Coagulation solvent oil out-feed tank industry big data system

Abstract

The present application relates to the field of petrochemical tank monitoring and industrial big data processing, in particular to a condensed solvent oil out-tank industrial big data system, comprising a continuous state data acquisition terminal, a discrete event acquisition terminal, a data interaction server and a remote terminal; by acquiring the continuous spatial layered state time series data of the out-tank and the valve asynchronous discrete intervention operation log, time deviation analysis, time axis logical alignment and space-time semantic binding are carried out to generate continuous-discrete event coupling mapping data; further, state mutation features are extracted and feature evolution trajectory data is generated, and after comparison with reference trajectory data, it is determined whether the target fluid container has state abnormalities, and state abnormality alarm and trajectory reconstruction completeness data are sent to realize continuous tracking and early warning of the cause of solvent oil water value fluctuation.

Description

Technical Field

[0001] This invention relates to the field of petrochemical storage tank monitoring and industrial big data processing, specifically to an industrial big data system for condensed solvent oil delivery tanks. Background Technology

[0002] Condition monitoring of condensed solvent oil delivery tanks refers to the identification and analysis of the operational status during the delivery process based on the stratification changes of solvent oil and free water within the delivery tank and the on-site water-cutting operation. Current methods for monitoring the condition of delivery tanks include three types: monitoring based on single-point instrument data, analysis based on manual inspection records, and comprehensive judgment based on conventional production information systems. However, when monitoring the status of delivery tanks using existing technologies, on the one hand, the correlation between the spatial stratification of multiphase flow inside the tank and the discrete operation of valves is not accurate enough; on the other hand, the reliability of anomaly tracing and early warning judgment is reduced due to issues such as delays in manual log entry, heterogeneous data sources, and incomplete reference trajectories. Summary of the Invention

[0003] The purpose of this invention is to provide an industrial big data system for condensed solvent oil delivery tanks, solving the following technical problems: By unifying the continuous physical stratification process and the discrete manual intervention process under the same retrieval semantic framework for data modeling, the system can continuously track the causes of solvent oil-water value fluctuations and provide early warning of downstream refining stability. Furthermore, it can quantitatively express the credibility of the traceability results by splicing the evidence chain of abnormal evolution across databases, enabling the system to continuously revise the knowledge base as the equipment evolves to adapt to new operating conditions and new anomalies.

[0004] The objective of this invention can be achieved through the following technical solutions: The industrial big data system for solvent oil delivery tanks includes: a continuous state data acquisition terminal, a discrete event acquisition terminal, a data interaction server, and a remote terminal; The continuous state data acquisition terminal and the discrete event acquisition terminal respectively acquire the continuous spatial layered state time series data of the target fluid container, which serves as the condensed solvent oil delivery tank, and the asynchronous discrete intervention operation log for the valve of the container, and send them to the data interaction server. The data interaction server extracts the first timestamp of the time-series data and the second timestamp of the operation log to calculate the time deviation; if the deviation is greater than the preset alignment deviation threshold, the second timestamp is shifted and corrected to the closest first timestamp to perform time axis logical alignment and generate an alignment log; otherwise, the original log is retained. The data interaction server performs spatiotemporal semantic binding between the aligned or original log and the time-series data to generate continuous discrete event coupling mapping data. The state change rate of the mapped data is calculated based on a preset feature retrieval model and a time sliding window with a preset step size. Data that is greater than a preset mutation threshold is extracted as state mutation features and associated queries are performed to generate feature evolution trajectory data. The reference trajectory data pre-configured and stored locally or in the cloud is compared with the feature evolution trajectory data. If they are inconsistent, the target fluid container is determined to be in an abnormal state, and trajectory reconstruction completeness data is generated. An abnormal state alarm and the trajectory reconstruction completeness data are sent to the remote terminal, which then displays or broadcasts the abnormal state alarm and the trajectory reconstruction completeness data. If they are consistent, it is determined that no abnormal state has occurred.

[0005] In one possible implementation, the data interaction server is also used for: Count the number of successfully bound records in the continuous discrete event coupling mapping data; Obtain the total number of records of the continuous spatial hierarchical state time series data; If the total number of records is greater than zero, divide the number of successfully bound records by the total number of records to generate a continuous discrete event coupling mapping rate; if the total number of records is equal to zero, output an empty result marker indicating no valid continuous state records to the remote terminal.

[0006] In one possible implementation, the data interaction server, when performing correlation queries, is also used for: Record the start time of the associated query; Record the completion time for generating the feature evolution trajectory data; Calculate the difference between the completion time and the start time to generate spatial hierarchical feature retrieval delay data.

[0007] In one possible implementation, the system also includes a heterogeneous database; When generating the trajectory reconstruction completeness data, the data interaction server is also used for: Based on the heterogeneous database, node data representing state transitions are extracted from the feature evolution trajectory data; The node data is compared with the standard nodes in the preset complete evolutionary chain model; Calculate the number matching ratio between the node data and the standard nodes, use the number matching ratio as a confidence score, and generate the trajectory reconstruction completeness data.

[0008] In one possible implementation, the target fluid container includes a physical space with an extended discharge port and an anti-clogging shield; The continuous spatial layered state time series data includes multiphase flow spatial layered state data; The asynchronous discrete intervention operation log for the target fluid container valve includes a log of manual periodic dual-valve water cut-off operations.

[0009] In one possible implementation, the system also includes an expert terminal; The expert terminal is used to send access requests to the data interaction server. The data interaction server is further configured to: determine whether the expert terminal has access to the data interaction server, and, if the expert terminal has access to the data interaction server, process the access request of the expert terminal. Alternatively, if the expert terminal does not have access to the data interaction server, the expert terminal's access request may be denied.

[0010] In one possible implementation, the data interaction server is also used for: Extract the account information of the expert terminal from the access request of the expert terminal; Retrieve reference account information from the local or cloud storage space; The reference account information is compared with the account information of the expert terminal; In response to the inconsistency between the reference account information and the account information of the expert terminal, it is determined that the expert terminal does not have access to the data interaction server; or, In response to the fact that the reference account information matches the account information of the expert terminal, it is determined that the expert terminal has access to the data interaction server.

[0011] In one possible implementation, the expert terminal is also used for: Update the feature retrieval model and the reference trajectory data stored on the data interaction server.

[0012] In one possible implementation, the remote terminal is also used for: Receive the status anomaly alarm and the trajectory reconstruction completeness data, and perform visual display and voice broadcast of the status anomaly alarm and the trajectory reconstruction completeness data.

[0013] The beneficial effects of this invention are: 1. This invention addresses the problems of delayed data entry and inaccurate correlation representation in manual logs. By extracting the time deviation between time-series data and operation logs and performing translation correction, this invention achieves spatiotemporal semantic binding between discrete operations and continuous hierarchical data. This can accurately restore instantaneous valve intervention to the actual process segment it affects, avoid misjudging state changes as isolated events, and significantly improve the accuracy of the correlation representation between operations and fluid states. 2. This invention addresses the problem of difficulty in measuring the reliability of anomaly tracing. It generates a coupling mapping rate by statistically analyzing the number of successfully bound records in the coupling mapping data of continuous discrete events and combining it with the total number of records in the continuous state time series data. This quantitative indicator can intuitively reflect the proportion of continuous state changes that are effectively explained by discrete operations, helping process engineers to objectively assess the current data model's ability to support anomaly analysis and prevent overreaction. 3. This invention addresses the problem of incomplete tracing chains caused by heterogeneous data sources. It introduces heterogeneous databases to extract key nodes representing state transitions and compares them with a preset standard evolution chain to generate trajectory reconstruction completeness. This mechanism combines multi-source evidence such as real-time sampling and operation records under a unified benchmark, avoiding misjudgments caused by single data blind spots and significantly enhancing the reliability of early warning by utilizing a closed causal evidence chain. 4. This invention addresses the problem that single-point instruments cannot reflect the complex flow state inside the tank. This invention combines the spatial characteristics of the inner extension outlet and the anti-clogging cover to focus on monitoring the coordinated response of the multiphase flow spatial stratification state and the manual dual-valve water cutting operation. This enables the underlying data model to truly reflect the spatial differences between the material suction area and the settling water accumulation area, and effectively identify the risk of local fluid stagnation and free water approach under specific process structures. 5. To address the problem of incomplete reference trajectories that are prone to failure due to changes in operating conditions, this invention introduces a controlled expert terminal interaction mechanism. Authorized experts can dynamically update the feature retrieval model and reference trajectory data based on the actual flow evolution patterns on site, newly confirmed equipment changes, or abnormal patterns. As a result, the system has the ability to continuously iterate process knowledge, ensuring the accuracy of identifying new anomalies and fluctuations during long-term operation. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram of the modules of the industrial big data system for the condensed solvent oil delivery tank provided in the embodiments of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 The industrial big data system for solvent oil delivery tanks includes: a continuous state data acquisition terminal, a discrete event acquisition terminal, a data interaction server, and a remote terminal. The continuous state data acquisition terminal and the discrete event acquisition terminal acquire the continuous spatial layered state time series data of the target fluid container, which serves as the condensed solvent oil delivery tank, and the asynchronous discrete intervention operation logs for the valves of the container, respectively, and send them to the data interaction server. The data interaction server extracts the first timestamp of the time-series data and the second timestamp of the operation log to calculate the time deviation; if the deviation is greater than the preset alignment deviation threshold, the second timestamp is shifted and corrected to the closest first timestamp to perform time axis logical alignment and generate an aligned log; otherwise, the original log is retained. The data interaction server performs spatiotemporal semantic binding between aligned or original logs and time-series data to generate continuous discrete event coupled mapping data. The state change rate of the mapping data is calculated based on the preset feature retrieval model and the time sliding window with a preset step size. Data that is greater than the preset mutation threshold is extracted as state mutation features and associated queries are performed to generate feature evolution trajectory data. The reference trajectory data pre-configured and stored locally or in the cloud is compared with the feature evolution trajectory data. If they are inconsistent, the target fluid container is determined to be in an abnormal state, and trajectory reconstruction completeness data is generated. An abnormal state alarm and the trajectory reconstruction completeness data are sent to a remote terminal, which then displays or broadcasts the abnormal state alarm and the trajectory reconstruction completeness data. If they are consistent, it is determined that no abnormal state has occurred.

[0018] This embodiment provides the operation mechanism of the industrial big data system for the coagulation solvent oil delivery tank; specifically, the system is deployed in the coagulation post-processing section of the butadiene rubber production line to continuously track the spatial stratification changes of solvent oil and free water in the delivery tank, and to perform unified data modeling of the valve operation records performed by the inspection personnel and the stratification changes, so as to complete the source tracing and early warning before the water value fluctuation occurs in the downstream refining unit. The continuous status data acquisition terminal can be set at different height positions on the external delivery tank, and can also be connected to liquid level, interface, temperature, pressure, density or water content detection components; Since solvent oil and water will form layers with different properties from top to bottom after standing, and the local fluid residence zone, slow flow zone and settling zone do not completely overlap after the discharge structure inside the tank changes, this terminal prefers to collect time-series data according to spatial layers, rather than just collecting single point values. The data generated in this way not only reflects the current moisture content, but also reflects which layer the water is concentrated on, whether it is moving towards the discharge area, and whether there is local stagnation near the protective cover; Discrete event acquisition terminals can consist of operator handheld terminals, inspection station machines, valve position acquisition devices, or control room input interfaces. They are used to record discrete operations such as valve opening and closing, water cut-off start, water cut-off end, and abnormal reset. Since on-site manual operations usually involve memory delays, input delays, or situations where the operation is performed first and then the data is recorded later, the time of such logs may not strictly fall on the sensor sampling time. If this time difference is not addressed, subsequent systems may misinterpret a real and effective water-cutting action as an isolated event without any prior cause, or incorrectly attribute a slowly forming stratification process to another subsequent shift. After receiving the two types of data, the data interaction server first extracts the first timestamp from the continuous time series data and the second timestamp from the operation log, and performs time deviation analysis. The translation correction here refers to establishing a searchable logical correspondence on a unified time axis without changing the original record content. The server prefers to retain the original second timestamp and adds a logically aligned time field for subsequent retrieval, binding and trajectory reconstruction. Specifically, assuming the continuous sampling times are T1, T2, T3, and T4, and the discrete event recording time is E1, if E1 is between T2 and T3 and the deviation time is greater than the preset alignment deviation threshold, then the server logically attaches E1 to the nearest sampling time; if E1 is already within the preset allowable time deviation from a certain sampling point, then the original log remains unchanged. The engineering basis for this processing is that the liquid phase stratification change in the external delivery tank is continuous, while the valve action is triggered instantaneously, and the system needs to map the instantaneous action to the process segment it actually affects. Furthermore, the closest preferred time is determined according to the principle of minimizing the absolute time difference; when the absolute time difference between two adjacent first timestamps and second timestamps is the same or indistinguishable, the first timestamp that ranks earlier in the time series can be selected as the logical alignment point to maintain the process interpretation order of operation preceding subsequent state response; The above time deviation analysis and translation correction process establishes the following logical alignment model: Let the timestamp sequence of continuous state time series data be... ,in This represents the total number of timestamps in the time-series data. Represents the first in the sequence Each sampling timestamp ( The timestamps of discrete events recorded in the operation log are: The preset alignment deviation threshold is Calculate the minimum absolute time deviation. Let the closest sampling timestamp at this time be... The formula for calculating the time axis translation correction is: when At that time, the corrected aligned timestamp ;when At that time, keep the original log. .

[0019] Example operation: In actual monitoring of the delivery tank, the continuous state system is set to a sampling period of 10 seconds. , , The operator's handheld terminal records the timestamp of the water cut-off event. The system has a preset alignment deviation threshold. Seconds. Calculate the absolute deviation, distance. The deviation is 6 seconds, and the distance is... The deviation is 4 seconds. Minimum deviation This triggers a translation correction. for The system generates an alignment log and logically aligns and corrects the occurrence time of the water-cutting action to... This enables precise connection between instantaneous actions and continuous sampling windows.

[0020] After completing the timeline processing, the server performs spatiotemporal semantic binding; spatiotemporal semantic binding is to establish a relationship between a valve action and the changes in the water content of which layer before and after the action. For example, a water-cutting action can be coupled with several consecutive sampling segments before and after it into a unit. If the water content of the bottom layer decreases, the interface position falls back, and the discharge layer is stable in the unit, then the action is marked as an effective drainage intervention. If the valve action exists but the rate of change of the state characteristics of the bottom layer is lower than the preset response threshold, then it is marked as the rate of change of state characteristics not reaching the preset response standard or an invalid intervention. Specifically, the consecutive segments corresponding to three adjacent sampling windows are S1, S2, and S3, respectively. After alignment, an operation log event L1 is attached to S2, and the system forms a coupled record of [S1-S2 combined with L1-S3], and the semantic tags of the pre-operation state, operation point, and post-operation state are written in this record. S2 is the continuous state segment to which the operation point belongs, and S1 and S3 are the forward background segment and the backward response segment, respectively. If a longer window is used, it can also be extended to the same structure of [S0-S1-S2 combined with L1-S3-S4], but its naming convention still maintains the consistent expression of operation point segment + adjacent segments. Furthermore, during semantic binding, it is preferable to write at least the following fields to each coupling unit: target fluid container identifier, operation event identifier, logical alignment time, fragment start and end time, involved spatial layer, state response direction, and binding confidence label, so that the subsequent retrieval model can directly call them without having to repeatedly parse the original logs and original sampling records; Based on a preset feature retrieval model, the server extracts the state change rate and identifies state change characteristics by sliding window over time; the state change rate here physically corresponds to the migration speed of the oil-water separation interface, the accumulation speed of free water in the bottom layer, or the rate of change of the water content in the discharge neighborhood. If the rate of change in a certain sliding window is lower than the preset stability threshold, it indicates that the tank is still in a relatively stable settling state; if the rate of change in a certain sliding window is greater than the preset mutation threshold, it indicates that there is external operating condition disturbance, water cutting intervention, rapid liquid level change or local flow state change in the shield area; the server uses this type of mutation segment as the retrieval entry point, continues to trace back the formation conditions along the time chain and track the results, thereby generating feature evolution trajectory data; Here, the object of the state change rate calculation is preferably the change of state characterization value in the same spatial segment or the same semantic aggregation segment under adjacent time sliding windows. The state characterization value can be water content characteristic value, interface position value, density characteristic value, or a combination of characterization values ​​composed of multiple detection quantities. As long as the sliding window size and calculation caliber are kept consistent in the same retrieval process, the needs of subsequent mutation identification and trajectory association can be met. To quantify and identify characteristics of state abrupt changes, a state change rate calculation model based on a time sliding window is established: Let the width of the time sliding window be W, and the step size be S. The k-th sliding window contains N sampling points, and its set of state representation values ​​is... ,in This indicates the first [section] within the sliding window. State characterization value of each sampling point The average state characterization value of the sliding window The formula for calculating the rate of change of state of adjacent sliding windows is: ,in For the first The average state characterization value of each sliding window This represents the time interval between adjacent sliding windows.

[0021] Example calculation: Set the width of the sliding window. Minutes, sliding step Minutes are used to characterize the free water phase height at the bottom of the tank (unit: mm) obtained by the ultrasonic interface analyzer. Within a certain operating cycle, the average water phase height is calculated during the first sliding window (10:00-10:05). The average water phase height was calculated using the second sliding window (10:05-10:10). Time interval Minutes. Substitute into the formula to calculate the rate of change of state during this period. If the preset free water rise abrupt change threshold in the system is 15 mm / min, due to The system determines that a state change has occurred during this stage, automatically extracts the 5-minute data as a characteristic mutation fragment, and initiates subsequent expert tracing and comparison.

[0022] The reference trajectory data can be stored locally or in the cloud. The reference trajectory used to directly perform consistency / inconsistency determination with the current feature evolution trajectory is preferably the normal evolution reference trajectory that should appear under the current operating conditions, such as the baseline trajectory of normal stratification - scheduled water cut-off - stable external transmission. As for the abnormal trajectory templates such as stratified instability, delayed water shedding, and downstream water value fluctuation, they can be used as auxiliary data for the interpretation of abnormality types and subsequent expert review, without changing the basic logic of determining abnormalities based on whether they deviate from the normal reference trajectory in this embodiment; the server compares the currently generated feature evolution trajectory with the normal reference trajectory. If the current trajectory lacks the necessary buffer phase, fails to recover as expected after water cut, or anomaly nodes appear in spatial segments where they should not be, it is judged as an abnormal state; conversely, if the evolution sequence, key nodes, and direction of influence all fall within the allowable range of the reference trajectory, it is judged as a normal state. Furthermore, when multiple normal reference trajectories exist, one set can be selected based on the current liquid level range, production level, seasonal temperature, or equipment structural status, and then a consistency comparison can be performed to avoid misjudging trajectories that are actually reasonable under different operating conditions as abnormal. Consistency or inconsistency is preferably determined according to the following rules: at least four types of factors should be compared: node sequence, existence of key nodes, spatial segment corresponding to nodes, and direction of state change. When all of the above elements fall within the allowable range of the selected reference trajectory, it is judged as consistent; when any key element exceeds the allowable range, it is judged as inconsistent; this can avoid drawing abnormal conclusions directly based on a single numerical fluctuation. Furthermore, during the time deviation analysis phase, if an operation log is missing time information, has an abnormal time format, or cannot establish a proximity relationship with the continuous sampling period, the server will first mark the log as an event to be verified and will not directly participate in high confidence binding, but will still retain the original record for subsequent manual review. During the semantic binding phase, if there is instantaneous data loss or the data loss duration is less than the preset duration, adjacent valid segments can be used to construct a degraded coupling unit, and a low-confidence label can be added to the unit to avoid the interruption of the entire trajectory due to the lack of local acquisition. During the reference trajectory comparison phase, if the local reference trajectory is missing, the cloud reference trajectory can be called. If neither the local nor the cloud is available, the system can still output a warning of the existence of unexplained abrupt segments, but the integrity will be marked as pending confirmation to avoid outputting a single definitive judgment result. During the night shift operation of a butadiene rubber plant, the temperature of the upper oil layer in the solvent oil delivery tank remained stable, while the water content in the middle and lower layers gradually increased. The inspection personnel performed a water-cutting operation at 22:16, but the log entry time in the handheld terminal was 22:19. The server aligned the log with the continuous sampling points near 22:16 and bound multiple sampling windows before and after the water-cutting operation. The system detected that although the bottom layer dropped briefly, the adjacent layer continued to show abnormal water content increase in subsequent windows. This was inconsistent with the pattern of stable recovery of the discharge layer after water cut in the historical normal trajectory. Therefore, it was determined that there was a risk of abnormal stratification or insufficient water cut in the tank, and the abnormal alarm and trajectory reconstruction integrity data were sent to the remote terminal. The purpose of this step is to unify the continuous physical stratification process with the discrete manual intervention process under the same retrieval semantic framework, so as to achieve continuous tracking of the causes of solvent oil-water value fluctuations and early warning of downstream refining stability.

[0023] In a preferred embodiment of the present invention, the data interaction server is further configured to: count the number of successfully bound records of continuous discrete event coupling mapping data; obtain the total number of records of continuous spatial hierarchical state time series data; when the total number of records is greater than zero, divide the number of successfully bound records by the total number of records to generate the continuous discrete event coupling mapping rate; and when the total number of records is equal to zero, output an empty result marker indicating that there are no valid continuous state records to the remote terminal.

[0024] This embodiment provides a mechanism for generating the coupling mapping rate of continuous discrete events; specifically, after completing the spatiotemporal semantic binding of continuous sampled data and operation logs, the server further calculates the binding coverage to determine whether the current data model is sufficient to support reliable anomaly tracing. Specifically, a single anomaly identification result cannot characterize the confidence level of the anomaly association data chain; because when the operating conditions of the delivery tank are complex, the team records are inconsistent, or some sampling segments are interrupted for a short time, although the system can still output a trajectory, the trajectory may be based on a small number of valid associations. The server counts the number of successfully bound records and combines it with the total number of records in the continuous spatial hierarchical state time series data to generate a mapping rate. This mapping rate reflects the proportion of continuous discrete event coupled mapping records that can be used for retrieval in a continuous state segment. The number of successfully bound records here is preferably limited to the number of data units that have actually been written into the continuous discrete event coupling mapping data, that is, including data units that are coupled with a certain discrete operation log, as well as data units that are explicitly marked by the server as stable and static, without corresponding operations but interpretable. The reason why the latter type of data unit can be included is because it still undergoes semantic judgment and enters the coupling mapping result, rather than simply counting the unprocessed fragment directly as successful; in terms of statistical caliber, successful binding is preferably understood as the data unit that successfully enters the continuous discrete event coupling mapping data, which can be further subdivided into two categories: event binding unit and binding unit without operation background. The former indicates that a correspondence has been established with a specific discrete operation log, while the latter indicates that although no actual discrete operation was triggered, it has been interpreted by the system according to unified semantic rules and written into the continuous discrete event coupling mapping data. Therefore, it still belongs to the mapping success and not to the uninterpreted fragment. Specifically, if there are 10 consecutive state segments in a certain shift, denoted as S1 to S10, among which segments S2 and S3 are successfully coupled with the water cutting event L1, segment S6 is successfully coupled with the valve reset event L2, and segments S8 and S9, although not corresponding to manual actions, are determined by the model to be stable static segments and written into the continuous discrete event coupling mapping data with no operation background label, then these can all be included in the category of successful binding. If segments S4 and S5 cannot be explained due to missing logs, and segment S10 cannot be confirmed due to lost sampling, they will not be included. The server generates a mapping rate based on this to represent the completeness of data association in the current time period. To avoid confusion in statistical methods, the total number of records and the number of successfully bound records preferably use the same slicing granularity. For example, they are both counted based on continuous state segments formed under the same time window, instead of mixing the original second-level sampling points with window-level bound records. The specific calculation model for the Coupling Mapping Rate (CMR) of Continuous Discrete Events is as follows: The formula is: .

[0025] in, This represents the number of event binding units that have been successfully associated with specific discrete operation logs. The number of successfully bound units that conform to the continuous and stable law and are labeled by the model as being in a static state without operational background; This represents the total number of records of valid continuous spatial hierarchical state time series data extracted during this accounting period.

[0026] Operational example: The calculation is based on the night shift (22:00 to 06:00 the next day, a total of 8 hours) operation data of the butadiene rubber plant's external delivery tanks. The system is divided into continuous segments every 10 minutes for binding and calculation, with a total of [number missing] records. Segments. Among them, the number of segments successfully semantically bound to the water cut-out, drainage, and reset operations. The number of segments that conform to normal stratified settlement logic and are marked as static segments. The remaining 12 segments lacked semantic closure due to omissions in manual log entries. Substituting these into the formula yields the night shift coupling mapping rate. The 75% indicator, output with the analysis report, indicates that 25% of the current state segments are in an "unexplainable" state, assisting experts in assessing the strength of their conclusions regarding anomalies.

[0027] Furthermore, when the system needs to output more detailed auxiliary statistics, the number of event-bound units and the number of no-operation-background-bound units can be given separately without changing the definition of the continuous discrete event coupling mapping rate, so that process personnel can distinguish between the coverage of the intervention explanation and the coverage of the static state explanation. However, the numerator of the mapping rate written into the embodiment still uniformly adopts the caliber of the number of successfully bound records. Furthermore, the total number of records is preferably limited to the number of consecutive state segments that have undergone basic validity verification within the current mapping rate calculation period; segments with missing timestamps, unconfirmed segment boundaries, or invalid quality markings are not included in the denominator; this limitation ensures that both the numerator and denominator are within the same valid data range, avoiding distortion of the mapping rate due to invalid original segments being included in the total number of records; In engineering, this indicator is not simply a statistical value, but reflects the degree to which continuous processes on site are covered and explained by discrete management actions. A mapping rate greater than or equal to a preset mapping rate threshold means that the completeness of the knowledge chain between process behavior and state change meets the preset threshold. A mapping rate less than the preset mapping rate threshold often means that the team has insufficient supplementary data, unreasonable sampling layout, or temporary interventions that have not been institutionalized and recorded. These will reduce the interpretability of abnormal conclusions. Furthermore, when the total number of records is zero, the server does not perform ratio generation, but directly outputs an empty result marker indicating that there are no valid continuous state records in the current time period, to avoid meaningless results; if the number of successfully bound records is greater than the total number of records and exceeds the preset reasonable ratio range, it indicates that there is an anomaly in the underlying data deduplication or window segmentation, and the server can trigger a self-check and reorganize the fragment boundaries before recalculating. If some records can only establish low-confidence bindings, a hierarchical confidence label can be attached in addition to the mapping rate to prevent a single proportion from masking quality differences; Whether the aforementioned low-confidence bindings are included in the number of successfully bound records is preferably determined by a unified rule pre-set by the system: when such records have been written into the continuous discrete event coupling mapping data according to the degraded coupling unit, they can be included in the number of successfully bound records, and the low-confidence percentage will be output separately; when such records are only retained as fragments to be verified and have not been written into the coupling mapping data, they will not be included in the number of successfully bound records. In the aforementioned abnormal night shift scenario, the server sliced ​​the layered data from 22:00 to 24:00, forming a total of 24 continuous segments. Among them, 18 segments could be reliably interpreted with water cut-off, valve position changes, or stable static states, while the other 6 segments could not be fully bound due to delays in inspection records and partial communication interruptions. The system generates the continuous discrete event coupling mapping rate for that period and uses it as an auxiliary field in the anomaly report. When subsequent process personnel see an anomaly alarm accompanied by a mapping rate less than the preset mapping rate threshold, they can determine that the tracing result still needs to be verified on-site, rather than directly modifying all process parameters based on it. The purpose of this step is to quantify the degree of coupling between continuous state data and discrete intervention logs, thereby enabling an auxiliary assessment of the reliability of anomaly analysis results.

[0028] In a preferred embodiment of the present invention, when performing a correlation query, the data interaction server is further configured to: record the start time of starting the correlation query; record the completion time of generating feature evolution trajectory data; calculate the difference between the completion time and the start time, and generate spatial hierarchical feature retrieval delay data.

[0029] This embodiment provides a monitoring mechanism for spatial hierarchical feature retrieval latency; specifically, the server synchronously records the query start point and completion time during the feature evolution trajectory generation process to obtain retrieval latency data corresponding to the current query complexity; In the anomaly analysis of the external delivery tank, not all query tasks have the same complexity; if only short time segments before and after a single water cut are queried, the number of time segments involved in the retrieval path is less than the preset value; if it is necessary to span multiple shifts, trace the local stagnation near the protective cover, the water accumulation at the bottom of the tank, and then to the water value feedback of the downstream refining unit; If the time span of the query is greater than the preset time span threshold, the number of spatial segments involved is greater than the preset number of segments, or the number of data types included is greater than the preset number of types, the system will trigger a delay monitoring mechanism. At this time, recording the retrieval delay can help the system determine whether the current index structure is suitable for this composite retrieval task of continuous state + discrete events. Specifically, when the server receives a request to trace the upstream trajectory corresponding to a certain abnormal water value at the start time Q1, it first retrieves the layer segments A, B, and C, then associates the events L1 and L2, and finally generates the trajectory R1 at the completion time Q2. The time difference between time Q2 and Q1 constitutes the retrieval delay data. If another query only needs to obtain the recovery fragment after a single water cut, the path required to form trajectory R2 is shorter, and the corresponding delay data generated is less than the preset time threshold; by continuously recording these differences, the system can identify which query types are prone to causing processing backlog. The physical significance of this delay lies in the fact that spatial stratification itself has obvious hierarchical dependence, especially near the modified discharge area, where the local flow pattern cannot be fully expressed by ordinary linear time series and requires more refined segment association; therefore, this indicator can be regarded as an operational signal of whether the current database structure and index organization can support the actual process traceability task. In an abnormal execution scenario, if the query is manually aborted during execution, the node times out, or the downstream storage does not respond, the server marks the task as abnormally terminated and does not directly include it in the normal retrieval delay statistics to avoid polluting the performance assessment. If the task uses segmented queries to execute in parallel, the total completion time can be recorded, as well as the time taken for each subquery, for subsequent index optimization. If the completion time is earlier than the start time, it indicates that the system clock is drifting, and the time should be calibrated first and the record should be discarded. In the aforementioned night shift anomaly handling, the control room on-duty engineer initiated a full-chain tracing query of the anomaly evolution from 22:00 to 02:00 the next day; the server recorded the query start time and, after forming the evolution trajectory including four types of nodes such as bottom water accumulation and rise, water cutting action, material discharge section anomaly, and downstream water value fluctuation, recorded the completion time, thus obtaining the spatial layer feature retrieval delay data for this instance; if this data exceeds the preset delay threshold for multiple consecutive shifts, maintenance personnel can use this to determine that the existing indexing method is not conducive to rapid backtracking of complex process scenarios; The purpose of this step is to enable quantifiable monitoring of the related query process, thereby achieving an evaluation of database structure adaptability and anomaly tracing response efficiency.

[0030] In a preferred embodiment of the present invention, the system further includes a heterogeneous database; when generating trajectory reconstruction completeness data, the data interaction server is also used to: extract node data representing state transitions from the feature evolution trajectory data based on the heterogeneous database; compare the node data with standard nodes in the preset complete evolution chain model; calculate the number matching ratio between the node data and the standard nodes, and use the number matching ratio as a confidence score to generate trajectory reconstruction completeness data.

[0031] This embodiment provides a trajectory reconstruction completeness generation mechanism based on heterogeneous databases. Specifically, the system not only relies on real-time sampling and a single log library, but also calls heterogeneous data sources such as historical process library, equipment maintenance library, shift operation library, and quality test library to extract key nodes that can characterize state transitions and compare them with a preset complete evolution chain model. In actual factories, evidence of abnormalities in delivery tanks is not all stored in the same database; continuous layer data may be stored in a real-time database, valve actions may be stored in an operation log database, shift change instructions may be stored in an electronic shift handover system, and subsequent water value results from the refining unit may be located in a quality analysis database. If a single database is still used for retrieval, it often only yields traces that reflect anomalies but are not fully explained. By introducing heterogeneous databases, the server can piece together event nodes from different sources into a more complete causal chain according to a unified timeline and semantic relationships. Specifically, the standard chain of a certain abnormal evolution can be set as including the bottom water accumulation node N1, the interface upward movement node N2, the artificial water cutting node N3, the discharge layer recovery node N4 or the recovery failure node N4a, and the downstream water value response node N5; if the actual nodes obtained by the current retrieval are N1, N2, N3, and N5, it means that node N4 or N4a is missing, and the server can determine that the trajectory is not yet complete; If a node with abnormal water valve opening is extracted from the maintenance database, this node can also explain why normal recovery did not occur after node N3; the server matches the actual node with the standard node accordingly, and the more comprehensive the node coverage, the higher the generated integrity score. Furthermore, the above-mentioned quantity matching ratio is preferably calculated based on the set of standard nodes that should appear under the current working condition. That is, the standard nodes that should be applicable in this case are first selected from the complete evolution chain model, and then the proportion of the number of nodes covered by the actual nodes is calculated. For mutually exclusive branches like N4 and N4a, either one can be chosen as the node to be matched this time, rather than requiring both to appear simultaneously; for supplementary explanatory nodes such as abnormal water valve opening, they can be used to enhance the causal explanation, but in principle, they should not directly replace standard nodes, nor should they be counted repeatedly because of their repeated appearance, so as to avoid the completeness being artificially inflated by non-standard supplementary information. Furthermore, before node comparison, node standardization processing is preferably performed first, converting records in different databases that point to the same process fact into a unified node identifier, a unified time field, and a unified spatial segment field; For example, valve action records in the control log and valve action records of the same event in the manually supplemented log can be merged into the same node if they are close in time and the equipment objects are the same, and then participate in subsequent matching to prevent the same event from being counted repeatedly. In engineering terms, the completeness score here represents the degree to which the current anomaly explanation chain is supported by evidence, rather than simply indicating the severity of the anomaly. Severity depends on the process impact, while completeness depends on whether the traceability chain is closed. For example, if the downstream water value fluctuation is greater than the preset fluctuation threshold, but there is a lack of upstream water cut records and intermediate layer change nodes, the system should indicate that the anomaly has been observed, but the explanation chain is incomplete. Conversely, even if the abnormal amplitude is less than the preset fluctuation threshold, as long as the upstream and downstream nodes are fully closed, a trajectory reconstruction completeness greater than or equal to the preset mapping rate threshold can be obtained. To make the technical meaning of this score clearer, the quantity matching ratio is preferably limited to: the number of standard nodes that are actually successfully matched divided by the total number of standard nodes that should be matched under the current working condition. The number of standard nodes that are actually successfully matched is only counted for nodes that have completed deduplication and passed the timeline consistency check. Supplementary explanation nodes are not included in the numerator. After this processing, the trajectory reconstruction completeness data can directly reflect the coverage of the standard evidence chain. The above trajectory reconstruction completeness data (denoted as confidence score) The calculation model formula for ) is: .

[0032] Where M is the total number of standard causal nodes corresponding to the current working condition in the preset complete evolution chain model; To match the indicator function; Index number for standard nodes : When the When a standard node successfully matches (through timeline alignment and duplicate removal) the actual transition node data in the evolution trajectory of a heterogeneous database, ,otherwise .

[0033] Running example: The "solvent oil with water anomaly evolution chain" model currently invoked by the system is set to contain 5 standard nodes. The nodes are: Node 1 (accelerated rise of bottom-layer water), Node 2 (oil-water interface approaching the discharge port), Node 3 (sudden change in flow eddy current in the anti-clogging hood), Node 4 (failure to respond to water cutting action), and Node 5 (downstream refining system water level exceeding the standard alarm). In an actual anomaly alarm tracing, after the system was spliced ​​across databases, the real-time database supported Node 1, the ultrasonic density meter data supported Node 2, and the LIMS quality analysis system supported Node 5; however, due to log loss and the lack of sensors on the anti-clogging hood, the corresponding nodes for Node 3 and Node 4 were not properly configured. The value is 0. Then... Substituting into the formula yields the trajectory reconstruction completeness score. When broadcasting an alarm report, the remote terminal will include this 60% confidence level data, indicating to experts that the current chain of evidence is not closed and requires manual review.

[0034] Furthermore, if a heterogeneous database is temporarily inaccessible, the server can first generate a phased completeness based on the existing nodes and indicate the missing data source type in the results; If there are multiple process versions of the standard chain model, such as different delivery rhythms under different seasons and different production levels, the server should first select the corresponding version according to the current working conditions, and then perform node matching to avoid misjudging the completeness using an incorrect template. If the extracted nodes contain duplicate descriptions, such as the same valve action appearing in both the control log and the manual entry log, the server can remove duplicates before comparison to prevent artificially inflated scores. If some nodes can only confirm that an event has occurred but cannot confirm its exact time or spatial segment, then the node can be marked as a weak matching node for auxiliary explanation, but it is preferable not to include it in the number of successful matches of the standard node. In the aforementioned abnormal night shift scenario, the real-time database provides the time sequence nodes for the rise in water content in the bottom layer and the abnormality in the discharge layer, the operation log library provides the water cutting record, the quality library provides the result that the refined solvent oil and water value is greater than the preset delay threshold the next day, and the maintenance library adds maintenance work orders for the delayed action of the water cutting valve. When the server compares these nodes with the standard chain of accumulation-upward movement-water cutting-recovery or recovery failure-downstream response, if the existing evidence can directly show that the discharge layer has not recovered to stability after water cutting, then the node will be counted as N4a recovery failure and matched, instead of N4 recovery to normal being regarded as a node that should be matched. If the existing evidence can only prove that the downstream has responded, but is not enough to confirm whether the intermediate part is a normal recovery or a failed recovery, then N4 / N4a will be marked as the branch node is missing. Taking the former case as an example, the system can determine that N1, N2, N3, N4a and N5 have all obtained the corresponding evidence, thereby generating a trajectory reconstruction completeness that is greater than the preset scoring threshold and less than the preset maximum matching value. If it is the latter case, then generate a trajectory reconstruction completeness that is greater than or equal to the preset mapping rate threshold but not full value, and indicate in the result that the missing part is the evidence of the recovery branch, rather than generally stating that the recovery normal node is missing. The purpose of this step is to quantitatively express the credibility of the source tracing results by splicing the evidence chain of abnormal evolution across databases, so as to facilitate process personnel to judge whether to take immediate action or continue to supplement evidence.

[0035] In a preferred embodiment of the present invention, the target fluid container includes a physical space with an extended discharge port and an anti-clogging shield; the continuous spatial stratification state time series data includes multiphase flow spatial stratification state data; and the asynchronous discrete intervention operation log for the valves of the target fluid container includes a manual periodic dual-valve water cutting operation log.

[0036] This embodiment provides an industrial big data modeling mechanism for a specific external delivery tank physical structure; specifically, the target fluid container is a condensed solvent oil external delivery tank equipped with an extended discharge port and an anti-clogging shield, the continuous data collection object is the multiphase flow spatial stratification state, and the discrete event object is the manual periodic double valve water cutting operation. The external delivery tank is not a standard hollow container; after the discharge port extends into the tank, the actual material intake position is no longer close to the tank wall, but enters the tank a certain distance, which will change the spatial relationship between the material intake area and the bottom water accumulation area; the presence of the top or surrounding anti-clogging shield will further change the local flow velocity distribution and the range of sedimentation disturbance. As a result, a relatively safe zone suitable for stable oil phase delivery may form inside the tank, while a local stagnation zone where water accumulation is difficult to identify in a timely manner may also form. Therefore, continuous state data should not only describe the overall liquid level, but should describe the stratification state of multiphase flow in different spatial sections. Manual periodic dual-valve water shut-off is a typical asynchronous discrete intervention in this scenario; the dual-valve configuration is used to meet the preset sequence requirements of on-site safety isolation and discharge control; the periodic operation is configured as a non-closed-loop control operation triggered according to a preset shift cycle or manual input command; These types of operations have obvious human factors characteristics. Even if the actual physical effect occurs at the moment the valve is opened, the log recording may lag behind the action, or only record the start and not the end. For this reason, this scenario is particularly suitable for a data structure that couples continuous state with discrete intervention, rather than relying solely on ordinary alarm thresholds. Specifically, the tank space is simplified into an upper layer Z1, a middle layer Z2, a bottom layer Z3, and a discharge neighborhood Z4. Multiphase flow data records the status of the four areas at times t1, t2, and t3. During a certain shift, a dual-valve water cutting operation is performed, and the log records the first valve opening event D1, the second valve opening event D2, and the closing completion event D3. The server does not treat events D1 to D3 as isolated actions, but observes whether the bottom layer Z3 drops first and whether the discharge neighborhood Z4 subsequently recovers to stability. If only the bottom layer Z3 drops and the discharge neighborhood Z4 does not improve, it may indicate that although the water has been partially drained, the discharge neighborhood is still affected by disturbances. If the interval between events D1 and D2 is abnormal, it may also indicate that the operation rhythm does not meet the process requirements. In an abnormal handling scenario, if only the action of a single valve is recorded on site, while the log of the other valve is missing, the server can mark the water cut-off as an incomplete intervention and not directly evaluate it as a standard two-valve process. If the anti-clogging shield is disassembled, relocated, or its scaling condition changes, the system should allow the spatial topology configuration to be updated; otherwise, the old spatial semantic model may become invalid. If the multiphase flow stratification data can only cover part of the layers temporarily, the server can still perform coupling analysis on the covered layers, but this will reduce the ability to interpret the risks in the discharge neighborhood. After several days of continuous operation of the same production unit, free water gradually accumulates at the bottom of the external discharge tank. Because the discharge port extends into the tank, it can normally avoid the sedimentation area close to the bottom of the tank. However, when the water cutting interval is lengthened at night, the middle and lower interface begins to move upward and approach the discharge area. The operator performs double valve water cutting according to the procedure and records the log. Based on this, the system binds the multiphase flow changes of the bottom layer Z3 and the discharge neighborhood Z4 with the entire process of the dual valve operation to determine whether the water cut has truly restored the space state that can be safely delivered, rather than drawing conclusions based solely on a single valve opening. The purpose of this step is to align the data model with the actual tank topology and operating procedures, thereby achieving a more on-site semantic expression of the risks associated with multiphase flow stratification.

[0037] In a preferred embodiment of the present invention, the system further includes an expert terminal; the expert terminal is used to send an access request to the data interaction server; the data interaction server is further used to: determine whether the expert terminal has access rights to the data interaction server, and if the expert terminal has access rights to the data interaction server, process the access request of the expert terminal; or, if the expert terminal does not have access rights to the data interaction server, reject the access request of the expert terminal.

[0038] This embodiment provides an expert terminal access control mechanism; specifically, based on the abnormal trajectory and reference model already formed in the system, an expert terminal is introduced for process verification, model maintenance and trajectory interpretation, but the server only processes access requests from experts with the required permissions; In the aforementioned analysis of abnormalities in the delivery tanks, ordinary remote terminals mainly undertake the functions of receiving and displaying alarms, while expert terminals may further request to view the original trajectory across shifts, modify trajectory templates, supplement process descriptions, or verify explanations of special working conditions; therefore, the preset permission level corresponding to this terminal is higher than that of ordinary terminals. Without access control, ordinary users or unrelated devices may obtain the complete abnormal evolution chain, and even rewrite the subsequent judgment basis, affecting production decisions; Upon receiving an access request, the server first performs a permission determination. This determination can be based on account identity, terminal type, access source, time range, or authorization status. Only if the server confirms that the user is qualified to access the data will it open the interface for viewing or modifying historical data, model parameters, and reference data. If the user is not qualified, the server will directly reject the access request and simultaneously record the reason for rejection and the source of the request for subsequent auditing. Specifically, two requests were received at the same time. Request P1 came from the control room's general display terminal and only requested to view the current alarm page. Request P2 came from the process expert terminal and requested to retrieve the abnormal trajectory of the past three months and prepare to revise the reference template. At the access control layer, the server classifies request P1 as display permission and sends request P2 to the expert permission judgment process; if request P2 is authorized, processing continues; if not, the underlying model data is not returned. In an abnormal handling scenario, if the network environment of the expert terminal is abnormal, the certificate is invalid, or the request fields are incomplete, the server can first ask the request to be re-initiated, rather than directly entering high-privilege processing. If the system is in an emergency or abnormal period, in order to prevent accidental operation, the server can also restrict the expert terminal to only view and not modify; if the access permission service is temporarily unavailable, a conservative strategy will be adopted by default, only allowing remote alarm display and not opening the expert side model update function. Following the aforementioned night shift anomaly, the day shift process expert needs to retrieve the tank's continuous stratification trajectory and water cut-off log for the past week to determine whether the mistimed water cut-off by the dual valves was due to improper timing or recent changes in the flow regime near the shield causing the settling zone to expand. The expert initiates an access request through a dedicated terminal. The server first confirms the expert's permission status before granting access to the relevant trajectory page. If other unauthorized terminals attempt to view the same batch of historical trajectories at the same time, the server rejects their requests to prevent unauthorized modification or leakage. The purpose of this step is to separate model maintenance permissions from ordinary alarm viewing permissions, thereby enabling controlled access to the critical process knowledge base and abnormal trajectory data.

[0039] In a preferred embodiment of the present invention, the data interaction server is further configured to: extract the account information of the expert terminal from the access request of the expert terminal; obtain reference account information from local or cloud storage space; compare the reference account information with the account information of the expert terminal; determine that the expert terminal does not have access to the data interaction server in response to the inconsistency between the reference account information and the account information of the expert terminal; or determine that the expert terminal has access to the data interaction server in response to the consistency between the reference account information and the account information of the expert terminal.

[0040] This embodiment provides an expert access authentication mechanism based on account information comparison; specifically, the server reads account information from the expert terminal's access request and compares it with the reference account information stored locally or in the cloud to determine whether access is allowed. Identifying the device type alone is not enough to guarantee security; because the same device may be used by different people, and there may be situations where the device is borrowed, remotely logged in, or permissions are not revoked in a timely manner. Comparing account information with reference account information can further narrow down the legitimacy of the device to the legitimacy of the actual user; the reference account information may include account identifier, job category, authorization period, device to which it belongs, and range of accessible data, etc. Specifically, the reference account library contains account A and account B, where account A corresponds to a process expert and has the right to view and update the model, while account B only has the right to view. If an access request carries account A information and the server finds it matches the reference database, it is allowed to proceed to the expert processing flow. If another access request carries account C information, but it does not exist in the reference database or the fields are inconsistent, it is determined that there is no access permission. If account A exists but the authorization period has expired, it is also considered to be inconsistent. The practical significance of this account comparison mechanism is to ensure that the reference trajectory and feature retrieval model are maintained only by authorized process personnel. Since these models directly affect the anomaly interpretation path and early warning basis, once they are mistakenly modified, the system may misjudge the trajectory that should be considered abnormal as normal, or conversely amplify harmless fluctuations. In an abnormal handling scenario, if the local reference account database is temporarily inaccessible, the server can switch to the cloud reference account database for comparison; if both are unavailable, high-privilege access is denied by default, and only the ability to view ordinary alarms is retained. If the account information in the request is missing, such as the account identifier or signature verification field, the server will not perform fuzzy matching but will directly reject the request. If the same account sends requests from multiple sources simultaneously within a short period of time, the server can trigger secondary confirmation or audit flags to reduce the risk of impersonation. When the day shift expert reviews the aforementioned anomaly, their terminal sends an access request to the server, which includes the expert's account identifier and authorization information. The server first searches for the corresponding reference record in the local account database. If the local database is not updated, it then calls the cloud account database. After the comparison confirms that they match, the system allows them to view the complete evolution trajectory that caused the anomaly that night. If another person who has been transferred attempts to access the system with the old account, the system will refuse the high-privilege access because the reference account information is no longer consistent. The purpose of this step is to establish an auditable authorization entry point through account consistency verification, thereby enabling fine-grained control over expert access behavior.

[0041] In a preferred embodiment of the present invention, the expert terminal is also used to: update the feature retrieval model and reference trajectory data stored on the data interaction server.

[0042] This embodiment provides a mechanism for expert terminals to participate in the updating of models and reference trajectories; specifically, after expert access is authorized, the expert terminal can write back the newly formed process knowledge, corrected abnormal patterns and confirmed historical trajectories to the server for updating the feature retrieval model and reference trajectory data. Relying solely on the reference trajectory established during initial deployment is insufficient to adapt to changes in real-world operating conditions over the long term. As seasonal temperature changes, raw material fluctuations, equipment scaling, adjustments to the protective cover structure, or changes in shift operating habits occur, the boundaries between the original normal and abnormal trajectories may shift. If the model is not updated, the system will gradually develop two types of problems: one is misjudging newly emerging but actually reasonable trajectories as anomalies, and the other is lacking the ability to identify new anomalies; therefore, it is necessary to introduce an expert feedback mechanism, in which personnel familiar with the operating rules of the device make evidence-based revisions to the model. Specifically, the original reference trajectory library contains different types of reference trajectory templates, such as template R1, which shows the bottom layer descending and the discharge layer recovering after water cutting, and template R2, which shows the downstream water value fluctuating due to water cutting lag. When experts discover that after a certain equipment modification, a new buffer layer is formed near the shield, the normal recovery process will have an extra transition node between the water cutting action and the recovery to stability, denoted as buffer node N. Then the original template R1 can be updated to the new version template R1': that is, after the water cutting, the buffer node appears first, and then the stability is restored. Similarly, the feature retrieval model, which originally only focused on the correlation between underlying changes and water cutting events, can now add discharge neighborhood retention features as a new retrieval entry point after the update. The update here is not an arbitrary modification, but is based on the verified abnormal cases, stable operation cases and equipment change information; after the updated model is put back into operation, the system's recognition of similar scenarios in the future will be closer to the actual state of the current device. In an abnormal handling scenario, if the difference between the updated content submitted by the expert and the existing model exceeds a preset difference threshold, the server can first place it in the review area instead of directly replacing the online model; if insufficient key evidence is found during the update process, such as only having abnormal results for downstream water values ​​but lacking upstream stratification evidence, only trajectory notes can be added without performing adjustments to the main structure of the model; if the update write fails, the server should retain the original model to continue running and record the failed version to prevent online recognition from being interrupted. After several weeks of observation, process experts confirmed that the aforementioned night shift anomaly was not simply a delay in water cutting, but was related to a newly formed localized retention layer near the inner extension outlet. In the review interface, the experts added the anomaly trajectory as a new reference trajectory type and included the continuous increase in water content in the outlet neighborhood layer after water cutting into the feature retrieval model. Subsequently, when the system encounters similar scenarios, it can identify this type of evolution path in advance without waiting for the downstream water value to exceed the preset water phase safety parameter range. The purpose of this step is to enable the system to continuously update its knowledge base as the device evolves, thereby adapting to new operating conditions and new anomalies.

[0043] In a preferred embodiment of the present invention, the remote terminal is further configured to: receive status anomaly alarms and trajectory reconstruction integrity data, and to visualize and broadcast the status anomaly alarms and trajectory reconstruction integrity data via voice.

[0044] This embodiment provides a human-machine presentation mechanism for abnormal results; specifically, the remote terminal receives the status abnormality alarm and trajectory reconstruction integrity data sent by the server, and outputs them synchronously in a graphical interface and voice mode to serve the rapid handling in control room, inspection channel or mobile duty scenarios. In industrial sites, simply outputting anomaly alarms is often insufficient to support the handling; on-duty personnel also need to know which tank the anomaly is located in, which type of space layer the anomaly is related to, whether the trajectory evidence is sufficient, and whether it is appropriate to immediately perform water replenishment or arrange on-site re-inspection. Therefore, remote terminals should ideally display abnormal alarms together with the completeness of trajectory reconstruction; the visualization interface can use timelines, layered color bands, node links, or trend overlays to express the evolution process; and voice broadcasts are suitable for immediate reminders when on-duty personnel are not continuously monitoring the screen. For example, when the completeness is greater than or equal to the preset mapping rate threshold, the interface can emphasize that the completeness of the abnormal chain meets the preset threshold and recommends priority handling; when the completeness is in the preset low completeness range, it will indicate that abnormal signs have been found, but it is still recommended to conduct further review; thus, it can not only prompt the on-duty personnel about the abnormal status identified by the system, but also characterize the reliability of the explanation of the abnormal evolution path, thereby avoiding overreaction to low evidence results or slow response to high evidence results; Specifically, the objects displayed on the terminal page can consist of three parts: the first part is the alarm identifier A, the second part is the trajectory chain B, and the third part is the integrity label C; If alarm identifier A is the highest level alarm, trajectory chain B shows bottom water accumulation - water cut-off - recovery failure - downstream response, and integrity label C is high integrity, then the system will simultaneously trigger a voice broadcast prompting immediate review; if alarm identifier A is a yellow warning and integrity label C is low integrity, then the voice broadcast content will be more suggestive than mandatory. In an abnormal handling scenario, if the remote terminal screen malfunctions or the graphics component fails to load, the system will retain at least one of the following methods: voice broadcast or text push, to avoid alarm loss. If the terminal is in silent or unattended state, it can be forwarded to the duty mobile phone or central control broadcast; if the network is temporarily interrupted, the terminal can cache the latest alarm summary and retrieve detailed trajectory data after the network is restored; if the completeness data has not yet been generated, the interface will first display the abnormal alarm, and then place the status of trajectory evaluation in progress to prevent the alarm from being delayed due to waiting for the completeness data. In the aforementioned night shift anomaly, the server simultaneously sent alarm information to the control room screen and the shift leader's mobile terminal; the screen displayed the abnormal rise in water content in the lower layer of the delivery tank, insufficient recovery after water cut-off, and the downstream risk-related trajectory, along with the trajectory reconstruction completeness. The mobile terminal then broadcasts a voice message indicating that a certain delivery tank is suspected of being abnormal, and that the completeness of the evidence chain meets the preset threshold. The message requests a review of the water cutting effect and the discharge status. Based on this, the on-duty personnel can distinguish between normal fluctuations and conditions that require immediate on-site confirmation. The purpose of this step is to transform the anomaly judgments generated in the background into actionable information that can be directly understood and executed on-site, thereby achieving efficient communication and tiered response of alarm results.

[0045] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An agglomerated solvent oil out-feed tank industry big data system, characterized by, include: Continuous state data acquisition terminal, discrete event acquisition terminal, data interaction server, and remote terminal; The continuous state data acquisition terminal and the discrete event acquisition terminal respectively acquire the continuous spatial layered state time series data of the target fluid container, which serves as the condensed solvent oil delivery tank, and the asynchronous discrete intervention operation log for the valve of the container, and send them to the data interaction server. The data interaction server extracts the first timestamp of the time-series data and the second timestamp of the operation log to calculate the time deviation. If the deviation is greater than the preset alignment deviation threshold, the second timestamp will be shifted and corrected to the closest first timestamp to perform timeline logical alignment and generate an alignment log; otherwise, the original log will be retained. The data interaction server performs spatiotemporal semantic binding between the aligned or original log and the time-series data to generate continuous discrete event coupling mapping data. The state change rate of the mapped data is calculated based on a preset feature retrieval model and a time sliding window with a preset step size. Data that is greater than a preset mutation threshold is extracted as state mutation features and associated queries are performed to generate feature evolution trajectory data. The reference trajectory data pre-configured and stored locally or in the cloud is compared with the feature evolution trajectory data. If they are inconsistent, the target fluid container is determined to be in an abnormal state, and trajectory reconstruction completeness data is generated. An abnormal state alarm and the trajectory reconstruction completeness data are sent to the remote terminal, which then displays or broadcasts the abnormal state alarm and the trajectory reconstruction completeness data. If they are consistent, it is determined that no abnormal state has occurred. The system also includes a heterogeneous database; When generating the trajectory reconstruction completeness data, the data interaction server is also used for: Based on the heterogeneous database, node data representing state transitions are extracted from the feature evolution trajectory data; The node data is compared with the standard nodes in the preset complete evolutionary chain model; Calculate the number matching ratio between the node data and the standard nodes, use the number matching ratio as a confidence score, and generate the trajectory reconstruction completeness data.

2. The industrial big data system for condensed solvent oil delivery tanks according to claim 1, characterized in that, The data interaction server is also used for: Count the number of successfully bound records in the continuous discrete event coupling mapping data; Obtain the total number of records of the continuous spatial hierarchical state time series data; If the total number of records is greater than zero, divide the number of successfully bound records by the total number of records to generate a continuous discrete event coupling mapping rate; If the total number of records is zero, an empty result marker indicating that there are no valid continuous state records is output to the remote terminal.

3. The industrial big data system for condensed solvent oil delivery tanks according to claim 1, characterized in that, When performing correlation queries, the data interaction server is also used for: Record the start time of the associated query; Record the completion time for generating the feature evolution trajectory data; Calculate the difference between the completion time and the start time to generate spatial hierarchical feature retrieval delay data.

4. The industrial big data system for condensed solvent oil delivery tanks according to claim 1, characterized in that, The target fluid container includes a physical space with an extended discharge port and an anti-clogging shield; The continuous spatial layered state time series data includes multiphase flow spatial layered state data; The asynchronous discrete intervention operation log for the target fluid container valve includes a log of manual periodic dual-valve water cut-off operations.

5. The industrial big data system for condensed solvent oil delivery tanks according to claim 1, characterized in that, The system also includes an expert terminal; The expert terminal is used to send access requests to the data interaction server. The data interaction server is further configured to: determine whether the expert terminal has access to the data interaction server, and, if the expert terminal has access to the data interaction server, process the access request of the expert terminal. Alternatively, if the expert terminal does not have access to the data interaction server, the expert terminal's access request may be denied.

6. The industrial big data system for condensed solvent oil delivery tanks according to claim 5, characterized in that, The data interaction server is also used for: Extract the account information of the expert terminal from the access request of the expert terminal; Retrieve reference account information from the local or cloud storage space; The reference account information is compared with the account information of the expert terminal; In response to the inconsistency between the reference account information and the account information of the expert terminal, it is determined that the expert terminal does not have access to the data interaction server; or, In response to the fact that the reference account information matches the account information of the expert terminal, it is determined that the expert terminal has access to the data interaction server.

7. The industrial big data system for condensed solvent oil delivery tanks according to claim 5, characterized in that, The expert terminal is also used for: Update the feature retrieval model and the reference trajectory data stored on the data interaction server.

8. The industrial big data system for condensed solvent oil delivery tanks according to any one of claims 1 to 7, characterized in that, The remote terminal is also used for: Receive the status anomaly alarm and the trajectory reconstruction completeness data, and perform visual display and voice broadcast of the status anomaly alarm and the trajectory reconstruction completeness data.

Images