the method predicts the next location and variable that will be polluted (congested) within a given time window, it means a hit. In this invention, the occurrence probability is used to indicate the effectiveness of the cascade pattern discovered by the model. The calculation formula of the occurrence probability is as follows:

[0124] (9)

[0125] in Represents the prediction data in the cascade network The number of correct predictions of the propagation relationship between It represents the number of important spatiotemporal events of variable p in location i in historical data. In order to verify the correctness of the cascade pattern generated by the method, the present invention defines score as an evaluation criterion.

[0126] (10)

[0127] Where E G is the set of edges in the cascade network G, is the number of edges in the causal cascade network.

[0128] In addition, in order to analyze the operation efficiency of the algorithm, the average running time of cascade pattern mining is used as an evaluation indicator.

[0129] In order to verify the effectiveness of the algorithm, the following spatiotemporal cascade pattern mining algorithm and spatiotemporal association pattern mining algorithm were selected to compare and analyze the performance of the algorithm.

[0130] 1) CasInf. The algorithm first discretizes the traffic data, models the influences existing in traffic propagation, and uses a data-driven approach to mine cascade patterns.

[0131] 2) ST-MST. This algorithm mines temporal patterns from multivariate spatiotemporal data using information theory methods and uses statistical significance tests to identify valuable patterns.

[0132] 3) pg-Causality. This algorithm combines frequent pattern mining with Bayesian learning. It first mines frequent patterns and then uses a Gaussian Bayesian network-based model to encode spatiotemporal causality.

[0133] 4) GC-CasInf. The proposed VLGC-CasInf algorithm replaces the variable-delay Granger causal inference part with the traditional Granger causal inference method.

[0134] In order to evaluate the impact of two important parameters introduced in the VLGC-CasInf algorithm, the causal cascade propagation probability threshold γ and the frequency threshold min_freq, on the accuracy of the mined patterns, the score values ​​calculated by the VLGC-CasInf algorithm when the causal cascade propagation probability threshold is 0.2 to 0.9 (frequency threshold min_freq=0.5) and the frequency threshold is 0.1 to 0.8 (causal cascade propagation probability threshold γ=0.7) are compared on three data sets in the embodiment of the present invention. The experimental results are as follows: Figure 4 As shown in (a) and (b).

[0135] from Figure 4 From the results in (a), we can see that the causal cascade patterns mined by the VLGC-CasInf algorithm are significantly affected by the causal cascade propagation probability threshold γ. When the value of γ is small, the accuracy of the patterns mined by the algorithm is low, and it is difficult to mine reliable causal cascade patterns. When the value of γ is between 0.7 and 0.8, the accuracy of the mining pattern tends to be flat. Therefore, selecting γ=0.7 can mine more reliable causal cascade patterns. Selecting too large a value will filter out some reliable patterns. Figure 4 From the result in (b), we can see that the causal cascade pattern mined by the VLGC-CasInf algorithm is less affected by the frequency threshold min_freq, and the impact on accuracy tends to be flat when the frequency threshold is greater than 0.5. Therefore, selecting the frequency threshold min_freq=0.5 can mine a reliable normalized causal cascade network.

[0136] In order to evaluate the accuracy of VLGC-CasInf in mining normalized multivariate spatiotemporal causal cascade patterns, the score values ​​calculated by the patterns mined by various algorithms and the overall average values ​​of the three data sets are compared here, and the results are shown in Table 2. In the experimental setting, the causal cascade propagation probability threshold γ=0.7 and the frequency threshold min_freq=0.5 of VLGC-CasInf and the ablation comparison algorithm GC-CasInf are set according to the results of parameter evaluation; the prediction time window γ set for the Air-BJ data set in the experiment w The prediction time window γ is set to 6 hours for the two traffic datasets. w It should be noted that, since the CasInf algorithm can only mine single variable cascade patterns, the present invention uses the CasInf algorithm to mine the Air-BJ dataset, extracting a corresponding cascade network for each variable and obtaining a cascade network set of all variables.

[0137] Table 2 Comparison of the scores of mining patterns of various algorithms on three actual data sets

[0138]

[0139] From the results in Table 2, it can be seen that VLGC-CasInf outperforms the existing pattern mining algorithms in the three data sets and the overall results. Comparing the results on the Air-BJ data set and the two data sets in the traffic field, it can be seen that the CasInf algorithm can only mine the cascade relationship in the univariate data, so it shows the worst performance on this multivariate spatiotemporal data, while in the univariate spatiotemporal data sets of the two traffic data, its performance is similar to that of the ST-MST algorithm. In addition, VLGC-CasInf, GC-CasInf and pg-Causality can mine more accurate spatiotemporal causal relationships because they add causal inference mechanisms on the basis of pattern mining, and the accuracy of VLGC-CasInf using variable time-lag causal inference is significantly better than GC-CasInf using traditional Granger inference and pg-Causality algorithm using Gaussian Bayesian inference, which proves the effectiveness of variable time-lag Granger causal inference in causal cascade pattern mining. However, the CasInf and ST-MST algorithms only mine relevant patterns from spatiotemporal proximity and statistical methods, focusing on the spatiotemporal patterns of frequently interacting neighboring spaces, ignoring the correlation between the changing trends of spatiotemporal events, resulting in large deviations in their prediction results.

[0140] In order to evaluate the efficiency of VLGC-CasInf, the running time of the other three comparison algorithms and the ablation comparison algorithm on three actual data sets is compared here, using the average running time of spatiotemporal pattern mining as a metric. The experimental results are as follows: Figure 5 shown.

[0141] like Figure 5 As shown in the figure, compared with the single variable spatiotemporal cascade pattern mining algorithm CasInf, VLGC-CasInf takes longer running time because it considers the changing trend of the sequence in the process of variable time-lag causal inference. In addition, since the variable time-lag perception process is added to the variable time-lag Granger causal inference, the VLGC-CasInf algorithm needs to consume more running time compared with the GC-CasInf algorithm using traditional Granger causal inference. However, compared with the ST-MST and pg-Causality algorithms, the cascade pattern is first used to mine the patterns that may have cascade relationships, and the causal patterns are calculated from them, which greatly improves the running efficiency of the algorithm. Although the pg-Causality algorithm captures frequent evolutionary patterns through frequent pattern mining, the Gaussian Bayesian network is used to mine causal relationships, which increases its computational complexity. The ST-MST algorithm not only mines frequent item sets, but also uses clustering algorithms to cluster similar item sets, and then obtains significant patterns in different spaces through statistical methods. Due to the large number of algorithm processes, its computational complexity is the highest and it consumes the most computing time compared to other algorithms.

[0142] In the embodiment of the present invention, an example of a normalized multivariate spatiotemporal causal cascade network is as follows: Figure 6 As shown in Figure 2, the conventional multivariate spatiotemporal causal cascade network mined from the Beijing air pollution dataset is taken as an example, and the part of the network centered on the PM2.5 pollutants at the Temple of Heaven pollution monitoring station is selected, as shown in Figure 2. Figure 6 As shown in (a). In each event node Z in the graph i,p In the data, locations 2, 3, 11, 13, 15, 18, and 33 represent the Temple of Heaven, Guanyuan, Gucheng, Daxing, Tongzhou, Mentougou, and South Third Ring Road pollution monitoring stations, respectively, and variables 1, 2, and 3 represent PM2.5, PM10, and NO2 pollutants, respectively. Figure 6 The causal cascade network in (a) is visualized on the actual Beijing map as follows: Figure 6 The result of (b).

[0143] In summary, in order to extract dynamic spatiotemporal correlation information and normalized spatiotemporal correlation relationships in multivariate spatiotemporal data to mine the propagation process of spatiotemporal events, after completing the time-delay-aware segmentation, it is also necessary to effectively mine the spatiotemporal events obtained by segmentation. The present invention combines the variable time-delay Granger causal inference model with the cascade pattern mining algorithm to propose a multivariate spatiotemporal causal cascade pattern mining algorithm VLGC-CasInf. The algorithm first extends the existing single-variable cascade pattern mining algorithm to multivariate, and uses the idea of ​​cascade pattern mining to discover the cascade propagation probability between the occurrence time and location of spatiotemporal events between different spaces and different variables in multivariate spatiotemporal data, and further uses variable time-delay Granger causality to infer whether there is a reliable causal relationship between these spatiotemporal events, so as to mine dynamic causal cascade patterns. In order to further realize the traceability of spatiotemporal events, a normalized causal cascade pattern and a normalized causal cascade network are integrated on the basis of the dynamic causal cascade pattern.

[0144] The invention also constructs a multivariate spatiotemporal data causal cascade pattern mining system, which consists of three main parts: data collection, causal cascade pattern mining, and monitoring APP, which are interconnected to form a complete system, providing useful reference for research and development in related fields.

[0145] In an exemplary embodiment, the present invention further provides an electronic device, the electronic device comprising:

[0146] processor;

[0147] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are loaded and executed by the processor, the steps of the above-mentioned multivariate spatiotemporal data causal cascade pattern mining method are implemented.

[0148] In an exemplary embodiment, the present invention further provides a computer-readable storage medium, wherein at least one instruction is stored in the computer-readable storage medium, and the at least one instruction is loaded and executed by a processor to implement the steps of the multivariate spatiotemporal data causal cascade pattern mining method as described above. For example, the computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0149] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0150] References in the specification to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments", etc. indicate that the described embodiments may include a particular feature, structure, or characteristic, but not every embodiment may include the particular feature, structure, or characteristic. In addition, when a particular feature, structure, or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the relevant art to implement such feature, structure, or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0151] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0152] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0153] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0154] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0155] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0156] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0157] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0158] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0159] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for mining causal cascade patterns of multivariate spatiotemporal data, characterized in that: The following steps are involved: S1, collect multivariate spatiotemporal data; S2, performing time-lag aware segmentation on the multivariate spatiotemporal data to obtain multiple subsequences, and defining the multiple subsequences as a multivariate spatiotemporal event dataset; S3, establish the probability of causal cascade propagation between two different events, and select the causal cascade mode whose causal cascade propagation probability is greater than the set threshold; S4. Construct a dynamic causal cascade network based on all the mined causal cascade patterns; S5. Filter out the normalized causal cascade network from the dynamic causal cascade network according to the frequency of the causal cascade pattern.

2. The method for mining causal cascade patterns of multivariate spatiotemporal data according to claim 1, characterized in that: In step S1, the multivariate spatiotemporal data is a spatiotemporal data sequence with spatial information collected by sensors, including environmental monitoring data and traffic data.

3. The method for mining causal cascade patterns of multivariate spatiotemporal data according to claim 1, characterized in that: The step S2 specifically includes: Assume that the multivariate spatiotemporal data of P variables observed at N spatial positions within T time points is X; based on the MLVAMTS segmentation algorithm, the multivariate spatiotemporal data X is segmented with time lag awareness and divided into multiple subsequences; the subsequence of the pth variable at the i-th spatial position in the time interval [ts, te] in the segmentation result is defined as the event ; Among them, 1≤i≤N, 1≤p≤P, 1≤ts <te≤T。 4. The method for mining causal cascade patterns of multivariate spatiotemporal data according to claim 1, characterized in that: In step S3, establishing the causal cascade propagation probability between two different events specifically includes: For any two events in the multivariate spatiotemporal event data set, based on cascade pattern mining, the variable-delay Granger causality method is used to analyze whether there is a causal relationship between the subsequences corresponding to the two events, and the probability of causal cascade propagation between the two events is calculated; The calculation of the causal cascade propagation probability between two events includes cascade probability calculation and causal relationship inference.

5. The method for mining causal cascade patterns of multivariate spatiotemporal data according to claim 1, characterized in that: In step S3, selecting a causal cascade pattern whose causal cascade propagation probability is greater than a set threshold specifically includes: For any spatiotemporal event of variable q in any spatial position j in the time interval [ts',te'] , assuming that the variable p in spatial position i has a spatiotemporal event in the time interval [ts,te] The probability of causal cascade propagation is the largest and greater than the set threshold γ, then define A causal cascade pattern.

6. The method for mining causal cascade patterns of multivariate spatiotemporal data according to claim 1, characterized in that: The step S4 specifically includes: The spatial locations and variables of all mined causal cascade patterns are connected in series to construct a dynamic causal cascade network.

7. The method for mining causal cascade patterns of multivariate spatiotemporal data according to claim 1, characterized in that: The step S5 specifically includes: The frequency threshold min_freq of the causal cascade network connection is set, and the causal cascade patterns with a frequency greater than min_freq in the dynamic causal cascade network in all time periods are calculated as the normalized causal cascade patterns. All normalized causal cascade patterns are connected in series to obtain the normalized causal cascade network.

8. A multivariate spatiotemporal data causal cascade pattern mining system, the system being used to implement the method according to any one of claims 1 to 7, characterized in that: The system comprises: Data collection module, used to collect multivariate spatiotemporal data; A data processing module, used for performing time-lag aware segmentation on the multivariate spatiotemporal data to obtain multiple subsequences, and defining the multiple subsequences as a multivariate spatiotemporal event data set; The causal cascade pattern mining module is used to establish the probability of causal cascade propagation between two different events, and select the causal cascade patterns whose causal cascade propagation probability is greater than the set threshold; construct a dynamic causal cascade network based on all the mined causal cascade patterns; and screen out a regularized causal cascade network from the dynamic causal cascade network based on the frequency of the causal cascade patterns.

9. The multivariate spatiotemporal data causal cascade pattern mining system according to claim 8, characterized in that: The data acquisition module includes a sensor component, a serial port server, a management workstation, an interface server, an information center storage server, and a platform core switch; Among them, the sensor component is connected to the management workstation through the serial port server, the management workstation is connected to the interface server, the interface server is connected to the information center storage server through a unidirectional isolation gateway, the information center storage server is connected to the platform core switch through optical fiber, and the platform core switch is connected to multiple cloud computing nodes and web servers.

10. The multivariate spatiotemporal data causal cascade pattern mining system according to claim 8, characterized in that: The system also includes a monitoring APP, which includes: a client, a server and a system management background; The client is used for user registration and login, online query, modification and logout; the server is used for registration and login verification as well as data transmission, addition, modification and deletion; the system management background is used to manage the database.

Images