Multi-source data fusion engineering intelligent decision cockpit system

Abstract

This invention discloses an intelligent decision-making cockpit system for engineering projects based on multi-source data fusion, belonging to the fields of digital engineering management and control, artificial intelligence decision-making, multi-source data processing, and visualization. The system collects heterogeneous data from across the engineering domain through a multi-source data access layer, performing protocol conversion and preprocessing. A data fusion processing layer deeply correlates multi-source data through weighted fusion at the data layer, hybrid neural network extraction and fusion at the feature layer, and evidence-based reasoning fusion at the decision layer. An intelligent decision engine layer performs risk warnings, scheme deductions, and resource scheduling based on the fusion results and knowledge graphs, generating decision commands. A cockpit visualization layer dynamically displays and hierarchically manages the system based on a 3D digital twin model. A closed-loop management execution layer pushes commands to the terminal, tracks execution, and receives feedback to continuously optimize system parameters. This invention helps break down data silos in engineering projects, improving decision-making predictability and closed-loop execution capabilities.

Description

Technical Field

[0001] This invention relates to the fields of digital engineering management and control, artificial intelligence decision-making, multi-source data processing and visualization, and specifically to an intelligent engineering decision-making cockpit system that integrates multi-source data. Background Technology

[0002] As the engineering and construction industry transforms towards digitalization, intelligence, and lean manufacturing, the IoT sensing terminals, professional monitoring equipment, video surveillance systems, BIM modeling platforms, and business management systems deployed on construction sites continuously generate massive amounts of multimodal and heterogeneous data, covering all dimensions of information, including structural stress monitoring, environmental parameters, personnel location, equipment operating conditions, schedule and cost, safety hazards, geology and hydrology, and construction logs. However, the current engineering decision-making and control system still faces some issues that require improvement in practical applications.

[0003] First, multi-source data is often stored in different hardware terminals and software systems. Due to incompatible data protocols, inconsistent formats, and non-standardized data, as well as the lack of a unified access and collaborative scheduling mechanism, the decision-making process often relies on local data sources, making it difficult to form a comprehensive data perspective for the entire project and easily leading to biased judgments.

[0004] Secondly, existing engineering monitoring platforms mainly rely on simple data aggregation and static visualization in terms of data processing. They lack in-depth feature extraction and correlation fusion methods for engineering time-series monitoring data, unstructured videos, 3D models and text materials, making it difficult to uncover hidden risk patterns, schedule deviations and the root causes of potential problems behind the data.

[0005] Furthermore, in practice, project management still relies heavily on human experience and judgment. The capabilities in real-time perception, intelligent analysis, and proactive early warning need to be strengthened. Risk management is somewhat delayed, the means of scheme simulation are limited, and the accuracy of resource allocation also has room for improvement. It is often insufficient to cope with the dynamic and complex nature of project construction.

[0006] In addition, traditional decision-making dashboards typically only have data display functions and lack a linkage link between hierarchical control, scenario adaptation, instruction issuance and execution feedback. The human-computer interaction method is relatively simple, the control granularity is relatively coarse, and a closed-loop control mechanism covering the entire engineering process has not yet been formed.

[0007] To address the aforementioned issues, there is an urgent need for an engineering intelligent decision-making cockpit system that integrates multi-source data to solve the problems associated with traditional methods. Summary of the Invention

[0008] The purpose of this invention is to provide an engineering intelligent decision-making cockpit system that integrates multi-source data. Through multi-source data access and three-level progressive fusion, it effectively integrates heterogeneous data across the entire engineering domain, breaking down information silos. The intelligent decision engine, combined with knowledge graphs, provides proactive early warning and inference, enhancing risk predictability. The closed-loop execution link realizes instruction issuance, tracking feedback, and model iteration, promoting engineering management to form a continuous improvement cycle of perception, decision-making, execution, and optimization, thereby improving management efficiency and resource utilization.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An engineering intelligent decision-making cockpit system that integrates multi-source data includes: a multi-source data access layer, a data fusion processing layer, an intelligent decision engine layer, a cockpit visualization layer, and a closed-loop control and execution layer that are interconnected bidirectionally via a real-time data bus; The multi-source data access layer collects heterogeneous data from the entire engineering domain and performs protocol conversion, format standardization, and noise reduction and deduplication preprocessing on the heterogeneous data. The data fusion processing layer receives the preprocessed data and performs deep correlation on multi-source data through data layer weighted fusion, feature layer hybrid neural network extraction fusion, and decision layer evidence reasoning fusion. The intelligent decision engine layer, based on the output of the data fusion processing layer, performs risk identification and early warning, scheme deduction and resource scheduling, and generates decision instructions; The cockpit visualization layer is based on a three-dimensional digital twin model, which dynamically visualizes the data of the entire engineering field and presents differentiated control interfaces according to different permission levels. The closed-loop control execution layer pushes decision instructions to the corresponding terminals, tracks the execution process and receives execution feedback, and optimizes the system based on the feedback data.

[0010] Furthermore, the multi-source data access layer includes an IoT sensing access module, a business system docking module, an external data import module, and a data preprocessing unit; The IoT sensing access module connects to special engineering equipment through a general industrial protocol to collect sensing data in real time. The business system integration module automatically synchronizes structured business data and three-dimensional digital twin models through the business system; The external data import module is used to import external data in batches; The data preprocessing unit is used to perform noise reduction, deduplication, missing value filling, outlier removal, spatiotemporal alignment, and format standardization on the raw data collected by the IoT sensing access module, business system docking module, and external data import module.

[0011] Furthermore, the data fusion processing layer includes a data pool storage unit, a feature fusion module, a spatiotemporal correlation module, and an engineering knowledge graph construction unit; The data pool storage unit is used to store system-related data; The feature fusion module is used to perform data layer weighted fusion, feature layer hybrid neural network extraction fusion, and decision layer evidence reasoning fusion. The spatiotemporal correlation module is used to establish a three-dimensional correlation relationship between time, space, and data indicators; The engineering knowledge graph construction unit is used to integrate all data and build a knowledge graph specifically for the engineering field.

[0012] Furthermore, the weighted fusion of the data layer specifically refers to: Weighted average normalization is used to perform dimensionless and weighted merging on the preprocessed data, where the weight coefficients are adaptively calculated based on the inverse of the variance of each data source.

[0013] Furthermore, the feature layer hybrid neural network extraction and fusion specifically involves: A hybrid neural network combining CNN, LSTM, and multi-head attention mechanism is used to extract high-contribution features.

[0014] Furthermore, the decision-making level evidence reasoning fusion specifically includes: The DS evidence theory is used to synthesize evidence from preliminary decisions from different feature branches in order to output the final fused decision proposition.

[0015] Furthermore, the intelligent decision engine layer includes a risk warning module, an intelligent inference module, a resource scheduling module, and a decision generation module; The risk warning module has built-in multi-level warning thresholds and risk assessment models, and combines the final fusion decision proposition and engineering domain-specific knowledge graph to identify and warn of risks. The intelligent simulation module links the three-dimensional digital twin model and real-time sensing data to perform visualized dynamic simulations of different scenarios. The resource scheduling module is used to construct a resource optimization objective function and solve it to achieve dynamic optimization scheduling; The decision generation module combines all the system data and, after manual intervention, outputs decision instructions.

[0016] Furthermore, the cockpit visualization layer includes a panoramic visualization module, a hierarchical control module, a scene switching module, and an interactive operation unit; The panoramic visualization module displays data from a high-definition decision-making screen system. The hierarchical control module is configured with a differentiated visual interface based on three levels of authority: decision-making level, management level, and execution level. The scene switching module has multiple standardized scenes built-in; The interactive operation unit is used to enable interactive operation with the high-definition decision-making screen.

[0017] Furthermore, the closed-loop control execution layer includes an instruction issuance unit, a terminal execution module, a result feedback module, and an optimization iteration module; The instruction issuing unit is used to push decision instructions to the corresponding responsible persons; The terminal execution module is located on-site and is used to receive and execute decision instructions; The result feedback module is used to collect feedback data during the execution process; The optimization and iteration module is used to optimize and iterate the system based on feedback data from the execution process.

[0018] In summary, the present invention has at least one of the following beneficial technical effects: First, it helps break down data silos and improve the depth and quality of data fusion. By constructing a unified multi-source heterogeneous data access standard and adopting a three-level progressive fusion algorithm of "data layer - feature layer - decision layer", it effectively handles the problems of dimensional differences and semantic fragmentation of data from different sources and in different formats, providing more confident full-domain data support for subsequent intelligent decision-making.

[0019] Second, it helps to form a closed-loop management and control system throughout the entire process, ensuring the effective implementation of decision-making instructions. The system closely links the intelligent decision-making engine with the closed-loop management and execution layer, connecting the entire chain from data perception, intelligent analysis, instruction issuance to execution feedback and optimization iteration. This enables decision-making instructions to be pushed in a targeted manner, the execution process to be tracked throughout, and the execution results to be transmitted back in real time. This helps to solve the problem of the disconnect between decision-making and execution, and provides a foundation for the system's continuous self-optimization.

[0020] Third, it helps improve the predictability of decision-making and risk prevention capabilities. By incorporating multi-level early warning thresholds and risk assessment models, combined with real-time fusion data and engineering knowledge graph rules, the system can identify and issue graded early warnings for various engineering risk scenarios in real time. It also utilizes digital twin technology to visualize and simulate solutions, allowing for more sufficient response time for risk management and reducing reliance on human experience.

[0021] Fourth, it facilitates comprehensive visualization and refined management of the entire project. Utilizing 3D digital twin and large-screen interactive technology, managers can intuitively grasp core information such as project progress, risk distribution, and resource allocation. The tiered management mechanism provides differentiated management interfaces for personnel at different levels, balancing a holistic decision-making perspective with on-site execution details, effectively improving management efficiency and human-machine collaboration. Attached Figure Description

[0022] Figure 1This is a schematic diagram of the system structure of the present invention; Figure 2 This is a schematic diagram of the multi-source data access layer structure; Figure 3 This is a schematic diagram of the feature fusion module structure. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0024] like Figure 1 As shown, the present invention provides an engineering intelligent decision-making cockpit system for multi-source data fusion, comprising: a multi-source data access layer, a data fusion processing layer, an intelligent decision engine layer, a cockpit visualization layer, and a closed-loop control execution layer that are interconnected bidirectionally via a real-time data bus; The multi-source data access layer collects heterogeneous data from the entire engineering domain and performs protocol conversion, format standardization, and noise reduction and deduplication preprocessing on the heterogeneous data. The data fusion processing layer receives the preprocessed data and performs deep correlation on multi-source data through data layer weighted fusion, feature layer hybrid neural network extraction fusion, and decision layer evidence reasoning fusion. The intelligent decision engine layer, based on the output of the data fusion processing layer, performs risk identification and early warning, scheme deduction and resource scheduling, and generates decision instructions; The cockpit visualization layer is based on a three-dimensional digital twin model, which dynamically visualizes the data of the entire engineering field and presents differentiated control interfaces according to different permission levels. The closed-loop control execution layer pushes decision instructions to the corresponding terminals, tracks the execution process and receives execution feedback, and optimizes the system based on the feedback data.

[0025] Next, each layer will be introduced in detail: 1. Multi-source data access layer like Figure 2 As shown, the multi-source data access layer serves as the system data entry point, responsible for the standardized collection, protocol conversion, and preprocessing of heterogeneous data across the entire engineering domain. It shields the differences between the underlying hardware and systems, enabling comprehensive data access across the entire domain. It includes an IoT sensing access module, a business system interface module, an external data import module, and a data preprocessing unit. The IoT sensing access module is compatible with various industrial sensors, high-definition cameras, GNSS positioning devices, smart meters, tunnel boring machines / tower cranes and other special engineering equipment. It collects sensing data such as structural stress, ambient temperature and humidity, displacement and settlement, personnel positioning, equipment operating parameters, and video images in real time through common industrial protocols such as MQTT, Modbus, HTTP, and TCP / IP.

[0026] The business system integration module uses API gateway, database middleware, and ETL tools to integrate with business systems such as project progress management, cost estimation, quality management, safety management, BIM model, and supervision logs, and automatically synchronizes structured business data with the three-dimensional digital twin model. The external data import module supports batch import, OCR recognition, and text parsing of unstructured / semi-structured data such as geological survey reports, construction organization designs, weather forecasts, public opinion information, and standards, supplementing external decision-making data. The data preprocessing unit performs noise reduction, deduplication, missing value filling, outlier removal, spatiotemporal alignment, and format standardization on the raw data collected by the IoT sensing access module, business system docking module, and external data import module to eliminate data noise and errors and ensure the data quality for subsequent fusion analysis.

[0027] 2. Data Fusion Processing Layer The data fusion processing layer is the core data processing unit of the system. It constructs a hierarchical and multi-dimensional data fusion mechanism to realize feature extraction, semantic association and deep fusion of multi-source data, and break down data silos. It includes a data pool storage unit, a feature fusion module, a spatiotemporal association module and an engineering knowledge graph construction unit. The data pool storage unit adopts a distributed cloud storage architecture, which is divided into a real-time data cache area, a historical data warehouse, a model database, and a knowledge graph library. These respectively store real-time perception stream data, historical business data, BIM models and algorithm models, and engineering domain knowledge rules, supporting efficient reading, writing, and retrieval of massive amounts of data. like Figure 3 As shown, the feature fusion module adopts a three-level progressive fusion algorithm of "data layer - feature layer - decision layer" to achieve accurate fusion for the characteristics of multi-source heterogeneous data in engineering. The fusion rules and mathematical formulas for each level are as follows: (1) Weighted fusion of data layers Dimensionless and weighted merging is performed on multi-source sensing data to eliminate dimensional differences and local errors. The fusion formula is as follows: ,in, For the data layer fusion result, For the number of data sources, For the first The weight coefficients of each data source (satisfying) ), For the first The weighting coefficients for the standardized single-value data from each data source are adaptively calculated using the inverse variance method, as follows: In the formula For the first The data variance of each data source For the first The variance of data from each data source is used to characterize data stability.

[0028] (2) Feature layer hybrid neural network extraction and fusion A hybrid neural network combining CNN, LSTM, and multi-head attention is employed to extract high-contribution features. Specifically, a convolutional neural network (CNN) is used to extract spatial features from videos and images, while a long short-term memory network (LSTM) extracts dynamic features from time-series monitoring data. A multi-head attention mechanism is used to dynamically assign feature weights, enhancing the extraction of key risk features. The formula for calculating the attention weights is as follows: in: For query matrix, For the key matrix, For value matrices, The feature vector dimension is used to normalize the feature weight distribution using the Softmax function, and high-contribution features are then selected.

[0029] Next, we will explain in detail that this invention adopts a parallel architecture of "CNN feature extraction branch + LSTM feature extraction branch + multi-head attention fusion branch". The three branches work together and progress layer by layer to ensure comprehensive extraction and efficient fusion of spatial and dynamic features. Among them, the CNN branch and LSTM branch process different types of data in parallel, outputting spatial feature maps and temporal feature sequences respectively. Then, the multi-head attention branch dynamically assigns weights and fuses the two types of features, finally outputting high-contribution fused features. The overall structure is compact and adaptable to the multimodal characteristics of engineering data, avoiding the limitations of a single model in feature extraction.

[0030] Based on the characteristics of data in the engineering field, a targeted training scheme was designed to ensure that the model training effect is adapted to engineering scenarios, as detailed below: ① Training Dataset Adaptation: Construct a dedicated feature extraction dataset for engineering projects, covering video and image data (no less than 10,000 records) and time-series monitoring data (no less than 50,000 records) from various engineering sites. The data comes from different engineering types (construction, municipal engineering, rail transit), including data on normal and abnormal operating conditions. Core feature labels (such as hidden danger features, equipment failure features, and schedule deviation features) are annotated to ensure the diversity and representativeness of the dataset and avoid insufficient generalization ability of the model.

[0031] ② Training parameter adaptation: The Adam optimizer is used with a learning rate of 0.001 and a step decay strategy (decreasing by 10% every 10 epochs) to avoid gradient vanishing or oscillation during training. The loss function is a hybrid loss function combining cross-entropy loss and mean squared error loss. Cross-entropy loss is used to optimize feature classification accuracy, while mean squared error loss is used to optimize the rationality of feature weight allocation, adapting to the dual requirements of "classification + weight allocation" in engineering feature extraction. The training batch size is set to 32, and the training epochs are set to 50. An early stopping strategy is adopted (training is stopped when the validation set loss does not decrease for 5 consecutive epochs) to prevent model overfitting.

[0032] ③ Transfer Learning and Fine-tuning: Transfer learning is performed based on the pre-trained ResNet-50 model (adapted for image feature extraction) and the BiLSTM model (adapted for temporal feature extraction). The parameters of the pre-trained model are used as initial parameters, and only the top convolutional layer, LSTM hidden layer and multi-head attention layer of the model are fine-tuned to reduce training costs and improve the feature extraction accuracy of the model in engineering scenarios. Fine-tuning strategies are designed to meet the differentiated needs of different engineering types. For example, for rail transit projects, the feature extraction parameters of tunnel images and shield machine time series data are fine-tuned, and for high-rise building projects, the feature extraction parameters related to high formwork and tower cranes are fine-tuned to achieve scenario-based adaptation of the model.

[0033] ④ Training, Validation and Optimization: The training set and validation set are divided into an 8:2 ratio. After each epoch, the feature extraction accuracy of the validation set is tested. For features with low recognition accuracy in the validation set (such as features of subtle hidden dangers or features of sudden changes in time under complex working conditions), corresponding datasets are supplemented for targeted training. After training, the model is tested with actual engineering field test data. Based on the test results, parameters such as the model convolution kernel size, number of attention heads, and learning rate are adjusted to ensure that the model feature extraction accuracy is not less than 95%, which meets the needs of intelligent decision-making in engineering.

[0034] Through the above adaptation modifications and targeted training, the hybrid neural network can accurately extract spatial features (such as hazard location, equipment status, and construction scene) from videos and images, as well as dynamic features (such as parameter change trends and risk mutation signals) from time-series monitoring data. Furthermore, a multi-head attention mechanism is used to allocate dynamic feature weights, enhancing the extraction of key risk features. The formula for calculating the attention weights is as follows: in: For query matrix, For the key matrix, For value matrices, The feature vector dimension is used to normalize the feature weight distribution using the Softmax function, and high-contribution features are then selected.

[0035] (3) Evidence-based reasoning fusion at the decision-making level The DS evidence theory is used to synthesize evidence from preliminary decisions based on different feature branches, reducing decision uncertainty. The basic probability allocation function synthesis formula is as follows: in: The conflict coefficient represents the degree of conflict among multiple pieces of evidence. , Basic probability assignment for different feature branches For the decision proposition after integration, The basic probability function of the first feature branch A proposition, The basic probability function of the first feature branch A proposition.

[0036] The data layer completes the unification of multi-source data formats, deduplication and merging, and preliminary correlation; the feature layer extracts deep features and filters redundant information through the above-mentioned hybrid neural network; the decision layer realizes multi-feature fusion decision-making through dynamic weight allocation and evidence reasoning, and completely solves the problems of heterogeneous data fusion bias and low confidence.

[0037] The spatiotemporal correlation module binds spatiotemporal tags to all access data based on engineering GIS geographic information and construction timeline, and establishes a three-dimensional correlation relationship of "time-space-data indicators" to realize full-domain positioning, tracing and penetrating analysis of engineering site data; The engineering knowledge graph construction unit integrates data such as engineering construction specifications, risk cases, handling procedures, and construction methods to build a knowledge graph specifically for the engineering field. This enables semantic data association, risk rule reasoning, and reuse of experiential knowledge. The specific construction process will be described in detail below: The engineering knowledge graph construction unit revolves around multi-source, multi-modal engineering data such as construction specifications, risk cases, handling procedures, and construction methods. It strictly follows eight full-process steps: ontology design, data preprocessing, knowledge extraction, knowledge fusion, knowledge storage, quality verification, inference rule configuration, and iterative update to build an engineering-specific knowledge graph. This enables data semantic association, intelligent risk inference, and reuse of industry experience. The specific process of knowledge graph construction is as follows.

[0038] Ontology Design: A top-down approach is adopted, relying on the Neo4j tool to build a graph skeleton, defining 8 major categories of core entities (construction specifications, risk types, hidden danger locations, processes and methods, disposal measures, equipment models, personnel positions, and construction locations), semantic relationships and entity and relationship attributes, and setting constraint rules to build a standardized domain knowledge system.

[0039] Data preprocessing: Classify and process multimodal data such as PDF, text, images, and flowcharts, and unify the data format through noise reduction, format standardization, word segmentation normalization, OCR parsing, and key information extraction.

[0040] Knowledge extraction: A hybrid approach combining deep learning and rule engines is adopted, using algorithms such as ALBERT and BiLSTM-CRF to accurately extract engineering entities, relationships, and attributes, forming standardized knowledge triples to ensure the accuracy of professional knowledge extraction.

[0041] Knowledge integration: Through operations such as entity alignment, relationship sorting, and semantic disambiguation, the problems of redundancy, conflict, and duplication of multi-source knowledge are solved, and a unified and coherent knowledge system is formed.

[0042] Knowledge storage: A hybrid architecture of Neo4j graph database and Highgo relational database is adopted to store graph association knowledge and structured attribute data respectively, and with multi-dimensional indexes, efficient retrieval and multi-hop association queries are achieved.

[0043] Quality verification: A dual mechanism of automatic algorithm verification and manual review by industry experts is implemented to identify problems such as logical conflicts and knowledge gaps, correct and optimize knowledge, and ensure high accuracy of map queries.

[0044] Inference rule configuration: Combine norms and expert experience to set rules such as association reasoning, experience reuse, and normative constraints, and integrate domain small models and graph RAG (Graph Retrieval-Augmented Generation) to achieve an upgrade from data query to intelligent reasoning.

[0045] Iterative Updates: A dynamic update mechanism is established to simultaneously add new engineering data to supplement the knowledge map, and to optimize entity, relationship and reasoning rules based on business feedback, so as to continuously adapt to the needs of engineering construction and intelligent decision-making.

[0046] 3. Intelligent Decision Engine Layer The intelligent decision engine layer, based on the fused data and engineering knowledge graph, performs real-time analysis, intelligent inference, and decision generation to replace human experience-based decision-making. It includes a risk warning module, an intelligent inference module, a resource scheduling module, and a decision generation module. The risk warning module has built-in multi-level warning thresholds and risk assessment models. Combined with real-time fusion data and knowledge graph rules, it can identify scenarios such as structural safety, environmental hazards, schedule delays, equipment failures, personnel violations, and fire hazards in real time. It divides the warnings into four levels: red, orange, yellow, and blue, according to the severity of the risks, and accurately locates the risk sources and pushes related data. The intelligent simulation module is based on digital twin technology, linking a three-dimensional digital twin model with real-time sensing data to perform visualized and dynamic simulations of scenarios such as construction plan optimization, risk management, and emergency rescue, simulating the engineering implementation effects under different working conditions and predicting potential problems.

[0047] The resource scheduling module is based on integrated data such as project schedule, personnel and equipment distribution, material inventory, and on-site working conditions. It uses genetic algorithms and linear programming models to construct a resource optimization objective function, realize dynamic optimization scheduling of personnel, equipment, and materials, improve resource utilization, and reduce construction costs. The decision generation module integrates early warning information, simulation results, scheduling schemes and knowledge graph rules to automatically generate standardized decision instructions, supporting manual intervention, correction and confirmation by managers, and forming a final executable decision scheme.

[0048] 4. Cockpit Visualization Layer The cockpit visualization layer adopts three-dimensional digital twin and large screen interaction technology to realize dynamic visualization display of engineering data and human-machine collaborative management, taking into account both overall overview and detailed penetration. It includes a panoramic visualization module, a hierarchical management and control module, a scene switching module and an interactive operation unit. The panoramic visualization module is based on a high-definition decision-making screen, which displays the overall three-dimensional twin model of the project, real-time core indicators, risk distribution heat map, progress bar chart, resource allocation distribution map, and early warning information list, so as to achieve a "one-screen overview" of the entire project. The hierarchical control module is based on three levels of authority: decision-making, management, and execution. It features a customized and differentiated visual interface. The decision-making level focuses on global indicators and major decisions, the management level focuses on specific control data, and the execution level focuses on on-site operation instructions, thus achieving clear responsibilities and hierarchical control. The scenario switching module has built-in standardized scenarios such as security management, progress management, quality management, cost management, equipment management, and emergency command, and supports one-click switching and custom scenario configuration to adapt to different business decision-making needs. The interactive operation unit supports functions such as large-screen touch control, voice interaction, remote control, data drill-down, model rotation, and layer switching. It can quickly retrieve data details, view the reasons for warnings, and issue handling instructions.

[0049] 5. Closed-loop control execution layer The closed-loop control and execution layer connects the "last mile" of decision-making and execution, enabling precise issuance of decision instructions, tracking of the execution process, feedback of results, and system iteration, forming a complete control and execution closed loop. It includes an instruction issuance unit, a terminal execution module, a result feedback module, and an optimization and iteration module. The instruction issuing unit will push the confirmed decision instructions to the corresponding responsible persons through mobile APP, on-site smart terminals and business systems, along with early warning information, handling requirements and time limits; The terminal execution module is used to allow on-site operators to receive instructions via mobile devices and carry out operations such as risk management, resource scheduling, schedule adjustment, and hazard rectification, while recording the entire execution process. The result feedback module collects execution process data, rectification results, and on-site images in real time and automatically transmits them back to the data fusion processing layer to form an execution ledger. The optimization and iteration module continuously optimizes the weights, warning thresholds, and decision model parameters of the three-level fusion algorithm based on execution feedback data and effect evaluation, thereby continuously improving the system's decision-making accuracy and adaptability.

[0050] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0051] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0052] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0053] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0054] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. An engineered intelligent decision cockpit system for multi-source data fusion, characterized in that, include: The system consists of a multi-source data access layer, a data fusion and processing layer, an intelligent decision engine layer, a cockpit visualization layer, and a closed-loop control and execution layer, all interconnected through a real-time data bus. The multi-source data access layer collects heterogeneous data from the entire engineering domain and performs protocol conversion, format standardization, and noise reduction and deduplication preprocessing on the heterogeneous data. The data fusion processing layer receives the preprocessed data and performs deep correlation on multi-source data through data layer weighted fusion, feature layer hybrid neural network extraction fusion, and decision layer evidence reasoning fusion. The intelligent decision engine layer, based on the output of the data fusion processing layer, performs risk identification and early warning, scheme deduction and resource scheduling, and generates decision instructions; The cockpit visualization layer is based on a three-dimensional digital twin model, which dynamically visualizes the data of the entire engineering field and presents differentiated control interfaces according to different permission levels. The closed-loop control execution layer pushes decision instructions to the corresponding terminals, tracks the execution process and receives execution feedback, and optimizes the system based on the feedback data.

2. The multi-source data fused engineering intelligent decision cockpit system according to claim 1, wherein, The multi-source data access layer includes an IoT sensing access module, a business system docking module, an external data import module, and a data preprocessing unit; The IoT sensing access module connects to special engineering equipment through a general industrial protocol to collect sensing data in real time. The business system integration module automatically synchronizes structured business data and three-dimensional digital twin models through the business system; The external data import module is used to import external data in batches; The data preprocessing unit is used to perform noise reduction, deduplication, missing value filling, outlier removal, spatiotemporal alignment, and format standardization on the raw data collected by the IoT sensing access module, business system docking module, and external data import module.

3. The multi-source data fused engineering intelligent decision cockpit system according to claim 2, wherein, The data fusion processing layer includes a data pool storage unit, a feature fusion module, a spatiotemporal correlation module, and an engineering knowledge graph construction unit; The data pool storage unit is used to store system-related data; The feature fusion module is used to perform data layer weighted fusion, feature layer hybrid neural network extraction fusion, and decision layer evidence reasoning fusion. The spatiotemporal correlation module is used to establish a three-dimensional correlation relationship between time, space, and data indicators; The engineering knowledge graph construction unit is used to integrate all data and build a knowledge graph specifically for the engineering field.

4. The multi-source data fused engineering intelligent decision cockpit system according to claim 3, wherein, The weighted fusion of the data layer specifically refers to: Weighted average normalization is used to perform dimensionless and weighted merging on the preprocessed data, where the weight coefficients are adaptively calculated based on the inverse of the variance of each data source.

5. The multi-source data fused engineering intelligent decision cockpit system according to claim 4, wherein, The feature layer hybrid neural network extraction and fusion specifically involves: A hybrid neural network combining CNN, LSTM, and multi-head attention mechanism is used to extract high-contribution features.

6. The multi-source data fused engineering intelligent decision cockpit system according to claim 5, wherein, The aforementioned decision-making level evidence reasoning fusion specifically refers to: The DS evidence theory is used to synthesize evidence from preliminary decisions from different feature branches in order to output the final fused decision proposition.

7. The engineering intelligent decision-making cockpit system based on multi-source data fusion according to claim 6, characterized in that, The intelligent decision engine layer includes a risk warning module, an intelligent inference module, a resource scheduling module, and a decision generation module; The risk warning module has built-in multi-level warning thresholds and risk assessment models, and combines the final fusion decision proposition and engineering domain-specific knowledge graph to identify and warn of risks. The intelligent simulation module links the three-dimensional digital twin model and real-time sensing data to perform visualized dynamic simulations of different scenarios. The resource scheduling module is used to construct a resource optimization objective function and solve it to achieve dynamic optimization scheduling; The decision generation module combines all the system data and, after manual intervention, outputs decision instructions.

8. The engineering intelligent decision-making cockpit system based on multi-source data fusion according to claim 7, characterized in that, The cockpit visualization layer includes a panoramic visualization module, a hierarchical control module, a scene switching module, and an interactive operation unit; The panoramic visualization module displays data from a high-definition decision-making screen system. The hierarchical control module is configured with a differentiated visual interface based on three levels of authority: decision-making level, management level, and execution level. The scene switching module has multiple standardized scenes built-in; The interactive operation unit is used to enable interactive operation with the high-definition decision-making screen.

9. The multi-source data fused engineering intelligent decision cockpit system according to claim 8, wherein, The closed-loop control execution layer includes an instruction issuance unit, a terminal execution module, a result feedback module, and an optimization iteration module; The instruction issuing unit is used to push decision instructions to the corresponding responsible persons; The terminal execution module is located on-site and is used to receive and execute decision instructions; The result feedback module is used to collect feedback data during the execution process; The optimization and iteration module is used to optimize and iterate the system based on feedback data from the execution process.

Images