Ecological value anomaly event tracing method, system and device and storage medium

Abstract

The invention relates to the technical field of natural language processing, in particular to an ecological value anomaly event tracing method, system and device and a storage medium, and the method comprises the following steps: processing log text data to obtain a local expression of an event and a global expression of the event, and establishing an event expression structure; then, obtaining summary information of the real-time actuarial system by utilizing the event expression structure; and finally, according to summary information of the real-time actuarial system, performing ecological value abnormal event tracing. According to the method, the accuracy and comprehensiveness of event identification are enhanced, local fine fluctuation in the region is fully considered, it is ensured that the generated abstract information of the real-time actuarial system can overview the change process of the ecological environment of the system, and the occurrence process of the ecological event can be effectively traced. The problem that in the prior art, due to the fact that local dynamic changes are difficult to capture through a traditional global analysis method, when abnormal events in a complex system are processed, the evolution process of the events is difficult to reconstruct systematically is solved.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing technology, and specifically to a method, system, device and storage medium for tracing the source of abnormal ecological value events, and in particular to a method, system, device and storage medium for tracing the source of abnormal ecological value events based on local-global. Background Art

[0002] In the context of multiple challenges facing the global environment today, how to evaluate and manage the market value of ecological products in real time has become the key to ecosystem research. Issues such as climate change, desertification, water shortages, and biodiversity loss pose serious threats to the sustainable development of ecosystems. In response to these threats, the real-time actuarial system for the market value of ecological products has emerged as a core tool for monitoring and maintaining the health of ecosystems. Through smart sensors, the Internet of Things (IoT), remote sensing and other technical means, the system can continuously and dynamically collect and calculate the market value data of ecological products, covering air quality, water quality, soil conditions, and multi-dimensional ecological resource change information such as agriculture, forestry, and animal husbandry. The advantages of the real-time actuarial system in quantifying the value of ecological products not only help promote the market circulation of ecological products, but also provide a scientific basis for the protection and utilization of environmental resources.

[0003] However, although the real-time actuarial system has made significant progress in calculating the value of ecological products, the existing system is still insufficient in identifying abnormal events and tracing their causes. Current ecological event analysis methods usually rely on global data trends, focusing mainly on macro-level environmental changes and resource utilization analysis, and often ignoring local subtle fluctuations within the region. Due to the complex interactions and feedback mechanisms within the ecosystem, local ecological events are often early signals of global changes. Ignoring these local events may lead to delayed anomaly detection and difficulty in taking timely response measures. Especially in scenarios with highly complex ecological environments, traditional global analysis methods often find it difficult to capture local dynamic changes, limiting the application effect of the system in early warning and cause tracing of ecological events.

[0004] In recent years, natural language processing (NLP) technology based on deep learning has made significant progress, especially in text processing and sequence modeling. As a pre-trained language model, the BERT (Bidirectional Encoder Representations from Transformers) model has been widely used in various tasks, which can extract contextual information from massive text data and capture complex semantic relationships. However, relying solely on the BERT model to process the semantic information of ecological log texts is still challenging, because the data of ecological events is not only text information, but also contains time series data and local environmental change information, which requires us to integrate and apply multiple technologies to achieve efficient anomaly detection and event tracing. Summary of the invention

[0005] In order to solve the problem in the prior art that traditional global analysis methods are difficult to capture local dynamic changes, resulting in difficulty in systematically reconstructing the evolution process of events when dealing with abnormal events in complex systems, the present invention provides a method, system, device and storage medium for tracing the source of abnormal events of ecological value.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: The present invention provides a method for tracing the source of abnormal ecological value events, comprising: Processing log text data to obtain local expressions of events; the log text data is ecological log text data collected by a real-time actuarial system based on ecological product value; According to the local expression of the event, the global expression of the event is obtained; Establish an event expression structure based on the local expression of the event and the global expression of the event; Use event expression structure to obtain summary information of real-time actuarial system; Based on the summary information of the real-time actuarial system, the source of abnormal ecological value events is traced.

[0007] Furthermore, the method of processing the log text data to obtain the local expression of the event is: Extract semantic information from log text data to obtain word expressions of events; According to the word expression of the event, the event information is captured and the local expression of the event is obtained.

[0008] Furthermore, the method of extracting semantic information from log text data to obtain word expressions of events is: Take the log text data as text input, use the BERT module to decompose the log text data to obtain several tokens; Map each word unit into the vector space, convert it into word embedding, and introduce segment embedding to obtain the embedding representation of each word unit; The embedding representation of each word unit is processed to obtain the word representation of the event.

[0009] Furthermore, the method of capturing event information and obtaining a local expression of an event according to the word expression of the event is: The word representation of the event is used as a sequence input, and the input gate is calculated to control the current information representation to obtain the candidate hidden state of the time step; According to the candidate hidden state of the time step, obtain the hidden state of the current time step; Get a local representation of the event based on the hidden state at the current time step.

[0010] Furthermore, the method for obtaining the global expression of the event based on the local expression of the event is: Input the local expression of the event into the GRU unit as a sequence to obtain the hidden state of the GRU unit at each time step; Obtain a global representation of the event based on the hidden state of the GRU unit at each time step.

[0011] Furthermore, the method for establishing the event expression structure according to the local expression of the event and the global expression of the event is: With the timeline as the key and the local expression of the event and the global expression of the event as the value part, the local expression of the event and the global expression of the event are stored respectively to establish an event expression structure.

[0012] Furthermore, the method for obtaining summary information of the real-time actuarial system by using the event expression structure is: The expressions of all events of the event expression structure are serially connected and input into the initialized LSTM unit to obtain the starting state for calculating the hidden state; According to the starting state of the hidden state, the context vector is obtained through the attention mechanism; The context vector is input into the LSTM unit to generate hidden states at different time steps, and the hidden states and context vectors at different time steps are input into the output projection layer to obtain the vocabulary distribution and generate summary information of the real-time actuarial system.

[0013] The present invention provides a method system for tracing the source of abnormal ecological value events, comprising: The local expression acquisition module of the event is used to process the log text data and obtain the local expression of the event; the log text data is the ecological log text data collected by the real-time actuarial system based on the ecological product value; Event global expression acquisition module: used to obtain the global expression of the event based on the local expression of the event; Event expression structure establishment module: used to establish event expression structure according to local expression of events and global expression of events; Summary information acquisition module of real-time actuarial system: used to obtain summary information of real-time actuarial system by using event expression structure; Abnormal event tracing module: used to trace the abnormal events of ecological value based on the summary information of the real-time actuarial system.

[0014] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0015] A computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for tracing the source of abnormal ecological value events. The method processes log text data, obtains local expressions and global expressions of events, and establishes an event expression structure; then, using the event expression structure, obtains summary information of a real-time actuarial system; finally, based on the summary information of the real-time actuarial system, the method traces the source of abnormal ecological value events. By extracting the local expression of events and the global expression of events, event information in log texts can be identified more accurately. The local expression of events focuses on the specific details of a single event, while the global expression of events provides a comprehensive understanding of the entire event environment. This dual recognition mechanism enhances the accuracy and comprehensiveness of event recognition, fully considers the local subtle fluctuations within the region, and constructs an event expression structure indexed by time by associating the local expression of events with the global expression of events. The combination of the time-attention mechanism and the local-global model ensures the deep integration of temporal relationships and semantic understanding, and ensures that the summary information of the generated real-time actuarial system can provide an overview of the change process of the system's ecological environment and effectively trace the occurrence process of ecological events. It not only helps to clearly present the correlation and causal relationship between events, but also provides strong support for subsequent event summaries and traceability work. Through the event expression structure, users can intuitively understand the development context and potential impact of events, and can efficiently detect event summaries related to ecological anomalies. This efficient detection mechanism enables users to quickly locate key information, providing an accurate and reliable basis for subsequent problem solving and decision-making, so as to achieve systematic reconstruction of the evolution process of events. This method not only improves the accuracy of event identification and the clarity of expression, but also improves the efficiency of abnormal event detection, achieves rapid tracing, and promotes the scientific nature of decision-making and the innovation and development of technological applications. It has important value in ecological environmental monitoring and other related fields.

[0017] Furthermore, the method detects the environment through the real-time actuarial system of the market value of ecological products, and inputs the ecological log data generated by recording the ecological environment status into the BERT module for processing, and then uses the selective reading unit (SRU) module to extract the semantic expression of key information, thereby generating a local expression of the event. The event vector sequence is input into the gated recurrent unit (GRU), and a new hidden state is constructed by mixing with the existing hidden state to capture the temporal dependency in the event sequence and the global association between events, thereby generating a global expression of the event. Based on the expression of the local and global expressions of the event, a storage structure is constructed with the event axis as the key and the value subdivided into the local expression of the event and the global expression of the event. The long short-term memory network (LSTM) unit is used and the event vector representation is concatenated, and the context vector is dynamically aggregated through the attention mechanism (including word level and event level) as input to provide an accurate starting state for the decoding process. By integrating the time-attention mechanism with the local-global model, making full use of time series information, and combining local and global information through a gating mechanism, the vocabulary distribution is finally generated through the output projection layer, ensuring that the summary information of the generated real-time actuarial system can provide an overview of the changes in the system's ecological environment and effectively trace the occurrence of ecological events.

[0018] Furthermore, in the present invention, the problem of tracing the root cause of anomalies is transformed into a timeline summary problem in the field of deep learning. The word-level embedding module based on BERT and SRU accurately captures the local feature information of the text, and constructs a GRU-based architecture to effectively extract feature information between events. At the same time, a double-layer attention of word level and event level is designed to ensure that the event-level attention plays a correct guiding role on the word-level attention. Finally, the local and global features are passed into the timeline summary generator to obtain the final text summary, ensuring that the generated summary is deeply retained and integrated in terms of temporal and global relationships, ensuring that the generated summary text is complete and detailed, and effectively tracing the occurrence process of the event.

[0019] The present invention also provides a method system for tracing the source of abnormal events of ecological value, including a local expression acquisition module for events, a global expression acquisition module for events, an event expression structure establishment module, a summary information acquisition module of a real-time actuary system, and an abnormal event tracing module; through the local expression acquisition module for events and the global expression acquisition module for events, the system can quickly extract key information from a large amount of log text data to form an event vector. This not only improves the speed of data processing, but also ensures the accuracy and completeness of information; the event expression structure establishment module can construct a clear and intuitive event expression structure based on the event vector. This structure helps to better understand and analyze the correlation and causal relationship between events, and provides a solid foundation for the subsequent abnormal event summary and tracing work; using the event expression structure, the summary information acquisition module of the real-time actuarial system can quickly extract event summaries related to ecological value anomalies. These summary information is concise and clear, which is convenient for users to quickly understand the overview and key information of abnormal events; the abnormal event tracing module is based on the summary information of the real-time actuarial system, using advanced algorithms and technologies, and can accurately trace the root events that cause ecological value anomalies. Users can quickly obtain key information and root events related to ecological value anomalies, so as to make more accurate and timely decisions, which helps to improve the stability and sustainability of the entire ecosystem, and is of great significance for timely discovery and resolution of problems and prevention. The system is not only applicable to the field of ecological environment monitoring, but can also be extended to other fields that need to process and analyze large amounts of log text data, such as finance, medical care, transportation, etc.

[0020] The present invention also provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program. The device has a simple structure, low modification cost, and small resource occupation.

[0021] A computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the above method when executed by a processor. The storage medium has good portability and strong versatility. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The figure is a flow chart of a method for tracing the source of abnormal events of ecological value according to the present invention.

[0023] Figure 2 The figure is a schematic diagram of the overall structure of a method for tracing the source of abnormal events of ecological value according to the present invention.

[0024] Figure 3 The present invention processes the log text data to obtain the overall structure of the local expression of the event Figure 4 This is a schematic diagram of the structure of a generator for obtaining summary information of a real-time actuarial system using an event expression structure according to the present invention.

[0025] Figure 5 It is a structural schematic diagram of an ecological value abnormal event tracing system of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] The present invention is further described in detail below in conjunction with specific embodiments, which are intended to explain the present invention rather than to limit it.

[0029] See also Figure 1 and Figure 2 The present invention discloses a method for tracing the source of abnormal ecological value events, comprising: S1: Processing the log text data to obtain the local expression of the event is to pass the ecological log text data into the event embedding module, obtain the word vector representation of the event text in the BERT module, and use the SRU selective reading unit module based on the word vector representation to extract the local expression with high semantic understanding ability; the log text data is collected based on the real-time actuarial system of ecological product value; the ecological log text data exists in the form of structured text, including timestamp, event description, environmental parameters and other information; see Figure 3 , specifically: S1.1: Extract semantic information from log text data to obtain word expressions of events. The method is as follows: S1.1.1: Take the log text data as text input and use the BERT module to decompose the log text data to obtain several tokens: Enter the log text data as First, the text Will be broken down into tokens one by one, Suppose the sequence after word segmentation is:

[0030] in, is the sequence length, For word element.

[0031] S1.1.2: Map each word unit into the vector space, convert it into word embedding, and introduce position embedding to obtain the embedding representation of each word unit: Each word Mapped into a fixed-size vector space and converted into word embedding , the BERT module introduces position embedding ,in Indicates the position of the word in the sequence, adds the word embedding and the position embedding to get the final embedding representation of each word , which can be expressed as:

[0032] S1.1.3: Process the embedding representation of each word to obtain the word representation of the event, that is, to obtain the final embedding representation of each word , processed by the n Transformer encoding layers of the BERT module, the final output is the word expression of the event.

[0033] S1.2: Based on the word expression of the event, capture the event information and obtain the local expression of the event. The method is: S1.2.1: Take the word representation of the event as a sequence input and calculate the input gate to control the current information representation to obtain the candidate hidden state of the time step: The word expression of the event is used as a sequence input, and the SRU module is used to control the amount of information of the current input by calculating the input gate, where the input gate For each time step The calculation can be expressed as:

[0034] in, is the time step Input, is the weight matrix of the input gate, is the bias term, is the sigmoid activation function.

[0035] The hidden state that has not been adjusted by the input gate, that is, the candidate hidden state The calculation of is as follows:

[0036] That is, time step Candidate hidden states of By entering With weight The product of plus the bias term After that, the hyperbolic tangent activation function is applied To calculate it.

[0037] S1.2.2: Based on the candidate hidden state of the time step, get the hidden state of the current time step: The SRU module uses the hidden state of the previous time step and the current candidate hidden state , and the input gate To update the hidden state of the current time step , the process is shown as follows:

[0038] Here, ⊙ represents element-wise multiplication.

[0039] S1.2.3: Get the local expression of the event based on the hidden state of the current time step: Hidden State The local event-level data vector is obtained through the linear transformation shown in the following formula:

[0040] in, is the weight matrix, is the bias term, the local expression of the event As the output of this module.

[0041] S2: Based on the local expression of the event, the global expression of the event is obtained. It is based on the local expression of the event, uses the GRU unit to capture the timing information, continuously mixes the input local expression sequence with the existing hidden state and constructs a new hidden state, thereby capturing the temporal dependency in the event sequence and the relationship between different events, and obtaining the global expression of the event. Specifically: S2.1: Input the local expression of the event as a sequence into the GRU unit to obtain the hidden state of the GRU unit at each time step: The local expression of the event arrive As a sequence input to the GRU unit, the GRU unit at each time step Receive the local expression of the current event and the hidden state at the previous time step , and then update the hidden state of the current time step ; The update process of the GRU unit includes two main gating mechanisms, namely the update gate and reset gate ; Among them, the update gate Control the previous hidden state In the current hidden state The amount of retention in the reset gate Determines how to combine the new input information with the previous hidden state. The process can be expressed as:

[0042] in, Represents the sigmoid activation function, which is used to control the strength of the gating signal; and For the current input The weight matrix of and is for the previous hidden state The weight matrix of and is the bias term.

[0043] Compute candidate hidden states , which is based on the reset gate Adjusted previous hidden state and current input The combination of is calculated as:

[0044] in, For the current input The weight matrix of is for the previous hidden state The weight matrix of is the bias term.

[0045] Using Update Gate to mix the hidden state of the previous time step and candidate hidden states , thus obtaining the hidden state of the current time step :

[0046] S2.2: Obtain a global representation of the event based on the hidden state of the GRU unit at each time step.

[0047] The hidden state of the last time step will be output As a global expression of events.

[0048] S3: According to the local expression of events and the global expression of events, an event expression structure is established. It is based on the local expression and global expression generated by the ecological log text data. An event expression structure with time as the index is established. The value part is subdivided into local expression and global expression. The event expression is stored with the time axis as the key, reflecting the semantic understanding and timing information of the event. Specifically: With the timeline as the key and the local expression of the event and the global expression of the event as the value part, the local expression of the event and the global expression of the event are stored respectively, and the event expression structure is established to reflect the semantic understanding and timing information of the event; S4: Using the event expression structure, we can obtain summary information of the real-time actuarial system by randomly initializing LSTM units and concatenating the local and global expressions of the event as input to provide an accurate starting state for the decoding process. We use the attention mechanism (including word level and event level) to dynamically aggregate context vectors, integrate the time-attention mechanism with the local-global model, combine local and global information through the gating mechanism, and finally generate the vocabulary distribution through the output projection layer to obtain the summary information of the real-time actuarial system. See Figure 4 , specifically: S4.1: Serially connect the expressions of all events in the event expression structure and input them into the initialized LSTM unit to obtain the starting state for calculating the hidden state: The event representation structure established by S3 is used as input, and all representations of the selected events are serially connected using the method of randomly initializing LSTM units. The output of this input is used as the decoder starting state, namely:

[0049] in, is the hidden state of the decoder starting state, is a random variable, For serial connection event expression; S4.2: Based on the starting state of the hidden state, the context vector is obtained through the attention mechanism: According to the classic attention mechanism, the input event expression structure is transformed into a context vector through dynamic aggregation:

[0050] in, It is The hidden state of the decoding step; A vector representation of the event at the previous time step; is the hidden state of the previous decoder state; Represents the representation of the jth word in the event; is the time-dependent weight matrix; To calculate the intermediate amount of word attention distribution; is the weight matrix for the previous time step in the word attention mechanism; is the weight matrix for the current time step in the word attention mechanism; is the attention time step of the event, The word attention time step of event i; for attention distribution; The weighted sum of the states is obtained by applying the obtained attention distribution as the context vector .

[0051] In the current model framework, the generated context vector is calculated The focus is mainly on the word-level attention mechanism, which leads to insufficient feature capture of event-level information. Since the model's understanding of the specific events currently being narrated is crucial in the task of summarizing time series data, this defect may cause information from different events to be mixed together incorrectly, resulting in a summary result that is inconsistent with the facts. Therefore, an event-level attention mechanism similar to the word-level attention mechanism is introduced in the module. , and use this mechanism to refine word-level attention to ensure the accuracy of the summary and the consistency of the information:

[0052] in, is the weight matrix for event attention distribution; is the hidden state for the previous time step The weight matrix of For global expression The weight matrix of is the global expression of the ith event, and the result is is the attention distribution of the ith event, ; is the intermediate variable of attention distribution of the i-th event; is the intermediate variable of the attention distribution of the jth event; Combine event-level attention distribution with word-level attention distribution:

[0053] in, is a mixed information body of word attention distribution and event attention distribution calculated in time step t, , Represent the position of the event and the position of the word in the event respectively; After adding event-level attention, the new context vector It can be expressed as follows:

[0054] in, is the context vector; S4.3: Input the context vector into the LSTM unit to generate hidden states at different time steps, and input the hidden states and context vectors at different time steps into the output projection layer to obtain the vocabulary distribution and generate summary information of the real-time actuarial system: The initial step involves exploiting the hidden state To process the key values ​​in the module, these key values ​​represent the time position embedding, which coincides with the time axis of the time series information. Therefore, the purpose of the method is to ensure that the model can fully mine and use the sequence information, and derive the measure of temporal attention by measuring the correlation between the position encoding and the current hidden state. , which is calculated as:

[0055] in, For position encoding The weight matrix, For location Positional encoding, is the hidden state at time step t; Then use temporal-attention to obtain local values ​​in the local-global module and global values The weighted sum of is shown as:

[0056] in, and It is a storage body that carries multi-level information and has different roles and responsibilities in the generation process. Convert to , and finally applied to the subsequent projection layer. The transformation process is as follows:

[0057] in, is the first fusion gate, Fusion Gate The weight matrix of Method configures local values ​​in the projection layer , given that it contains detailed information about the input content rather than overall features, every word moment in the generation process should play a decisive role. It affects the entire generation process by storing the global features of the event in different locations, from The global information of middle, is the weight matrix, the process is as follows:

[0058] in, For the fusion The hidden state of It is the second fusion gate; Finally, an output projection layer is applied to obtain the final generated distribution over the vocabulary ,

[0059] in, The weight matrix for the vocabulary output; is bias; is the corresponding bias; In summary, the model concatenates the output of the decoder LSTM, the word context vector, and the memory vector to form the input data stream of the output projection layer and generate summary information for the real-time actuarial system.

[0060] S5: Trace the source of abnormal ecological value events based on the summary information of the real-time actuarial system.

[0061] See also Figure 5 The present invention also provides a method system for tracing the source of abnormal ecological value events, including: The local expression acquisition module of the event is used to process the log text data and obtain the local expression of the event; the log text data is the ecological log text data collected by the real-time actuarial system based on the ecological product value; Event global expression acquisition module: used to obtain the global expression of the event based on the local expression of the event; Event expression structure establishment module: used to establish event expression structure according to local expression of events and global expression of events; Summary information acquisition module of real-time actuarial system: used to obtain summary information of real-time actuarial system by using event expression structure; Abnormal event tracing module: used to trace the abnormal events of ecological value based on the summary information of the real-time actuarial system.

[0062] Through the local expression acquisition module of events and the global expression acquisition module of events, the system can quickly extract key information from a large amount of log text data to form an event vector. This not only improves the speed of data processing, but also ensures the accuracy and completeness of information; the event expression structure establishment module can build a clear and intuitive event expression structure based on the event vector. This structure helps to better understand and analyze the correlation and causal relationship between events, and provides a solid foundation for the subsequent abnormal event summary and tracing work; using the event expression structure, the summary information acquisition module of the real-time actuarial system can quickly extract event summaries related to ecological value anomalies. These summary information is concise and clear, which is convenient for users to quickly understand the overview and key information of abnormal events; the abnormal event tracing module is based on the summary information of the real-time actuarial system, using advanced algorithms and technologies, and can accurately trace the root events that cause ecological value anomalies. Users can quickly obtain key information and root events related to ecological value anomalies, so as to make more accurate and timely decisions, which helps to improve the stability and sustainability of the entire ecosystem, and is of great significance for timely discovery and resolution of problems and prevention. This system is not only suitable for the field of ecological environment monitoring, but can also be extended to other fields that need to process and analyze large amounts of log text data, such as finance, medical care, transportation, etc.

[0063] The present invention provides a terminal device including: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0064] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0065] The terminal device may be a computing device such as a desktop computer, a notebook, a palmtop computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0066] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0067] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0068] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0069] In summary, the present invention provides a method, system, device and storage medium for tracing abnormal events of ecological value. The method detects the environment through the real-time actuarial system of the market value of ecological products, and inputs the ecological log text data generated by recording the ecological environment status into the BERT module for processing to obtain the word vector representation of the ecological log text; then, based on the word vector representation, the SRU (selective reading unit) module is used to extract the semantic expression of key information, thereby generating a local expression of the event. The local expression sequence of the event is input into the GRU unit, and a new hidden state is constructed by mixing with the existing hidden state, so as to capture the temporal dependency in the event sequence and the global correlation between events, and generate a global expression of the event. Based on the local expression and global expression of the event, a storage structure with the time axis as the key and the value subdivided into local expression and global expression is constructed. LSTM units are used and event representations are connected in series to provide an accurate starting state for the decoding process. Context vectors are dynamically aggregated through the attention mechanism (including word level and event level). Time series information is fully utilized by integrating the time-attention mechanism with the local-global model, and local and global information are combined through the gating mechanism. Finally, a vocabulary is generated through the output projection layer to obtain summary information of the real-time actuarial system to trace the abnormal events of ecological value. The present invention fully mines the semantic information of log text based on BERT and SRU modules, uses GRU units to retain the global relationship and temporal dependency of the time series, constructs an event representation structure indexed by the time axis, combines LSTM and attention mechanisms to ensure efficient reuse of event sequence information, and through the combination of time-attention mechanisms and local-global models, ensures the deep integration of temporal relationships and semantic understanding, and ensures that the generated summary can provide an overview of the changes in the system's ecological environment and effectively trace the occurrence of ecological events.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to impose any limitation on the technical solution of the present invention. Those skilled in the art should understand that, without departing from the spirit and principles of the present invention, the technical solution can also be subjected to several simple modifications and substitutions, and these modifications and substitutions are also within the scope of protection covered by the claims.

Claims

1. A method for tracing the source of abnormal ecological value events, characterized in that: include: Processing log text data to obtain local expressions of events; the log text data is ecological log text data collected by a real-time actuarial system based on ecological product value; According to the local expression of the event, the global expression of the event is obtained; Establish an event expression structure based on the local expression of the event and the global expression of the event; Use event expression structure to obtain summary information of real-time actuarial system; Based on the summary information of the real-time actuarial system, the source of abnormal ecological value events is traced.

2. The method for tracing the source of abnormal ecological value events according to claim 1 is characterized in that: The method for processing the log text data to obtain the local expression of the event is: Extract semantic information from log text data to obtain word expressions of events; According to the word expression of the event, the event information is captured and the local expression of the event is obtained.

3. The method for tracing the source of abnormal ecological value events according to claim 2 is characterized in that: The method of extracting semantic information from log text data to obtain word expressions of events is as follows: Take the log text data as text input, use the BERT module to decompose the log text data to obtain several tokens; Map each word unit into the vector space, convert it into word embedding, and introduce segment embedding to obtain the embedding representation of each word unit; The embedding representation of each word unit is processed to obtain the word representation of the event.

4. The method for tracing the source of abnormal ecological value events according to claim 2 is characterized in that: The method of capturing event information and obtaining a local expression of an event based on the word expression of the event is: The word representation of the event is used as a sequence input, and the input gate is calculated to control the current information representation to obtain the candidate hidden state of the time step; According to the candidate hidden state of the time step, obtain the hidden state of the current time step; Get a local representation of the event based on the hidden state at the current time step.

5. The method for tracing the source of abnormal ecological value events according to claim 1 is characterized in that: The method for obtaining the global expression of an event based on the local expression of the event is: Input the local expression of the event into the GRU unit as a sequence to obtain the hidden state of the GRU unit at each time step; Obtain a global representation of the event based on the hidden state of the GRU unit at each time step.

6. The method for tracing the source of abnormal ecological value events according to claim 1 is characterized in that: The method for establishing the event expression structure according to the local expression of the event and the global expression of the event is: With the timeline as the key and the local expression of the event and the global expression of the event as the value part, the local expression of the event and the global expression of the event are stored respectively to establish an event expression structure.

7. The method for tracing the source of abnormal ecological value events according to claim 1 is characterized in that: The method for obtaining summary information of the real-time actuarial system by using the event expression structure is: The expressions of all events of the event expression structure are serially connected and input into the initialized LSTM unit to obtain the starting state for calculating the hidden state; According to the starting state of the hidden state, the context vector is obtained through the attention mechanism; The context vector is input into the LSTM unit to generate hidden states at different time steps, and the hidden states and context vectors at different time steps are input into the output projection layer to obtain the vocabulary distribution and generate summary information of the real-time actuarial system.

8. A method system for tracing the source of abnormal ecological value events, characterized in that: include: The local expression acquisition module of the event is used to process the log text data and obtain the local expression of the event; the log text data is the ecological log text data collected by the real-time actuarial system based on the ecological product value; Event global expression acquisition module: used to obtain the global expression of the event based on the local expression of the event; Event expression structure establishment module: used to establish event expression structure according to local expression of events and global expression of events; Summary information acquisition module of real-time actuarial system: used to obtain summary information of real-time actuarial system by using event expression structure; Abnormal event tracing module: used to trace the abnormal events of ecological value based on the summary information of the real-time actuarial system.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images