CN-122021917-A - Intelligent water affair infrastructure health monitoring system based on multi-mode cognitive reasoning

CN122021917ACN 122021917 ACN122021917 ACN 122021917ACN-122021917-A

Abstract

The invention discloses a health monitoring system of an intelligent water service infrastructure based on multi-mode cognitive reasoning, which comprises a data acquisition and conversion module, a multi-mode cognitive reasoning engine and a report generation module, wherein the data acquisition and conversion module is used for acquiring multi-source heterogeneous time sequence data of the water service infrastructure and converting the multi-source heterogeneous time sequence data into a dynamic visual map, the multi-mode cognitive reasoning engine triggers a long-period association analysis mechanism when receiving an abnormal signal, searches and integrates historical and contemporaneous context data associated with a current abnormal point based on a dynamic knowledge base of the water service field, generates an enhanced diagnosis map comprising a multi-dimensional comparison analysis view, invokes a large visual language model through a guided prompt project to analyze and infer visual semantics of the enhanced diagnosis map, completes abnormal diagnosis and root cause analysis, and the report generation module is used for outputting a structured natural language health evaluation report. The system can improve diagnosis accuracy, interpretability and long-term hidden danger discovery capability.

Inventors

ZHAO YUMENG
WANG RUNZHI
Zhao Zitang
ZHONG BAOYI
PAN XINYU
ZHANG HANQI
MA JUN

Assignees

哈尔滨工业大学

Dates

Publication Date: 20260512
Application Date: 20260210

Claims (9)

1. The intelligent water affair infrastructure health monitoring system based on the multi-mode cognitive reasoning is characterized by comprising a data acquisition and conversion module, a multi-mode cognitive reasoning engine and a report generation module, wherein: the data acquisition and conversion module is used for acquiring multi-source heterogeneous time sequence data of the water service infrastructure through a deployed sensor network of the Internet of things and converting the time sequence data into a standardized dynamic visual map; The multi-mode cognitive reasoning engine is connected with the data acquisition and conversion module and is used for receiving the dynamic visual map; The multi-modal cognitive reasoning engine is integrated with a large visual language model and a water affair field dynamic knowledge base, when an abnormal signal is received, a long-period association analysis mechanism is triggered, history and contemporaneous context data associated with a current abnormal point are retrieved and integrated based on the water affair field dynamic knowledge base, an enhanced diagnosis map containing a multi-dimensional comparison analysis view is generated, and a large visual language model is called through guided prompt engineering to perform visual semantic analysis and reasoning on the enhanced diagnosis map, so that abnormal diagnosis and root cause analysis are completed; The report generation module is connected with the multi-mode cognitive reasoning engine and is used for receiving the diagnosis reasoning result and outputting a structured natural language health assessment report containing visual basis, a logic reasoning chain and uncertainty quantification.
2. The intelligent water service infrastructure health monitoring system based on multi-modal cognitive reasoning as set forth in claim 1, wherein the internet of things sensor comprises a pressure sensor, a flow meter, a water quality monitor.
3. The multi-modal cognitive reasoning-based intelligent water infrastructure health monitoring system of claim 1 wherein the dynamic visualization map is at least one of a graph, a thermodynamic diagram, or a spatiotemporal map.
4. The intelligent water service infrastructure health monitoring system based on multi-mode cognitive reasoning of claim 1, wherein the long-period association analysis mechanism takes the time point at which the abnormality is detected as the center, forwards and backwards intercepts data of a preset time length to form a first analysis window, simultaneously searches contemporaneous data in the same season or working condition as the current abnormal time point in a historical database to form a second analysis window, converts the data of the first analysis window and the data of the second analysis window into a visual map together, and performs parallel or overlapped display to form a multi-dimensional comparison analysis view.
5. The intelligent water affair infrastructure health monitoring system based on multi-modal cognitive reasoning as set forth in claim 1, wherein the guided hint engineering includes system instruction hints for defining roles and tasks for large visual language models, contextual example hints for providing a few typical case guided models to understand water affair failure modes, and domain knowledge hints for extracting entities, relationships, and rules related to current anomalies from a water affair domain dynamic knowledge base and injecting into a model reasoning process.
6. The intelligent water service infrastructure health monitoring system based on multi-modal cognitive reasoning as claimed in claim 1, wherein the water service domain dynamic knowledge base stores topology information, equipment physical parameters, historical fault case base, maintenance rules and causal reasoning rules based on expert experience of the water service infrastructure.
7. The intelligent water service infrastructure health monitoring system based on multimodal cognitive reasoning as set forth in claim 1 wherein the health assessment report includes an abnormal event description, associated visual evidence identification, type or pattern of fault diagnosed, potential root cause inferred, recommended countermeasures or maintenance advice, and a diagnostic conclusion uncertainty score calculated based on model confidence or evidence sufficiency.
8. A method for intelligent water infrastructure health monitoring based on multimodal cognitive reasoning using the system of any of claims 1-7, characterized in that the method comprises the steps of: Step 1, continuously collecting multi-source heterogeneous time sequence data of a water service infrastructure through an Internet of things sensor network; step 2, converting the acquired time sequence data into a standardized dynamic visual map in real time or according to a preset period according to a predefined template; Step 3, performing preliminary screening on the dynamic visual map or the original data by using an anomaly detection unit, and identifying an anomaly signal; Step 4, triggering a multi-mode cognitive reasoning engine when an abnormal signal is identified, performing long-period association analysis, and retrieving and integrating relevant context data to generate a multi-dimensional comparison analysis view; Step 5, retrieving a domain knowledge segment related to the current abnormal scene from a water affair domain dynamic knowledge base, taking the domain knowledge segment as a part of a prompt, forming an enhanced diagnosis map together with the multi-dimensional comparison analysis view, and inputting the enhanced diagnosis map into a large-scale visual language model; step 6, driving a large visual language model to perform visual semantic analysis and reasoning on the enhanced diagnosis map through a guided prompt project, and completing abnormality diagnosis and root cause analysis; step 7, carrying out structural analysis and verification on the reasoning output of the large visual language model, extracting key diagnosis elements and carrying out consistency verification with rules in a knowledge base; and 8, generating and outputting a structured natural language health assessment report.
9. The method for monitoring the health of the intelligent water service infrastructure based on the multi-modal cognitive reasoning according to claim 8, wherein in the step 6, under the condition that the marked fault data is scarce or zero, the diagnosis of the abnormality of zero samples or few samples can be realized through guided prompt engineering and context example prompt.

Description

Intelligent water affair infrastructure health monitoring system based on multi-mode cognitive reasoning Technical Field The invention belongs to the technical field of intelligent water affair and infrastructure health monitoring, relates to an intelligent water affair health monitoring system, and in particular relates to a water affair infrastructure intelligent health monitoring system which integrates Internet of things, visual analysis and large-scale visual language model to perform multi-mode cognitive reasoning. Background The urban water supply network is used as a key infrastructure, and the safe and stable operation of the urban water supply network is crucial. However, the leakage and pipe explosion risks are increased due to the factors of ageing, load increase, extreme climate and the like of the pipe network, and huge water resource loss and economic cost are caused. Currently, with the wide application of industrial internet of things technology in smart water, sensors deployed in a pipe network can continuously generate massive pressure, flow and other monitoring data. The current mainstream health monitoring and abnormality diagnosis method mainly depends on two technologies, namely a method based on a statistical threshold or rule, wherein the method is simple and direct, has poor adaptability, cannot identify complex and weak early failure modes, and has high false alarm rate and false missing report rate. And secondly, a method based on deep learning, such as modeling a sensor sequence by using a cyclic neural network or a time sequence convolution network. Although the method can capture complex modes, the method has the remarkable limitation that firstly, the model training extremely depends on a large amount of historical fault data accurately marked by experts, and the high-quality marked data in an actual water service scene is scarce and has high acquisition cost, so that the application and generalization capability of the model are severely restricted. Second, deep learning models are often considered as "black boxes" with opaque internal decision logic, and lack of interpretability of the diagnostic results, resulting in difficult understanding and output of the trust model by the operator, preventing adoption in actual decisions. More importantly, most of the existing methods focus on reacting to transient or short-term data anomalies, lacking efficient analysis of long-period, trending correlations. For example, a small continuous drop in pressure may be an early sign of a covert leak, but it is difficult to detect single point data from an isolated analysis. The prior art fails to systematically integrate data of months before and after the current outlier and historical contemporaneous data for comparison analysis, so that the long-term hidden danger caused by slow degradation of infrastructure performance or seasonal factors cannot be revealed. Therefore, a new intelligent water health monitoring technical scheme is needed, which has deeper data utilization, can perform long-period correlation reasoning, and has transparent and interpretable diagnosis process. Disclosure of Invention In order to overcome the defects that the existing intelligent water service monitoring system is shallow in data utilization, lacks long-period correlation analysis, is poor in model interpretation and is severely dependent on labeling data, the invention provides a multi-mode cognition reasoning-based intelligent water service infrastructure health monitoring system, which can improve diagnosis accuracy, interpretation and long-term hidden danger discovery capability. The invention aims at realizing the following technical scheme: A multi-modal cognitive reasoning-based intelligent water affair infrastructure health monitoring system comprises a data acquisition and conversion module, a multi-modal cognitive reasoning engine and a report generation module, wherein: the data acquisition and conversion module is used for acquiring multi-source heterogeneous time sequence data of the water service infrastructure through a deployed sensor network of the Internet of things and converting the time sequence data into a standardized dynamic visual map; The sensor of the Internet of things comprises a pressure sensor, a flowmeter and a water quality monitor; the dynamic visualization map is at least one of a graph, a thermodynamic diagram or a space-time distribution map; The multi-mode cognitive reasoning engine is connected with the data acquisition and conversion module and is used for receiving the dynamic visual map; The multi-modal cognitive reasoning engine is integrated with a large visual language model and a water affair field dynamic knowledge base, when an abnormal signal is received, a long-period association analysis mechanism is triggered, history and contemporaneous context data associated with a current abnormal point are retrieved and integrated based on the water affair field dyna