Intelligent environmental protection large model desulfurization scene identification interaction method, system, medium and product

Abstract

The application provides a smart environmental protection large model desulfurization scene identification interaction method, system, medium and product, which comprises the following steps: firstly, the desulfurization role scene is matched through multi-modal instructions, and a graphical sand table and a global conversation state are initialized; after a diagnosis report generation instruction is received, multi-dimensional operation parameters of equipment associated with a role ID and a space-time context are obtained from a real-time database; a diagnosis result is obtained through analysis of a diagnosis large model and is mapped to a corresponding equipment graph element of the sand table; fault keywords are extracted to match a three-level differentiated solution scheme of quick solution, medium-term optimization and long-term upgrade; in response to a target scheme selection instruction, a quantitative evaluation result is generated through an efficiency estimation model combined with real-time working condition data, and finally, the result is visually presented to an interaction terminal. The whole process realizes a closed-loop intelligent interaction from intention recognition to decision support, and improves operation and maintenance efficiency and precision.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to intelligent environmental protection large-scale model desulfurization scene recognition and interaction methods, systems, media and products. Background Technology

[0002] In coal-fired power plants and heavy industry, flue gas desulfurization systems are a core component for achieving ultra-low emissions. With the advancement of smart environmental protection initiatives, operation and maintenance management is becoming increasingly complex, requiring frequent multi-dimensional interactions between maintenance personnel and the system. Therefore, quickly and accurately identifying users' operational intentions and establishing efficient and precise intelligent interaction paths between humans and the control system based on operational data has become crucial for improving the operational efficiency and safety of environmental protection facilities.

[0003] In existing technologies, the interaction and monitoring of desulfurization systems mainly rely on the DCS (Distributed Control System) or SIS (Plant-Level Monitoring Information System) screens in the central control room. Maintenance personnel navigate through complex menus and numerous configuration screens using a mouse and keyboard to find the corresponding equipment monitoring page. The system primarily presents its operating status in the form of data lists, trend curves, and fixed threshold alarms. When the system issues an alarm or maintenance personnel need to query specific issues, they typically rely on expert experience for analysis, followed by cost-benefit estimation of the handling solution using independent software, and finally, manual formulation of an operation and maintenance strategy.

[0004] However, in the aforementioned traditional technologies, because the system can only respond to mechanical operations and struggles to understand users' natural language commands, users must spend time searching for target functions within complex interface layers, resulting in lengthy interaction chains and fragmented user experiences. This passive interaction method disrupts the deep connection between "user intent" and "business scenario," making it difficult for the system to proactively adapt to users' operational needs. Therefore, existing desulfurization systems lack the ability to recognize user interaction intent and desulfurization operation and maintenance scenarios, making it difficult to achieve rapid scenario identification and intelligent decision support through natural interaction. Summary of the Invention

[0005] This application provides a smart environmental protection large-scale model for desulfurization scenario identification and interaction, a method, system, medium, and product for quickly locking and interacting with desulfurization operation and maintenance scenarios, thereby improving the operating efficiency and convenience of operation and maintenance interaction of desulfurization environmental protection facilities.

[0006] Firstly, this application provides a smart environmental protection large-scale model for desulfurization scenario identification and interaction. The method includes: determining the desulfurization role scenario from a role list based on multimodal instructions sent from an interactive terminal; initializing a graphical sandbox interface and a global session state including the current role ID and session lifecycle identifier based on a preset configuration; receiving a diagnostic report generation instruction for the desulfurization role scenario locked based on the global session state; obtaining multidimensional operating parameters of the equipment associated with the current role ID and spatiotemporal context from a real-time database; inputting the multidimensional operating parameters into a preset diagnostic large-scale model to determine the diagnostic result; and mapping the diagnostic result to... The system retrieves the corresponding device element node from the graphical sandbox interface; extracts fault keywords from the diagnostic results, and then matches these keywords with a pre-defined solution knowledge base to obtain a set of differentiated solutions at multiple levels, including quick resolution, mid-term optimization, and long-term upgrade; after displaying the differentiated solution set according to the identifier corresponding to the level, in response to the selection command of the target solution in the differentiated solution set, the system inputs the target solution and the real-time operating data associated with the graphical sandbox interface into a pre-defined benefit estimation model to obtain a benefit quantification assessment result; and finally sends the benefit quantification assessment result to the interactive terminal for visualization.

[0007] By adopting the above technical solution, the desulfurization role scenario is first locked based on multimodal commands, and the graphical sandbox and global session state are initialized to achieve precise binding between roles and scenarios, ensuring targeted interaction. Then, multi-dimensional operating parameters of related equipment are accurately acquired, and diagnostic results are output and mapped to sandbox elements in combination with the diagnostic model, making faults visible and intuitive. Multi-level differentiated solutions are matched through fault keywords to take into account different operation and maintenance needs. Finally, the target solution is quantitatively evaluated using a benefit estimation model, achieving a closed-loop interaction from intent recognition and fault diagnosis to solution selection and benefit evaluation. This solution breaks the mechanical nature of traditional interaction, connects "user intent - business scenario - decision support," shortens the interaction chain, improves the efficiency and accuracy of operation and maintenance interaction, and simultaneously achieves intelligent and quantitative operation and maintenance decision-making, reducing reliance on human experience and ensuring the stability of the desulfurization system.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the step of obtaining the multi-dimensional operating parameters of the device associated with the current role ID and spatiotemporal context from the real-time database specifically includes: parsing the time anchor point of the diagnostic report generation instruction, and calculating the lag time window in combination with the volume of the desulfurization system absorber tower and the current slurry circulation velocity to construct an asymmetric time-series sliding window; parsing the graph structure of the graphical sandbox interface, and determining the first-level strongly associated device nodes with the target device primitive involved in the diagnostic report generation instruction as the center; performing bidirectional topological traversal based on the flue gas medium flow direction and the slurry medium flow direction to determine the second-level weakly associated environment nodes; obtaining the time-series data slices of the first-level strongly associated device nodes and the second-level weakly associated environment nodes within the asymmetric time-series sliding window respectively, and performing spatiotemporal alignment and fusion to obtain the multi-dimensional operating parameters of the device.

[0009] By adopting the above technical solution, the time anchor point of the diagnostic command is first analyzed, and an asymmetric time-series sliding window is constructed by combining the absorption tower volume and slurry flow rate to accurately define the time range of valid data and avoid interference from invalid data. Then, using the target equipment primitives as the core, primary strongly correlated equipment and secondary weakly correlated environmental nodes are distinguished to clarify the priority and scope of data collection. Finally, the time-series data slices of the two types of nodes are spatiotemporally aligned and fused to ensure that the acquired multi-dimensional operating parameters are both targeted and complete. These technical features work together to solve the problems of ambiguous data collection range and redundant invalid data in traditional methods, providing high-quality data support for the accuracy of subsequent diagnostic results and improving the reliability of fault diagnosis.

[0010] In some embodiments of the first aspect, the step of constructing an asymmetric time-series sliding window by calculating the lag time window in conjunction with the volume of the desulfurization system absorber and the current slurry circulation velocity specifically includes: collecting the physical length and current flow velocity of the slurry pipeline, calculating the physical transmission time of the slurry from the dosing point to the absorber; subtracting the physical transmission time from the time anchor point to obtain the latest effective time, so as to exclude invalid data that has not yet been transmitted to the absorber; collecting the volume of the absorber slurry pool, and calculating the chemical reaction residence time of the slurry in the tower in conjunction with the current flow velocity; subtracting the chemical reaction residence time from the latest effective time to obtain the earliest effective time; and extracting the time period between the earliest effective time and the latest effective time to construct the asymmetric time-series sliding window.

[0011] By adopting the above technical solution, the physical transport time of the slurry from the dosing point to the absorption tower is first calculated. Subtracting the time anchor point yields the latest effective time, and invalid data that has not been transported to the absorption tower is eliminated. Then, combining the volume of the absorption tower slurry pool with the current flow rate, the residence time of the slurry chemical reaction is calculated to further determine the earliest effective time. Finally, an asymmetric time-series sliding window is constructed by extracting the time interval between the two time segments. Each technical step is progressive, precisely defining the effective data time interval relevant to the diagnostic needs. This avoids interference from prematurely collected unreacted data and delayed invalid data, while ensuring a high degree of correlation between the data within the window and the current diagnostic scenario. This improves the accuracy of data collection and lays the foundation for the accuracy of subsequent fault diagnosis.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting the time period from the earliest valid time to the latest valid time to construct the asymmetric time-series sliding window specifically includes: obtaining the historical load change curves on the first-level strongly correlated equipment nodes during the asymmetric time-series sliding window; calculating the fluctuation variance of the historical load change curves; if the fluctuation variance exceeds a preset steady-state threshold, then initiating a volume integral backtracking correction strategy, including reverse accumulation of the historical outflow slurry volume from the current time until the accumulated volume equals the volume of the absorber slurry pool, and updating the accumulation termination time to the earliest valid time; traversing the discrete operation events within the corrected asymmetric time-series sliding window, identifying the control command nodes that cause the transition of operating conditions; dividing the asymmetric time-series sliding window into multiple sub-context segments based on the control command nodes, and adding semantic tags representing operating condition attributes to each sub-context segment to generate equipment multi-dimensional operating parameters with time-series semantic enhancement.

[0013] By adopting the above technical solution, the historical load change curves of the first-level strongly correlated equipment within the window are first obtained, the fluctuation variance is calculated, and volume integral backtracking correction is initiated to ensure the accuracy of the earliest effective time and correct the time deviation caused by operating condition fluctuations. Then, the control command nodes for operating condition transitions are identified, the window is divided into multiple sub-context segments, and semantic tags are added to achieve semantic enhancement of time-series data. The synergistic effect of these technical features not only solves the time window deviation problem caused by operating condition fluctuations, but also allows multi-dimensional operating parameters to carry operating condition attribute information through semantic tags, making the data more contextually meaningful. This facilitates the accurate identification of fault characteristics under different operating conditions by large diagnostic models, further improving the accuracy and scenario adaptability of fault diagnosis.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the step of inputting the target scheme and the real-time operating data associated with the graphical sand table interface into a preset benefit estimation model to obtain a benefit quantification evaluation result specifically includes: based on a preset process reverse deduction algorithm, converting the target scheme into an instantiated parameter group containing specific execution action variables and adjustment step sizes; while keeping the current operating strategy unchanged, using the benefit estimation model to extrapolate the real-time operating data backward, generating an energy consumption trend curve within a preset future period as the first benefit trajectory; inputting the instantiated parameter group into a virtual copy of the graphical sand table interface, and while keeping the inlet operating conditions unchanged, using digital twin simulation to extrapolate the second benefit trajectory within a preset future period; performing an integral difference calculation on the first benefit trajectory and the second benefit trajectory, and using the accumulated difference obtained from the calculation as the benefit quantification evaluation result.

[0015] By adopting the above technical solution, the target solution is first transformed into an instantiated parameter set, clarifying the specific execution details of the solution and ensuring the accuracy of the simulation. Then, the first benefit trajectory under the current strategy is simulated through a benefit estimation model, and the second benefit trajectory corresponding to the target solution is simulated by combining a graphical sandbox virtual copy and digital twin simulation. Finally, the benefit quantification evaluation result is obtained through integral difference calculation. The interconnected technical links achieve precise quantification of the benefits of the operation and maintenance solution, solving the pain point of traditional solution benefit evaluation relying on experience and lacking quantification. This allows operation and maintenance personnel to intuitively compare the benefit differences of different solutions, providing a scientific and accurate decision-making basis for solution selection.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of using digital twin simulation to extrapolate the second benefit trajectory within a future preset period, the method further includes: obtaining the environmental emission hard indicator threshold under the desulfurization role scenario; detecting whether the process parameters corresponding to the second benefit trajectory exceed the environmental emission hard indicator threshold at any time; if they exceed, automatically triggering a downgrade correction or blocking strategy for the instantiated parameter group, and adding a compliance risk warning mark to the benefit quantification assessment result.

[0017] By adopting the above technical solution, the mandatory environmental emission thresholds for the desulfurization scenario are first obtained, clarifying the criteria for judging the compliance of the solution. Then, the process parameters corresponding to the second benefit trajectory are checked to see if they exceed the thresholds, accurately identifying compliance risks. Finally, when the thresholds are exceeded, parameter downgrade correction or blocking strategies are triggered, and a compliance risk warning marker is added. These technical features form a complete compliance control chain, preventing environmental violations caused by the execution of unqualified operation and maintenance solutions, enabling early warning and timely handling of risks, ensuring the environmental compliance of the desulfurization system's operation and maintenance process, and reducing the risk of penalties and equipment damage caused by environmental violations.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after inputting the multidimensional operating parameters into a preset diagnostic model to determine the diagnostic result and mapping the diagnostic result to the corresponding device element node in the graphical sandbox interface, the method further includes: parsing the fault source device node and the affected node in the diagnostic result; retrieving the physical pipeline path connecting the fault source device node and the affected node in the graph structure of the graphical sandbox interface; obtaining the real-time opening and closing feedback signals of all control valves and shut-off devices on the physical pipeline path; if the feedback signal of any shut-off device is in a fully closed state, or the real-time operating feedback signal of the fault source device node is in a stopped state, then the diagnostic result is determined to be a hallucination output, and a regeneration command for the diagnostic model is triggered.

[0019] By adopting the above technical solution, the fault source and affected nodes in the diagnostic results are first analyzed to clarify the fault propagation path; then, the physical pipeline path between them is retrieved to obtain real-time feedback signals from control valves and shut-off devices on the pipeline; finally, based on the feedback signals, it is determined whether the diagnostic result is a hallucination output, and a model regeneration command is triggered. These technical steps work together to construct a validity verification mechanism for the diagnostic results, effectively avoiding the problem of large diagnostic models outputting false fault results. This solves the pain points of traditional diagnostic models lacking self-verification and prone to misjudgment, ensuring the authenticity and reliability of the diagnostic results and providing accurate support for the formulation of subsequent operation and maintenance strategies.

[0020] In a second aspect, this application provides a scene recognition interaction system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the scene recognition interaction system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, this application provides a computer-readable storage medium including instructions that, when executed on a scene recognition interactive system, cause the scene recognition interactive system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, this application provides a computer program product, including a computer program that, when run on a scene recognition interactive system, causes the scene recognition interactive system to perform the method described in the first aspect and any possible implementation thereof.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0024] 1. By employing a full-process technical approach that includes multimodal command role-scene locking, graphical sandbox initialization, diagnostic large-scale model fault analysis, multi-level solution matching, and benefit estimation model quantitative evaluation, with a focus on multimodal intent recognition and end-to-end intelligent interactive design, this system effectively solves the technical problems of existing desulfurization systems, such as mechanical interaction, inability to understand user natural language commands, lengthy interaction links, fragmented experience, and lack of intent recognition and intelligent decision support. This achieves a closed-loop intelligent interaction for desulfurization operation and maintenance, from intent recognition to solution selection and benefit evaluation, shortening the interaction link, improving the accuracy and efficiency of operation and maintenance interaction, reducing reliance on human experience, and ensuring the stability of desulfurization system operation.

[0025] 2. By employing technical means such as calculating the physical transport time of slurry and the residence time of chemical reactions, accurately defining the effective time, and constructing an asymmetric time-series sliding window, and by emphasizing the precise calibration design of the time interval based on the characteristics of the desulfurization process, this technology effectively solves the technical problems of ambiguous data acquisition time range and redundant invalid data in existing technologies, which lead to insufficient correlation between data and diagnostic scenarios. This achieves the technical effect of accurately screening effective operating data that is highly relevant to diagnostic needs, eliminating invalid interference data, improving the accuracy of data acquisition, and laying a solid foundation for the accuracy of subsequent fault diagnosis results.

[0026] 3. By employing techniques such as target scheme instantiation parameter transformation, dual-benefit trajectory (current strategy and target scheme) derivation, and integral difference quantification calculation, and by emphasizing digital twin simulation and benefit quantification comparison design, this approach effectively solves the technical problem that existing desulfurization operation and maintenance scheme benefit assessments rely on expert experience and cannot be quantified, leading to a lack of scientific basis for scheme selection. This achieves accurate quantification of operation and maintenance scheme benefits, allowing operation and maintenance personnel to intuitively compare the benefit differences of different schemes, and providing scientific and accurate decision support for target scheme selection. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the intelligent environmental protection large-scale model desulfurization scene recognition and interaction method in the embodiments of this application;

[0028] Figure 2 This is another flowchart illustrating the intelligent environmental protection large-scale model desulfurization scene recognition and interaction method in the embodiments of this application;

[0029] Figure 3 This is a schematic diagram of the physical device structure of a scene recognition interaction system in the embodiments of this application. Detailed Implementation

[0030] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0031] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0032] For ease of understanding, the method provided in this implementation is described in process below. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating the intelligent environmental protection large-scale model desulfurization scene recognition and interaction method in this application embodiment.

[0033] S101. Determine the desulfurization role scenario from the role list based on the multimodal instructions sent by the interactive terminal, and initialize the graphical sandbox interface and the global session state including the current role ID and session lifecycle identifier based on the preset configuration.

[0034] Among them, the interactive terminal refers to various operating devices used by desulfurization operation and maintenance personnel to send instructions to the scene recognition interaction system and receive system feedback results, serving as the carrier for human-system interaction; multimodal instructions refer to a set of instructions sent by operation and maintenance personnel to the system, including multiple presentation formats, used to express the specific interaction needs of operation and maintenance personnel, including not only text instructions but also voice and interface click operations; the role list refers to a set of all environmental protection operation and maintenance related roles pre-stored in the scene recognition interaction system, used by the system to quickly match the identity and corresponding permissions of operation and maintenance personnel, with four core categories: desulfurization roles, denitrification roles, dust removal roles, and zero wastewater discharge roles, each with unique role information and a clear scope of permissions; the desulfurization role scenario refers to a dedicated interactive scenario highly adapted to the operation and maintenance work of the desulfurization role, determined by the system based on the desulfurization role identity selected by the operation and maintenance personnel and combined with preset rules, used to accurately match the actual work needs of the operation and maintenance personnel.

[0035] This step is the initial step in the scene recognition and interaction system's execution of all desulfurization-related interaction processes. It forms the foundation of the entire intelligent desulfurization interaction. The timing of its execution is the instant the maintenance personnel send their first desulfurization-related interaction command via the interaction terminal after the scene recognition and interaction system has started normally and completed initialization. Regardless of whether the command is for fault diagnosis, parameter query, solution acquisition, or other types of desulfurization interaction, the system will execute this step first to ensure that subsequent interaction processes are accurately matched to the maintenance personnel's desulfurization roles and needs. This step covers all maintenance interaction scenarios of the desulfurization system, including normal inspections of desulfurization equipment, fault diagnosis, parameter query, solution formulation, benefit evaluation, and other maintenance work. Whenever maintenance personnel initiate desulfurization-related interactions with the system, they must complete the desulfurization role scenario matching and basic interaction initialization through this step.

[0036] The specific execution process is as follows: First, the scene recognition interaction system monitors the command transmission status of all interactive terminals in real time. When a multimodal command sent by a certain interactive terminal is detected, the system immediately parses the multimodal command, extracts the core interaction requirements (clearly defined as desulfurization-related requirements) and relevant information of the sending terminal, and simultaneously retrieves the role list stored in the system. The system matches the identity information of the command sender (such as login account, terminal-bound role) with the desulfurization role information in the role list to quickly determine the desulfurization role corresponding to the maintenance personnel (selecting desulfurization-related roles from the four major environmental protection roles in the role list), and then determines the corresponding desulfurization role scenario to ensure that the scenario highly matches the desulfurization work requirements of the maintenance personnel. After determining the desulfurization role scenario, the system will, according to the preset configuration, retrieve the interface configuration, desulfurization equipment parameter display rules, and other information corresponding to the desulfurization role scenario, and automatically... The system automatically initializes the graphical sandbox interface to ensure that the content displayed (showing only desulfurization-related equipment and pipelines) and layout conform to the operating habits and focus of the desulfurization role. For example, the system initializes a sandbox interface with detailed desulfurization equipment parameters and fault analysis tools for the desulfurization technology expert scenario, and a sandbox interface with desulfurization inspection point markings and simplified diagrams of on-site desulfurization equipment for the desulfurization inspector scenario. At the same time, the system automatically generates a global session state, assigning a unique current role ID (consistent with the desulfurization role code matched with the maintenance personnel) and a session lifecycle identifier to the current interaction session. These two core pieces of information are stored in the global session state, and all subsequent desulfurization-related interaction steps will call the information in the global session state to lock the desulfurization interaction scenario, ensure session continuity, avoid information confusion between different roles and different sessions, and eliminate interference from the operation permissions of non-desulfurization roles.

[0037] This step effectively addresses the technical pain points of existing desulfurization systems, such as mechanical interaction and inability to adapt to the needs of different desulfurization operation and maintenance roles. By recognizing multimodal commands and accurately matching them with desulfurization role scenarios, it breaks the limitation of traditional systems that can only respond to mechanical operations, allowing the system to quickly understand the desulfurization-related interaction intentions of operation and maintenance personnel. By initializing the graphical sandbox interface and global session state, it establishes a basic interaction framework specifically for desulfurization between humans and the system, avoiding time-consuming searches for target functions by operation and maintenance personnel in complex interface layers that include non-desulfurization equipment, shortening the interaction chain. At the same time, it provides a unified scenario benchmark and session support for all subsequent desulfurization-related interaction steps, ensuring the continuity and accuracy of the interaction process, and laying the foundation for realizing intelligent desulfurization interaction.

[0038] S102. After receiving the diagnostic report generation instruction for the desulfurization role scenario based on the global session state lock, retrieve the device multi-dimensional operating parameters associated with the current role ID and spatiotemporal context from the real-time database.

[0039] Among them, global session state locking refers to the scenario recognition interaction system fixing the current desulfurization interaction scenario and session scope by calling the current role ID (desulfurization role) and session lifecycle identifier in the global session state.

[0040] This step is the core pre-step for the scene recognition interaction system to realize desulfurization fault diagnosis. It is executed when the system completes the desulfurization role scene matching, graphical sand table interface and global session state initialization in step S101, and successfully receives the diagnostic report generation instruction sent by the operation and maintenance personnel in the locked desulfurization role scene.

[0041] If no diagnostic report generation instruction is received, the system will maintain a global session state and wait for maintenance personnel to send other types of desulfurization-related instructions. The data acquisition operation in this step will not be executed.

[0042] The specific execution process is as follows: First, the scene recognition and interaction system continuously monitors the command input in the locked desulfurization role scenario. Based on the current role ID (desulfurization role) and session lifecycle identifier in the global session state, it filters out commands that are not within the current desulfurization role's permission scope or are not in the current session, and only responds to desulfurization-related commands sent by the current desulfurization role within this session. When a diagnostic report generation command is received from the maintenance personnel, the system first parses the command and extracts core information such as the diagnostic target (e.g., specific desulfurization equipment, specific desulfurization area), diagnostic time range, and diagnostic type. At the same time, combined with the current role ID in the global session state, it determines the range of desulfurization equipment data that the desulfurization role is authorized to obtain, avoiding unauthorized data access.

[0043] Subsequently, the system analyzes the spatiotemporal context corresponding to the diagnostic report generation instruction, clarifying the time and spatial dimensions of the desulfurization equipment data collection. The time dimension is determined by the time anchor point of the parsed instruction (i.e., the time when the instruction is sent or the diagnostic start time specified by the instruction), while the spatial dimension is determined by the location of the target desulfurization equipment and the scope of attention of the desulfurization role scenario.

[0044] In addition, the system further analyzes the time anchor point of the diagnostic report generation command and calculates the lag time window by combining the volume of the desulfurization system's absorber tower and the current slurry circulation velocity. An asymmetric time-series sliding window is then constructed to define the effective data range in the time dimension, avoiding the collection of invalid data that did not participate in the desulfurization reaction. Simultaneously, the system analyzes the graph structure of the graphical sandbox interface (containing only desulfurization equipment and pipelines). Centered on the target desulfurization equipment primitive involved in the diagnostic report generation command, it identifies first-level strongly correlated equipment nodes (i.e., desulfurization equipment directly associated with the target desulfurization equipment and significantly impacting its desulfurization operation, such as the circulating pump corresponding to the absorber tower). Then, based on the flue gas medium flow direction and the slurry medium flow direction, a bidirectional topological traversal is performed to determine second-level weakly correlated environmental nodes (i.e., those indirectly associated with the target desulfurization equipment). (1) Desulfurization auxiliary equipment or environmental monitoring equipment that has a certain impact on the operating conditions of the target equipment); then, the system sends a data query request to the real-time database, requesting to obtain time-series data slices of first-level strongly correlated equipment nodes and second-level weakly correlated environmental nodes within an asymmetric time-series sliding window. The real-time database responds quickly to the request, extracts the corresponding desulfurization equipment data and returns it to the system; finally, the system performs spatiotemporal alignment and fusion on the obtained time-series data slices to eliminate data deviations in the time and spatial dimensions (such as data acquisition time differences and location correlation deviations of different desulfurization equipment), and integrates them to form multi-dimensional operating parameters of the equipment that are highly related to the current role ID and spatiotemporal context, ensuring the integrity, accuracy and relevance of the data, and providing high-quality data support for subsequent desulfurization fault diagnosis.

[0045] This step effectively addresses the technical pain points of existing desulfurization system diagnostic data acquisition, such as ambiguous scope, redundant invalid data, and insufficient correlation between data and diagnostic scenarios. By locking desulfurization scenarios and role permissions based on global session state, it ensures the relevance and security of data acquisition and avoids collecting irrelevant data from non-desulfurization equipment. By constructing an asymmetric temporal sliding window and distinguishing between first-level strongly correlated and second-level weakly correlated nodes, it accurately filters effective desulfurization data and eliminates invalid and interfering data, avoiding the low diagnostic efficiency caused by blindly acquiring a large amount of irrelevant data in traditional technologies. At the same time, through spatiotemporal alignment and fusion, it ensures the integrity and correlation of desulfurization equipment data, providing solid data support for the accurate diagnosis of subsequent large-scale diagnostic models.

[0046] S103. Input the multi-dimensional operating parameters into the preset diagnostic model to determine the diagnostic results, and map the diagnostic results to the corresponding device element nodes in the graphical sandbox interface.

[0047] The pre-set diagnostic model refers to an artificial intelligence model pre-trained and deployed in the scene recognition and interaction system for diagnosing and analyzing the operating status of desulfurization system equipment. The model is trained based on a large amount of historical operating data of the desulfurization system, including desulfurization fault case data with diagnostic results, and desulfurization expert experience data. It possesses the ability to analyze and identify multi-dimensional operating parameters of desulfurization equipment, accurately determining the operating conditions, fault types, fault locations, fault causes, and fault impact range of the desulfurization equipment. For example, by analyzing the time-series data of the absorber slurry pH value and flue gas concentration data, the pre-set diagnostic model can determine whether the absorber has a low desulfurization efficiency fault and whether the fault is caused by abnormal slurry pH value or abnormal operation of the circulating pump. The diagnostic results refer to the complete set of analysis results output by the diagnostic model after analyzing and processing the input multi-dimensional operating parameters of the desulfurization equipment, providing a complete picture of the desulfurization system's operating status or desulfurization equipment faults. This is used to present the actual operating conditions and fault-related information of the desulfurization system to maintenance personnel, including operating condition assessment results, fault determination results, fault source equipment nodes, affected equipment nodes, fault level, and fault cause analysis.

[0048] This step is executed after successfully acquiring the multi-dimensional operating parameters of the equipment relevant to the current desulfurization diagnostic needs, without waiting for additional instructions from the maintenance personnel; if the S102 step fails to acquire valid multi-dimensional operating parameters of the equipment, the system will return a data acquisition failure message and will not execute this step.

[0049] The specific execution process of the steps is as follows: First, the scene recognition interaction system preprocesses the multi-dimensional operating parameters of the equipment obtained in step S102 to ensure that the data format meets the input requirements of the preset diagnostic model, avoiding invalid data from affecting the accuracy of the diagnostic results; then, the system inputs the preprocessed multi-dimensional operating parameters of the equipment into the preset diagnostic model. Based on the trained algorithm and desulfurization-related data, the diagnostic model comprehensively analyzes the input multi-dimensional data, and combines the process characteristics and fault occurrence patterns of the desulfurization system to determine the current operating condition (normal operating condition, abnormal operating condition) of the desulfurization system. If it is an abnormal operating condition, it further identifies the fault type and the fault source equipment. The system generates a complete diagnostic result by identifying the nodes, affected equipment nodes, fault level, and fault cause. The level of detail in the diagnostic result matches the desulfurization role permissions and scenario requirements corresponding to the current role ID. For example, the diagnostic result for the desulfurization technology expert scenario includes a detailed analysis of the fault mechanism, while the diagnostic result for the desulfurization inspector scenario only includes the fault location and basic fault information. After the diagnostic result is generated, the system will parse the result, extract all desulfurization equipment node information (fault source equipment, affected equipment, etc.), and match this equipment node information with the equipment element nodes in the graphical sandbox interface, establishing a one-to-one correspondence through equipment identifiers.

[0050] In addition, the system further analyzes the fault source equipment node and affected nodes in the diagnostic results. In the graph structure of the graphical sandbox interface, it retrieves the desulfurization physical pipeline path connecting the two and obtains the real-time opening and closing feedback signals of all control valves and shut-off devices on the pipeline. If any shut-off device feedback is in a fully closed state, or the real-time operation feedback signal of the fault source equipment node is in a stopped state, the diagnostic result is determined to be a phantom output, triggering the regeneration command of the diagnostic model to regenerate the diagnostic result. If the diagnostic result is valid, the system maps the fault information, operating condition information, etc. in the diagnostic result to the corresponding equipment element nodes through color marking, icon prompts, floating text, etc. For example, the fault source equipment element is marked in red, and the affected equipment element is marked in yellow. When the mouse hovers over it, a detailed fault cause and fault level are displayed. At the same time, the complete diagnostic result text is displayed in a designated area of ​​the graphical sandbox interface for easy viewing by operation and maintenance personnel.

[0051] This step effectively addresses the technical pain points of existing desulfurization systems, such as reliance on expert experience, low diagnostic efficiency, inaccurate results, and unintuitive presentation of diagnostic results. Through a pre-defined diagnostic model, it achieves intelligent and automated desulfurization fault diagnosis based on multi-dimensional equipment operating parameters, reducing reliance on human experience and improving diagnostic accuracy and efficiency. By mapping diagnostic results to equipment element nodes in a graphical sandbox interface, it achieves visualized presentation of diagnostic results, allowing maintenance personnel to quickly locate desulfurization faults and understand fault details. This avoids the tedious operation of searching for fault information in large data lists and configuration screens, further shortening the interaction chain and improving fault diagnosis efficiency. Simultaneously, by validating the diagnostic results, it avoids misjudgments caused by illusory outputs, ensuring the reliability of the diagnostic results.

[0052] S104. After extracting fault keywords from the diagnostic results, obtain a set of differentiated solutions at multiple levels from the preset solution knowledge base based on the fault keywords. These levels include quick solutions, mid-term optimizations, and long-term upgrades.

[0053] This step is executed after a valid diagnostic result is successfully generated and mapped to the graphical sand table interface; if the diagnostic result generated in step S103 is normal operating condition (no fault), the system will not execute this step, but will only indicate that the desulfurization operating condition is normal in the graphical sand table interface.

[0054] The specific execution process of this step is as follows: First, the scene recognition interaction system performs text parsing on the diagnostic results generated in step S103. It uses a keyword extraction algorithm to extract fault keywords that can characterize the core features of desulfurization faults from the diagnostic results. During the extraction process, it combines the professional terminology database of the desulfurization system to remove irrelevant words, ensuring the accuracy and relevance of the fault keywords. At the same time, it associates information such as fault level and fault source equipment type to enrich the search conditions.

[0055] Subsequently, the system uses the extracted fault keywords as the core search criteria, combined with information such as the current desulfurization role scenario, fault level, and desulfurization equipment operating conditions, to perform a search and matching in the preset solution knowledge base. It employs a combination of precise and fuzzy matching to filter out all solutions highly compatible with the current desulfurization fault. The system then categorizes the filtered solutions according to preset levels (quick resolution, mid-term optimization, long-term upgrade), forming a multi-level differentiated solution set. Each level of solution is labeled with its corresponding level identifier, implementation cycle, implementation difficulty, and expected desulfurization effect, facilitating differentiation and selection by maintenance personnel. Simultaneously, the system filters out solutions that the current role ID does not have permission to view or implement, based on the desulfurization role permissions, ensuring the compatibility and security of the solutions.

[0056] This step effectively addresses the technical pain points of existing desulfurization systems, such as reliance on expert experience for fault handling, cumbersome solution acquisition, lack of specificity and differentiation, and low solution matching efficiency. By extracting fault keywords and accurately matching them with a pre-set solution knowledge base, it achieves automated and rapid matching of desulfurization fault solutions, reducing reliance on human experience and improving the efficiency of solution acquisition. Through a multi-level set of differentiated solutions, it meets the needs of different desulfurization operation and maintenance scenarios and different fault handling priorities. Operation and maintenance personnel can select appropriate solutions based on the urgency of the fault and on-site implementation conditions, avoiding the problem of single solutions and inability to adapt to diverse needs in traditional technologies.

[0057] S105. After displaying the differentiated solution set according to the identifier corresponding to the level, in response to the selection instruction of the target solution in the differentiated solution set, the target solution and the real-time working condition data associated with the graphical sand table interface are input into the preset benefit estimation model to obtain the benefit quantification evaluation result.

[0058] The pre-set benefit estimation model refers to a model pre-trained and deployed in the scenario recognition and interaction system for quantitatively evaluating the benefits of desulfurization fault handling schemes. The model is trained based on the desulfurization system's process characteristics, equipment operating patterns, cost accounting rules, and benefit evaluation standards. It can combine the target scheme and real-time operating data to quantitatively evaluate the benefits after the scheme's implementation. For example, the benefit estimation model can calculate quantitative indicators such as the increase in desulfurization efficiency, energy consumption reduction, and cost savings after the scheme's implementation, based on the implementation steps of the target scheme and real-time operating data. The benefit quantification evaluation result refers to the set of evaluation results, including multiple quantitative indicators, output by the benefit estimation model after quantifying the target scheme. This is used to intuitively present the economic, environmental, and operational benefits after the target scheme's implementation. For example, the benefit quantification evaluation result includes quantitative indicators such as a 5% increase in desulfurization efficiency, a monthly energy saving of 1000 kWh, an implementation cost of 200 yuan, return on investment, and an increase in environmental compliance rate.

[0059] This step starts immediately after the differentiated solution set is displayed by hierarchical identifier and the target solution selection instruction sent by the operations and maintenance personnel is successfully received. If the operations and maintenance personnel do not send a selection instruction, the system will maintain the display state of the solution and wait for the selection operation of the operations and maintenance personnel, and will not perform the benefit estimation of this step.

[0060] First, the scene recognition and interaction system categorizes and sorts the differentiated solution set completed in step S104 according to the corresponding identifiers of each level, and displays it in the designated display area of ​​the graphical sandbox interface or on the interactive terminal interface. During the display process, it ensures that the hierarchical identifiers of each solution are clear and the core information is complete. At the same time, it supports operation and maintenance personnel to click to view the detailed implementation process of each solution, making it convenient for operation and maintenance personnel to select the target solution.

[0061] Subsequently, the system continuously monitors the selection commands from maintenance personnel. Upon receiving a selection command, it immediately parses the target solution identifier in the command, accurately locates the target solution selected by the maintenance personnel, and retrieves detailed implementation parameters, steps, and other information of the target solution. Next, the system obtains the real-time operating data of the current desulfurization system associated with the graphical sandbox interface. This data comes from a real-time database and is associated with the multi-dimensional operating parameters of the equipment obtained in step S102. The system focuses on extracting real-time operating data related to the implementation effect of the target solution (such as current desulfurization efficiency, energy consumption, environmental emission indicators, etc.) to ensure the timeliness and relevance of the data.

[0062] Subsequently, based on a preset process reverse engineering algorithm, the system transforms the target scheme into an instantiated parameter set containing specific execution action variables and adjustment step sizes (e.g., transforming "increase the frequency of the circulating pump inverter" into "adjust the inverter frequency to 50Hz, with an adjustment step size of 5Hz / minute"). Then, while maintaining the current operating strategy, the system uses a benefit estimation model to extrapolate real-time operating data backward, generating an energy consumption trend curve for a preset future period, serving as the first benefit trajectory (i.e., the baseline benefit trajectory without implementing the target scheme). Simultaneously, the instantiated parameter set is input into a virtual copy of the graphical sandbox interface. While maintaining the inlet operating conditions (e.g., flue gas inlet concentration, flow rate, etc.) unchanged, the system uses digital twin simulation to extrapolate the second benefit trajectory for a preset future period (i.e., the expected benefit trajectory after implementing the target scheme). Next, the system performs an integral difference calculation on the first and second benefit trajectories, using the accumulated difference as the benefit quantification assessment result. This quantification assessment result covers multiple dimensions, including economic benefits, environmental benefits, and operational benefits. Finally, the system organizes the benefit quantification assessment results to ensure clarity, intuitiveness, and ease of understanding for maintenance personnel.

[0063] This step effectively addresses the technical pain points of existing desulfurization system fault handling scheme benefit assessments, which rely on expert experience, lack quantification, and are inaccurate, leading to a lack of scientific basis for scheme selection. Through a pre-set benefit estimation model, combined with target schemes and real-time operating data, it achieves intelligent and quantitative assessment of scheme benefits, allowing maintenance personnel to intuitively understand the implementation effects of target schemes. Digital twin simulation and trajectory difference calculation improve the accuracy of benefit assessment, avoiding biases from human estimation. Simultaneously, the quantitative benefit assessment results provide maintenance personnel with a scientific and accurate reference for scheme selection, helping them choose the optimal fault handling scheme based on the assessment results, balancing environmental compliance, cost savings, and operational stability, further enhancing the intelligence level of desulfurization system operation and maintenance decision-making.

[0064] In some embodiments, after extrapolating the trajectory of the second benefit within a predetermined period through digital twin simulation, a pre-emptive environmental compliance control step can be added. This is specifically divided into three coherent and logically clear steps, detailed below:

[0065] First, the system obtains the threshold values ​​for mandatory environmental emission indicators. It first calls the global session state generated in step S101 above, extracts the current role ID and locked desulfurization role scenario information, clarifies the current desulfurization role and scenario requirements of the maintenance personnel, and then automatically accesses the internally preset environmental indicator database. This database is classified and archived according to environmental role scenario, equipment type, and process requirements. The system will filter out the threshold values ​​that are highly relevant to the current desulfurization scenario, target scheme, and system operating conditions to avoid obtaining irrelevant indicators. At the same time, it adjusts the scope of threshold usage based on desulfurization role permissions. After obtaining the threshold values, it is temporarily stored and associated with the global session state and the second benefit trajectory to provide core comparison basis for subsequent detection.

[0066] Second, process parameter detection: The system extracts process parameters for all time nodes within a future preset period from the temporarily stored second benefit trajectory, organizes them into a complete time-series parameter list, marks the time node corresponding to each parameter, starts the preset comparison algorithm, compares the process parameters of each time node with the corresponding environmental protection threshold point by point, and synchronously records details such as the time node exceeding the standard, the parameter exceeding the standard, and the extent of exceeding the standard, and finally draws a detection conclusion of compliance or non-compliance.

[0067] Thirdly, regarding compliance control operations, if the detection conclusion is non-compliant, the system immediately initiates control measures. First, a downgrade correction strategy is initiated, reducing the execution intensity of the instantiated parameter group. After optimization, the system is re-analyzed and tested. If multiple corrections are ineffective or the exceedance is significant, a blocking strategy is initiated to prohibit the execution of the solution. At the same time, regardless of which control measure is initiated, a compliance risk warning mark will be added to the subsequent benefit quantification evaluation results. All operation records are stored in the system's historical database to provide data support for subsequent review and solution optimization.

[0068] S106. Send the quantitative evaluation results of the benefits to the interactive terminal for visualization.

[0069] In this embodiment, the system can accurately match desulfurization role scenarios, initialize the graphical sandbox interface and global session state to build a dedicated interactive foundation; accurately acquire multi-dimensional operating parameters of desulfurization equipment to provide high-quality data support for diagnosis; realize intelligent and accurate diagnosis and visualization of desulfurization faults to avoid misjudgment due to illusory output; match multi-level differentiated solutions to adapt to different operation and maintenance needs; quantitatively evaluate the benefits of target solutions to assist scientific decision-making; and realize the visualization of benefit evaluation results to complete the interactive closed loop. Therefore, it can break through the limitations of traditional desulfurization system operation and maintenance, which rely on manual experience, have inefficient diagnosis, lack targeted solutions, and cannot quantify benefits, thereby realizing the intelligent and efficient operation and maintenance of desulfurization systems.

[0070] Following the above embodiments, the method provided in this embodiment will now be described in more detail. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating the intelligent environmental protection large-scale model desulfurization scene recognition and interaction method in this application embodiment.

[0071] S201. Collect the physical length and current flow rate of the slurry pipeline, and calculate the physical transmission time of the slurry from the dosing point to the absorption tower.

[0072] Among them, "slurry pipeline" refers to the conveying pipeline connecting the limestone slurry preparation system (i.e., the dosing point) and the slurry pool of the absorption tower, which is the physical channel for the desulfurizing agent to enter the reaction system; "physical transmission time" is used to indicate the pure physical movement time that the desulfurizing agent must undergo from the moment it is injected into the pipeline until it actually reaches the absorption tower to participate in the chemical reaction.

[0073] Specifically, this step typically occurs when the system receives a diagnostic report generation instruction and attempts to retrieve historical data related to the current fault. The scene recognition interaction system first accesses the underlying industrial control database or digital twin model library via a communication interface to read pre-stored geometric parameters (physical length) of the slurry supply pipeline. Simultaneously, the system collects the electromagnetic flow count value on the slurry supply pipeline in real time to obtain the current instantaneous flow velocity. Subsequently, based on fundamental fluid mechanics formulas, the system divides the physical length by the current flow velocity to calculate the physical transmission time. In actual operating conditions, to improve anti-interference capabilities, the system may perform short-time averaging filtering on the flow velocity.

[0074] This step quantifies the physical delay between control commands (such as increasing slurry supply) and system responses (such as an increase in pH within the tower). By calculating this time, the system can identify operational data that has been issued but has not yet reached the target, laying the foundation for subsequent removal of invalid data.

[0075] S202. Subtract the physical transmission time from the time anchor point to obtain the latest valid time, so as to exclude invalid data that has not yet been transmitted to the absorption tower;

[0076] Among them, "latest effective time" is used to indicate the cutoff time of all data that can have a causal impact on the current state of the absorber tower on the time axis; "invalid data" refers to the dosing operation data generated during the physical transmission time. This part of the slurry is still flowing in the pipeline at this time and has not entered the tower. Therefore, it has no physical causal relationship with the current desulfurization efficiency or pH value state in the tower.

[0077] Specifically, this step is executed immediately after the physical transmission time is calculated. The scene recognition interaction system performs a backward shift operation on the timeline, that is, subtracts the physical transmission time (e.g., 10 minutes) calculated in step S201 from the current time anchor point (e.g., 10:00) to obtain the latest effective time (09:50). The system marks all drug dosing operation data after 09:50 (i.e., between 09:50 and 10:00) as "future effective data" and excludes them.

[0078] This step prevents large models from incorrectly attributing actions that haven't yet occurred to the current fault. For example, if the system doesn't perform this step, the large model might generate incorrect diagnostic conclusions because a drug was added some time ago (and is still in the pipeline), and the pH value hasn't reached the appropriate level yet. By eliminating invalid data, the logical purity of the input data is ensured.

[0079] S203. Collect the volume of the slurry pool in the absorption tower and calculate the residence time of the slurry in the tower for chemical reaction based on the current flow rate;

[0080] The “absorber slurry pool volume” refers to the effective volume of slurry at the bottom of the absorber used to store the gas-liquid reaction slurry. This value is usually calculated from the height of the level gauge and the cross-sectional area of ​​the bottom of the tower. The “chemical reaction residence time” is used to represent the average time period during which a specific unit volume of slurry stays in the tower to participate in the reaction from the time it enters the absorber until its chemical activity is exhausted or it is discharged by the circulation pump.

[0081] Specifically, this step aims to determine the temporal depth of the data backtracking. The scene recognition interaction system first collects the liquid level count value of the absorption tower from the real-time database and calculates the current liquid volume (volume) by combining it with the geometry of the tower bottom. Next, the system collects the total flow rate of all operating slurry circulation pumps (if there is no flow meter, it can be estimated based on the pump's rated flow rate and current load). The system uses the formula: T=V / Q, where V is the volume and Q is the total flow rate, to calculate the theoretical replacement cycle of the slurry, i.e., the residence time of the chemical reaction.

[0082] This step constructs a timescale model of the desulfurization reaction. Desulfurization is a long-lag process, and current emission indicators are often determined by the accumulated slurry quality over the past few hours. Calculating the residence time is to determine how much data the large model needs to look forward to fully reconstruct the current chemical reaction.

[0083] S204. Subtract the chemical reaction residence time from the latest effective time to obtain the earliest effective time;

[0084] The "earliest effective moment" refers to the starting boundary point of the data required to construct the current diagnostic model on the time axis. Data prior to this moment means that the corresponding slurry or reaction products have already flowed out of the absorption tower through the gypsum discharge pump and no longer contribute significantly to the current system state.

[0085] Specifically, this step establishes the left boundary of the data window. After obtaining the latest valid time in step S202 (e.g., 09:50) and the chemical reaction retention time in step S203 (e.g., 2 hours), the scene recognition interaction system performs a subtraction operation to obtain the earliest valid time (07:50). This means that the system determines the properties of the slurry in the current absorption tower based on the cumulative operations and reactions within the two hours from 07:50 to 09:50.

[0086] This step enables precise cropping of massive amounts of historical data. By truncating outdated data, the computational load on large models is reduced, and noise interference from historical operating conditions (operations outside the current cycle) is eliminated, ensuring that the input to the model is a "full lifecycle" data segment strongly correlated with the current fault.

[0087] S205. Extract the time period between the earliest valid time and the latest valid time to construct the asymmetric time-series sliding window.

[0088] Specifically, this step is the final stage of data preparation. The scene recognition interaction system initiates a precise query request to the underlying time series database based on the calculated start and end times (from the earliest valid time to the latest valid time). The system packages and extracts all related variables within this time period (such as pH value, inlet SO2 concentration, slurry supply, pump current, etc.) to form a multi-dimensional time series tensor, which is the so-called "asymmetric time series sliding window".

[0089] Compared to the traditional approach of directly extracting data from the past N hours, this asymmetric window eliminates "ineffective future data" (right side) and "expired data" (left side), enabling large models to reason based on pure causal relationships. This significantly improves the accuracy of desulfurization scenario identification and fault diagnosis, and avoids logical illusions caused by time sequence misalignment.

[0090] In some embodiments, under non-steady-state conditions, there may be a window drift problem caused by the failure of the fixed flow rate assumption. In this case, the scene recognition interaction system first accesses the time series database and retrieves the load data sequence of the first-level strongly correlated equipment (such as coal-fired units or booster fans) within the initially determined window period. The system performs statistical analysis on the sequence and calculates its variance or standard deviation. If the calculated fluctuation variance is less than the preset steady-state threshold, it indicates that the system is in a stable operating condition, and the error of the initially determined window (calculated based on the average flow rate) is within an acceptable range. However, if the fluctuation variance exceeds the threshold, it indicates that the desulfurization system has just experienced load increases, decreases, starts, shutdowns, or large fluctuations. At this time, the slurry replacement rate in the absorption tower changes drastically, and a simple arithmetic average algorithm will lead to a serious time axis misalignment.

[0091] At this point, the system automatically triggers the volume integral backtracking correction strategy. This is a reverse derivation process based on the infinitesimal element method. The system uses the current time... Starting from the origin, perform a step-by-step scan in the historical direction (step size). (Typically set to minute-level). For each historical moment... The system collects the instantaneous outflow rate at that time. (Including gypsum discharge velocity, overflow velocity, etc.), calculate the outflow volume within this infinitesimal time period. Vi= × The system sets an accumulator variable ΣV, which is... The process begins by progressively accumulating these infinitesimal volumes, ΣV = Σ(ΔVi). This cycle continues until the accumulated value ΣV first numerically equals or just exceeds the current real-time slurry tank volume Vtank of the absorber. This corresponds to the time point... This is the corrected "earliest effective time." This time is physically defined as the absolute time when the earliest drop of the mixed slurry currently retained in the tower enters the system. This strategy can adaptively cope with varying operating conditions: under high load and high flow rate, the accumulation rate is fast, and the calculated time window will automatically shorten; under low load and low flow rate, the window will automatically extend, thus achieving a perfect alignment between "physical time" and "logical time."

[0092] After correcting the physical boundaries of the window, the system enters the semantic enhancement processing stage. The system traverses all discrete signal channels within this corrected, precise window, using edge detection algorithms to identify control command nodes. For example, if the system detects the "A circulation pump operation feedback" signal changing from 0 to 1, or the "pH setpoint" jumping from 5.2 to 5.4, these nodes are marked as key features of the operating condition transition. Based on these control command nodes, the system divides the continuous time window into multiple sub-situation segments. For example, segment 1 is the "low-load three-pump operation stage," segment 2 is the "dosing valve opening ramp-up transition stage," and segment 3 is the "high-load four-pump operation stage." The system generates corresponding semantic labels for each segment. Finally, the system performs tensor concatenation or metadata association between the original sensor numerical matrix (such as pH value and SO2 concentration) and these semantic labels to generate "multi-dimensional operating parameters of the equipment with temporal semantic enhancement."

[0093] The above embodiment eliminates time-series attribution errors under varying operating conditions. In power production, load fluctuations are common. Traditional algorithms based on fixed residence time introduce deviations of tens of minutes or even hours during drastic load changes, causing the model to learn incorrect input-output relationships (e.g., associating high emissions errors under high load with chemical dosing operations under low load). The integral backtracking strategy ensures, mechanistically, that the data window is always locked within the lifecycle of the "same batch of materials." Furthermore, it improves the interpretability of the diagnostic model's reasoning: by discretizing continuous data into segments with "context labels," the system is essentially telling the large model: "This data was generated during load increase, and that data was generated under steady-state conditions." This allows the large model to distinguish between parameter fluctuations caused by normal operating condition changes (such as sudden current changes caused by pump start-up) and genuine fault fluctuations (such as artificially high pH caused by slurry poisoning), thereby significantly reducing the false alarm rate and outputting diagnostic conclusions with clear engineering semantics, such as "response lag during load increase."

[0094] In this embodiment, an asymmetric time-series sliding window construction strategy based on the dual constraints of the physical transmission time of the slurry pipeline and the residence time of the chemical reaction in the absorption tower is adopted. This strategy can dynamically trim and lock time slices with real physical causal relationship with the current operating conditions according to real-time fluid dynamic parameters. Therefore, it can accurately remove future invalid data that has been generated but has not yet been transmitted to the reaction zone, as well as old historical data that has long been discharged from the system, during the data preprocessing stage. This achieves strict alignment between input features and physical reaction cycles in the time dimension, effectively solving the problems of data time sequence misalignment, causal inversion, and the resulting illusion of large model diagnostic logic caused by traditional fixed time windows in large-delay industrial process scenarios. This results in a significant improvement in fault diagnosis accuracy and model reasoning interpretability in complex and ever-changing operating environments.

[0095] The scene recognition and interaction system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the physical device structure of a scene recognition interaction system in this application embodiment.

[0096] It should be noted that, Figure 3 The structure of the scene recognition interaction system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0097] like Figure 3 As shown, the scene recognition interaction system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 302 or programs loaded from storage section 308 into Random Access Memory (RAM) 303, such as executing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0098] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0099] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the various functions defined in the present invention.

[0100] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0101] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0102] Specifically, the scene recognition and interaction system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the intelligent environmental protection large model desulfurization scene recognition and interaction method provided in the above embodiment.

[0103] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the scene recognition interaction system described in the above embodiments; or it may exist independently and not assembled into the scene recognition interaction system. The storage medium carries one or more computer programs, which, when executed by a processor of the scene recognition interaction system, enable the scene recognition interaction system to implement the intelligent environmental protection large-scale desulfurization scene recognition interaction method provided in the above embodiments.

[0104] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0105] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0106] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A smart environmental protection large-scale model desulfurization scenario recognition and interaction method, characterized in that, The method includes: Based on the multimodal instructions sent from the interactive terminal, the desulfurization role scenario is determined from the role list, and the graphical sandbox interface and global session state, including the current role ID and session lifecycle identifier, are initialized based on the preset configuration. After receiving the diagnostic report generation instruction under the desulfurization role scenario based on the global session state lock, the device multidimensional operation parameters associated with the current role ID and spatiotemporal context are obtained from the real-time database; The multidimensional operating parameters are input into a preset diagnostic model to determine the diagnostic results, and the diagnostic results are mapped to the corresponding device element nodes in the graphical sandbox interface. After extracting fault keywords from the diagnostic results, a set of differentiated solutions at multiple levels is obtained by matching the fault keywords from a preset solution knowledge base. The levels include quick resolution, mid-term optimization, and long-term upgrade. After displaying the differentiated solution set according to the identifiers corresponding to the levels, in response to the selection instruction of the target solution in the differentiated solution set, the target solution and the real-time operating data associated with the graphical sand table interface are input into the preset benefit estimation model to obtain the benefit quantification evaluation result. The quantitative evaluation results of the benefits are sent to the interactive terminal for visualization.

2. The method according to claim 1, characterized in that, The step of obtaining the device multidimensional operating parameters associated with the current role ID and spatiotemporal context from the real-time database specifically includes: The time anchor point of the diagnostic report generation instruction is analyzed, and the lag time window is calculated in combination with the volume of the desulfurization system absorber tower and the current slurry circulation velocity to construct an asymmetric time-series sliding window; The graph structure of the graphical sandbox interface is analyzed, and the first-level strongly related device nodes are determined with the target device primitives involved in the diagnostic report generation instruction as the center; A two-way topological traversal is performed based on the flow direction of flue gas and slurry to determine the secondary weakly associated environmental nodes. The time-series data slices of the first-level strongly correlated device nodes and the second-level weakly correlated environment nodes are obtained within the asymmetric time-series sliding window, and spatiotemporal alignment and fusion are performed to obtain the multidimensional operating parameters of the device.

3. The method according to claim 2, characterized in that, The steps for constructing an asymmetric time-series sliding window, which involve calculating the lag time window based on the volume of the desulfurization system's absorber tower and the current slurry circulation velocity, specifically include: Collect the physical length and current flow rate of the slurry pipeline, and calculate the physical transport time of the slurry from the dosing point to the absorption tower; Subtract the physical transmission time from the time anchor point to obtain the latest valid time, so as to exclude invalid data that has not yet been transmitted to the absorption tower; Collect the volume of the slurry tank in the absorption tower and calculate the residence time of the slurry in the tower for chemical reaction based on the current flow rate; Subtracting the chemical reaction retention time from the latest effective time, we obtain the earliest effective time. The time period between the earliest valid time and the latest valid time is extracted and constructed as the asymmetric time-series sliding window.

4. The method according to claim 3, characterized in that, After the step of extracting the time period from the earliest valid time to the latest valid time and constructing it as the asymmetric time-series sliding window, the method further includes: Obtain the historical load change curves on the first-level strongly correlated device nodes during the asymmetric time-series sliding window; Calculate the fluctuation variance of the historical load change curve. If the fluctuation variance exceeds the preset steady-state threshold, then activate the volume integral backtracking correction strategy, which includes reversing the accumulation of the historical outflow slurry volume from the current moment until the accumulated volume equals the volume of the absorber slurry tank, and update the accumulation termination time to the earliest valid time. Traverse the discrete operation events within the corrected asymmetric timing sliding window to identify the control command nodes that cause transitions in the operating conditions; Based on the control command node, the asymmetric time-series sliding window is divided into multiple sub-context segments, and semantic tags representing operating conditions are added to each sub-context segment to generate multi-dimensional operating parameters of the device with time-series semantic enhancement.

5. The method according to claim 1, characterized in that, The steps of inputting the target scheme and the real-time operating data associated with the graphical sand table interface into a preset benefit estimation model to obtain the benefit quantification evaluation results specifically include: Based on a preset reverse engineering algorithm, the target scheme is transformed into an instantiated parameter group containing specific execution action variables and adjustment step sizes; While keeping the current operating strategy unchanged, the energy consumption trend curve generated in the future preset period is used as the first benefit trajectory by extrapolating the real-time operating data through the benefit estimation model. The instantiated parameter set is input into a virtual copy of the graphical sandbox interface. Under the premise of keeping the entry conditions unchanged, the second benefit trajectory within a preset period is deduced through digital twin simulation. The first benefit trajectory and the second benefit trajectory are subjected to integral difference calculation, and the cumulative difference obtained by the calculation is used as the benefit quantification evaluation result.

6. The method according to claim 5, characterized in that, Following the step of using digital twin simulation to extrapolate the trajectory of the second benefit within a predetermined future period, the method further includes: Obtain the environmental emission thresholds for desulfurization scenarios; Detect whether the process parameters corresponding to the second benefit trajectory exceed the environmental emission mandatory threshold at any time; If the threshold is exceeded, a downgrade correction or blocking strategy for the instantiated parameter group will be automatically triggered, and a compliance risk warning mark will be added to the benefit quantification assessment result.

7. The method according to claim 1, characterized in that, After inputting the multidimensional operating parameters into a preset diagnostic model to determine the diagnostic results, and mapping the diagnostic results to the corresponding device element nodes in the graphical sandbox interface, the method further includes: Analyze the fault source device node and affected nodes in the diagnostic results; In the graph structure of the graphical sandbox interface, retrieve the physical pipeline path connecting the fault source device node and the affected node; Obtain real-time opening / closing feedback signals of all control valves and shut-off devices along the physical pipeline path; If the feedback signal of any cutoff device is in a fully off state, or the real-time operation feedback signal of the fault source device node is in a shutdown state, then the diagnostic result is determined to be a hallucination output, and a regeneration instruction for the diagnostic large model is triggered.

8. A scene recognition and interaction system, characterized in that, The scene recognition interaction system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the scene recognition interaction system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the scene recognition interaction system, the scene recognition interaction system performs the method as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is run on the scene recognition interaction system, it causes the scene recognition interaction system to perform the method as described in any one of claims 1-7.

Images