Property facility intelligent prediction management system based on internet of things

Abstract

This invention relates to the field of smart property facility operation and maintenance management technology, specifically to an IoT-based intelligent predictive management system for property facilities. This system includes a multi-domain causal network construction module, a risk quantification and causal inference module, and a risk resource joint optimization and decision execution module. This invention integrates prior knowledge of property operation and maintenance with multi-source heterogeneous IoT data to construct a dynamic, directed acyclic causal network covering all elements, uncovering causal relationships between variables. Based on the causal network, it performs risk transmission inference for abnormal events, quantifies fault risks, and identifies intervened nodes, matching corresponding operation and maintenance intervention measures. Coupled with constraints on all operation and maintenance elements, it constructs a multi-objective optimization model to solve for the globally optimal maintenance scheduling scheme and achieve dynamic closed-loop scheduling, enabling proactive risk prevention and efficient allocation of operation and maintenance resources for property facilities.

Description

Technical Field

[0001] This invention relates to the field of smart property facility operation and maintenance management technology, and more specifically, to an intelligent predictive management system for property facilities based on the Internet of Things. Background Technology

[0002] With the deepening of smart community construction, property facility operation and maintenance management is transforming from the traditional reactive model of emergency repairs to a preventive and proactive operation and maintenance model based on the Internet of Things. Accurate prediction and efficient handling of facility risks have become core needs for the property industry to improve service quality and reduce operating costs.

[0003] Currently, most existing intelligent operation and maintenance technologies for property facilities are based on data correlation fitting and time series statistical models. They can only predict the failure probability of a single device's own deterioration trend, but cannot explore the causal relationships between multiple domain elements such as facility clusters, personnel operations, environmental changes, and public events. They are difficult to predict the risk of sudden failures and cascading failures triggered by external events, resulting in serious lag in risk prevention and control and an inability to achieve proactive intervention. At the same time, existing technologies generally suffer from the problem of disconnect between risk prediction and operation and maintenance scheduling. They fail to embed hard constraints such as operation and maintenance human resources, skill matching, spare parts inventory, and operation time windows into the core logic of prediction and decision-making. They can only formulate operation and maintenance plans based on simple ranking of risk levels, and cannot achieve the global optimization of multiple objectives such as risk prevention and control, cost control, and execution feasibility. This makes it difficult to implement the prediction results, forming a common industry problem of prediction and operation and maintenance being two separate entities.

[0004] To address the numerous shortcomings of the existing technologies, this invention proposes an intelligent predictive management system for property facilities based on the Internet of Things. Summary of the Invention

[0005] In view of the shortcomings of existing technologies, the purpose of this invention is to provide an intelligent predictive management system for property facilities based on the Internet of Things.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The Internet of Things-based intelligent predictive management system for property facilities includes a multi-domain causal network construction module, a risk quantification and causal inference module, and a risk resource joint optimization and decision execution module.

[0008] The multi-domain causal network construction module is used to integrate prior knowledge in the property operation and maintenance field with multi-source heterogeneous IoT data to construct a dynamic directed acyclic causal network covering all elements of property facility clusters, personnel operations, environmental status, and public events. It abandons the traditional data correlation fitting method, explores the causal relationship between variables in the property operation and maintenance scenario, and provides a causal structure basis for risk inference.

[0009] The risk quantification and causal deduction module is used to perform causal path deduction on real-time captured abnormal events based on the aforementioned dynamic causal network, quantify the real-time failure risk level of each property facility, identify the interventionable nodes in the risk transmission path, and match corresponding operation and maintenance blocking measures for each interventionable node.

[0010] The risk resource joint optimization and decision execution module is used to couple the constraints of all elements of property operation and maintenance with the aforementioned facility risk quantification results and candidate operation and maintenance blocking measures to construct a multi-objective joint optimization model to solve the maintenance scheduling scheme, and to realize dynamic closed-loop scheduling based on the changes in operation and maintenance status and risk events.

[0011] Furthermore, the multi-domain causal network construction module constructs an initial causal framework by extracting standardized causal rules from the property operation and maintenance field and transforms them into soft constraints for causal learning. It performs denoising, missing value imputation, and feature standardization on multi-source heterogeneous property operation and maintenance data, while performing time alignment to generate a standardized dataset adapted for causal learning. Combining the initial causal framework constraints, it infers the causal relationships between variables from the observation data and generates a directed acyclic causal graph.

[0012] Furthermore, the multi-domain causal network construction module dynamically adjusts the structure and causal association weight parameters of the causal network through two mechanisms: periodic full updates and triggered incremental updates. This maintains the matching degree between the causal network and the actual operating status of the property system. The triggered incremental updates are triggered by facility operation anomalies, environmental changes, or public events.

[0013] Furthermore, the multi-source heterogeneous IoT data accessed by the multi-domain causal network construction module includes facility operation sequence data, human operation event data, environmental and public event data, owner feedback text data, and security and fire protection event data. Corresponding preprocessing operations are performed for different types of data, including trend smoothing processing for time series data, keyword extraction processing for text data, and classification and encoding processing for event data. After feature extraction, a unified time alignment processing is performed on all features.

[0014] Furthermore, the risk quantification and causal inference module uses real-time captured abnormal events as root cause nodes, searches for the optimal risk transmission path in the dynamic causal network, calculates the cumulative causal effect of the event on each facility node, and combines the facility's own baseline health to quantify the facility's comprehensive real-time failure risk value.

[0015] Furthermore, the risk quantification and causal deduction module distinguishes between non-interventional and interventional nodes based on the node type of the risk transmission path, identifies core risk blocking points, and matches pre-configured operation and maintenance blocking measures for each interventional node. The operation and maintenance blocking measures are associated with corresponding operation and maintenance personnel skill requirements, spare parts types, and risk elimination efficiency parameters.

[0016] Furthermore, the risk resource joint optimization and decision execution module mathematically quantifies all constraints of property operation and maintenance, including human resources, skill matching, spare parts inventory, and operation time windows. It constructs a multi-objective optimization model with the goals of minimizing system residual risk and minimizing total operation and maintenance cost. The Pareto optimal solution set is obtained by solving the Pareto optimal solution set through the improved NSGA-III multi-objective optimization algorithm adapted to the property operation and maintenance scenario. The improved NSGA-III includes four core improvements: constraint-aware coding mechanism, hierarchical constraint domination mechanism, risk priority-oriented evolutionary operator, and dynamic rescheduling hot start mechanism. Finally, it outputs a maintenance scheduling scheme adapted to business needs.

[0017] Furthermore, when the status of operation and maintenance resources and the status of facility risk change, the risk resource joint optimization and decision execution module locks the issued unchangeable tasks, performs rapid rescheduling for the remaining tasks to be executed based on risk level priority, and incrementally updates the operation and maintenance execution instructions based on the rescheduling results, thereby realizing dynamic closed-loop adjustment of the scheduling scheme.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. This invention constructs a multi-domain dynamic causal network covering all elements, enabling the prediction of fault probability, the deduction of fault causal chains, and proactive risk prevention. Abandoning the traditional correlation fitting approach, this invention integrates prior knowledge from the property operation and maintenance field with multi-source heterogeneous IoT data to construct a directed acyclic causal network covering all elements of property facility clusters, personnel operations, environmental conditions, and public events. This network accurately uncovers the causal relationships between variables in the operation and maintenance scenario. Based on the causal network, risk transmission path deduction is completed, which not only quantifies the fault risk of facility degradation but also effectively predicts sudden faults and cascading failures triggered by external events that cannot be covered by existing technologies. Simultaneously, it identifies key, intervention-enabled nodes, achieving advanced risk prediction and proactive prevention, significantly improving the comprehensiveness and foresight of property facility risk management.

[0020] 2. This invention constructs a joint optimization technology system that inherently couples prediction and operation and maintenance, fundamentally solving the industry problem of the disconnect between prediction and implementation. This invention deeply couples the hard constraints of all elements in property operation and maintenance, such as human resources, skill matching, spare parts inventory, and work time windows, with risk quantification results and operation and maintenance intervention measures, embedding the boundary of operation and maintenance capabilities into the core logic of prediction and decision-making from the bottom layer. It constructs a multi-objective optimization model with the goals of minimizing system residual risk and minimizing total operation and maintenance cost, solving for a globally optimal maintenance scheduling scheme that balances risk, cost, and executability. Simultaneously, through a dynamic closed-loop scheduling mechanism to adapt to changes in scenarios, it achieves the optimal balance between risk prevention and control effectiveness, operation and maintenance cost control, and the feasibility of scheme implementation, thus connecting the entire chain from risk prediction to operation and maintenance implementation. Attached Figure Description

[0021] Figure 1 This is a block diagram of a property facility intelligent predictive management system based on the Internet of Things (IoT).

[0022] Figure 2 This is a flowchart illustrating the implementation of the risk quantification and causal deduction module of the present invention.

[0023] Figure 3 This is a flowchart illustrating the implementation of the risk resource joint optimization and decision execution module of this invention. Detailed Implementation

[0024] Example, refer to Figure 1 The IoT-based intelligent predictive management system for property facilities in this embodiment includes a multi-domain causal network construction module, a risk quantification and causal inference module, and a risk resource joint optimization and decision execution module.

[0025] M1, Multi-Domain Causal Network Construction Module: This module constructs a dynamic causal network covering all elements of facility clusters, people, environment, and events to deduce the causal transmission path of failures and the root causes of risks. This network abandons the traditional data correlation fitting method and instead mines the causal relationships between variables by integrating multi-source heterogeneous data and prior knowledge.

[0026] S11, Construction of the Prior Causal Framework:

[0027] First, based on the Failure Mode and Effects Analysis (FMEA) document, industry maintenance standards, and historical accident investigation reports, standardized causal rules are extracted. These rules are stored in a structured format of (precursor variable, causal effect direction, postcursor variable), such as (elevator traction machine, positive causality, bearing wear), (rainstorm, positive causality, ground water accumulation), etc., forming the initial causal framework.

[0028] This framework serves as a soft constraint for subsequent causal learning. It achieves the constraint effect by adding a prior penalty term to the optimization loss function, ensuring that the learned causal graph conforms to basic physical and engineering common sense, while allowing data-driven correction or supplementation of new causal paths. The specific implementation of the prior constraint is explained in detail in Section S13.

[0029] S12. Multi-source heterogeneous data access and vectorization:

[0030] The system receives five categories of data in real time:

[0031] Facility operation data: such as time series data of elevator vibration, motor current, water pump outlet pressure, etc. The sampling frequency is set according to the characteristics of the equipment, such as vibration data once per second and temperature data once per minute;

[0032] Human operation events: such as maintenance work order records, equipment start-up and shutdown operation logs, and decoration and construction application information. Each record includes a timestamp, operation type, operator ID, and associated equipment.

[0033] Environmental and public event data: Access to heavy rain and high temperature warnings from the meteorological bureau, water and power outage notices from the municipality, and environmental sensor data within the community, such as temperature, humidity, and water level, via API;

[0034] Homeowner feedback signals: Natural language processing is performed on repair work order texts and homeowner group chat records (with permission and privacy protection) to extract fault-related keywords, such as elevator noise and unstable water pressure, and generate semantic vectors. At the same time, the frequency of complaints per unit time is counted.

[0035] Security / Fire Incident Logs: such as smoke alarm and abnormal access control opening records, including event type, location, and time;

[0036] The above data undergoes unified preprocessing and time alignment, as follows:

[0037] 1. Continuous data processing: For continuous data of sensor values, a sliding time window with a step size of 1 minute is used. Within each window, statistical characteristics such as mean, variance, peak value, and range are calculated and normalized, and then used as the feature values ​​within that window.

[0038] 2. Processing of Discrete Events and Text Data: For event-type data, within the time range of the event's effectiveness, the feature value in each time window is set to the intensity value corresponding to the event, and is set to 0 in non-effective time periods; for text data, keyword matching based on a property fault terminology database is used, combined with a Chinese BERT-base pre-trained model, to embed the text content into a 768-dimensional semantic vector, and then the dimensionality is reduced to a 1-dimensional feature value through average pooling to represent the intensity of owner feedback within the corresponding time window;

[0039] 3. Multi-source data time alignment: All preprocessed features are aligned uniformly according to a time window of 1 minute step to form an n×d data matrix X, where each row corresponds to the observation data of a time window, and each column corresponds to a preprocessed feature variable, where n is the total number of time windows and d is the total number of feature variables.

[0040] S13. Causal structure learning based on NOTEARS:

[0041] This step infers causal relationships from observational data, rather than simply calculating correlation coefficients. It uses NOTEARS, a causal structure learning algorithm that incorporates prior knowledge constraints, to construct a causal graph.

[0042] This algorithm transforms the structure learning problem into a continuous optimization problem; for a system with d variables, we aim to find a directed acyclic graph (DAG) and its corresponding weighted adjacency matrix. ,in This represents the causal strength from variable u to variable v; combined with the NOTEARS algorithm of prior skeleton constraints, it is achieved by minimizing a least-squares loss function with DAG constraints and prior constraints:

[0043] ;

[0044] Where X is an n×d data matrix, which comes directly from the preprocessed multi-source heterogeneous data in S12 and contains the observation values ​​of all variables;

[0045] The first term is the least squares loss, which measures the error when predicting the current variable with all other variables as parent variables, consistent with the linear structural equation model (SEM).

[0046] The second term is the L1 regularization term. It is a regularization coefficient used to control the sparsity of the cause-effect graph and prevent overfitting;

[0047] The third and fourth terms are prior skeleton constraint penalty terms, where M is a d×d prior constraint matrix, constructed based on the causal skeleton of S11. If the prior rule determines that there is a causal relationship between variables u and v, then... If the prior rule determines that there is no causal relationship between variables u and v, then No clear a priori ; This represents the weighting coefficient for cases where there is no prior causal constraint, with a value range of [0.01~0.1] and a default value of 0.05. represents the weight coefficient for prior causal constraints, with a value range of [0.01~0.1], defaulting to 0.05, and can be adjusted independently according to the confidence level of the prior rule; ⊙ represents the Hadamard product;

[0048] The key constraint h(W) = 0 is a smooth function that forces the graph W to be acyclic. Its gradient can be computed, allowing it to be solved using a standard numerical optimizer. The specific form of h(W) is:

[0049] ;

[0050] in, h(W) = 0 holds if and only if the graph corresponding to W is acyclic.

[0051] Parameter description and determination method:

[0052] W represents the causal adjacency matrix, the core parameter to be optimized, and its non-zero elements. The variable u represents the direct cause of variable v, and its range of values ​​is... Positive values ​​represent positive promoting effects, while negative values ​​represent negative inhibiting effects. It is usually initialized as a zero matrix. If there are prior causal rules, the initial value can be set according to the prior strength.

[0053] X represents the input data matrix, n is the number of samples, i.e. the total number of time windows, usually the data collected in the last 30 days using a sliding window method. The window size is dynamically adjusted according to the system stability, such as taking the data of the last 4 weeks when recalculating weekly; d is the total number of feature variables, which is determined by the feature dimensions after preprocessing of the multi-source data.

[0054] This represents the regularization coefficient, obtained through cross-validation, such as the BIC criterion, among candidate values. Choose from these options to ensure that the model strikes a balance between fit and complexity;

[0055] This represents the prior constraint weight coefficient, with a value range of [0.01~0.1]. In this embodiment, the default value is 0.05, which can be adjusted according to the confidence level of the prior rule. The more authoritative the prior rule, the higher the confidence level. The larger the value;

[0056] M represents the prior constraint matrix, with the same dimension as W. It is obtained by transforming the prior causal skeleton of S11. The matrix elements take values ​​of 0, 0.5, or 1, corresponding to the three cases of no causal prior, no explicit prior, and causal prior, respectively.

[0057] h(W) represents the smooth DAG constraint function, which ensures that the learned graph structure is acyclic. This is the essential feature that distinguishes causal graphs from general correlation networks. The value of this function is calculated through matrix exponentiation, and its gradient is obtained by matrix exponent derivative.

[0058] Algorithm solution and cause-effect graph generation:

[0059] The L-BFGS optimizer is used to solve the constrained optimization problem described above. The maximum number of iterations is set to 1000, and the convergence threshold is set to... After optimization, a weighted adjacency matrix W is obtained. Effective edges are then selected from the W matrix, with elements having an absolute value greater than 0.05 considered as valid causal edges. This threshold is determined through cross-validation among candidate edges. The optimal choice is made from all valid causal edges, which together form a directed acyclic causal graph G, used for subsequent risk extrapolation.

[0060] S14. Dynamic updating of cause-effect graphs:

[0061] The causal structure of a property management system is not static. For example, the introduction of new equipment or seasonal changes may alter the original causal chain. Therefore, the system adopts an online learning approach, using both regular and triggered updates to dynamically maintain the accuracy of the causal graph.

[0062] 1. Regular updates: A full optimization is performed weekly, based on the sliding window data of the most recent 4 weeks, the S13 optimization algorithm is rerun, and the weights W and structure of the causal network are completely updated;

[0063] 2. Triggered Update: Real-time monitoring of data distribution changes. For each feature variable, using the data within the current sliding window as the test set and the sliding window data from the last causal graph update as the baseline set, a one-sample KS test is performed. When the p-value for more than 30% of the feature variables is less than 0.05, it is determined that the data distribution has changed significantly, triggering an incremental update. The incremental update uses the W matrix obtained from the last update as the initial value and performs optimization based on the latest sliding window data, reducing the maximum number of iterations to 200 and improving update efficiency.

[0064] The updated causal graph will be pushed to the risk quantification and causal deduction module to ensure that it always reflects the current system state.

[0065] M2, Risk Quantification and Causal Deduction Module: such as Figure 2 As shown, this module uses the dynamic causal graph G output by the multi-domain causal network construction module to perform causal inference on real-time events, predict a series of possible consequences, and quantify the risk level of each facility, providing input for subsequent joint optimization.

[0066] S21. Causal path search and risk transmission calculation based on Dijkstra's theory:

[0067] When a new abnormal event e, such as a red alert for heavy rain, occurs at the current moment... When captured by the system, it is regarded as a root cause node in the causal network. The system needs to find all facility nodes that may be affected by this event and quantify their risks.

[0068] This invention proposes a risk propagation model based on a weighted causal graph; for a causal path from event source node s to target facility node j... The probability that an event propagates along the path and eventually affects j is determined by the causal strength of all edges on the path. This embodiment adopts the cumulative causal strength model, which regards risk propagation as a probability decay process. Since the causal network is a directed acyclic graph, we can use the improved Dijkstra algorithm to convert the causal strength weights of the edges into distances to find the most likely causal path from s to all nodes and calculate the cumulative causal effect.

[0069] The specific implementation of the improved Dijkstra's algorithm is as follows:

[0070] To find the path from root cause node s to target node j that maximizes the product of causal effects, the causal weights are transformed as follows: For the edge (u to v), its distance value is set to... , where ln is the natural logarithm function; through this transformation, the product maximization problem is equivalent to the problem of minimizing the sum of path distances, which can be directly solved using Dijkstra's algorithm to obtain the optimal causal path from s to all reachable nodes and the corresponding cumulative causal effect;

[0071] To quantify facility node j at the current time Real-time risk value after being affected by the event Taking into account the cumulative effect derived from all root cause events and the baseline health of j itself; for a single event s, its causal effect contribution to j is the product of the weights of all edges on the path. When there are multiple concurrent events, we take the maximum effect value as the dominant risk, as shown in the following formula:

[0072] ;

[0073] in, Indicates that facility j is at the current time. The comprehensive risk value, ranging from 0 to 1, represents the probability of a failure occurring within the next 24 hours. This value will be used as the input for subsequent optimization modules.

[0074] Indicates that facility j is at the current time. The baseline health score, ranging from 0 to 1, represents the probability of failure caused by the facility's own deterioration and is independent of external events. This value is calculated using two optional methods:

[0075] 1. Wiener Degradation Process Model: Based on historical time-series data of key operating parameters of facility j, the drift coefficient and diffusion coefficient of degradation are fitted to calculate the first arrival time distribution of the current degradation reaching the fault threshold, thereby obtaining the probability of a fault occurring within the next 24 hours, as... The possible values ​​of ;

[0076] 2. LSTM time series prediction model: The input is the time series operation characteristics of facility j over the past 7 days, and the output is the failure probability in the next 24 hours. The output is constrained to the range of [0~1] by the sigmoid activation function. The model is trained offline in advance using historical failure data of property facilities, and the model parameters are updated once a quarter. The calculation is completed by online inference at the edge.

[0077] Indicates the current moment This is a collection of all external events that occurred within a short time window. The default time window is the past hour, while the time window for extreme weather events can be extended to 24 hours. These events are captured in real time by the event processing engine.

[0078] This represents the initial intensity of event s, with a value range of [0~1], determined by the event type and level. For example, a red alert for heavy rain... The value can be set to 0.8, and the value for decoration and construction declaration can be set to 0.3. This value can be dynamically calibrated based on historical data and updated using the exponential moving average method. The update formula is:

[0079] ;

[0080] in This represents the percentage of related facilities that failed within 24 hours of a similar incident occurring in the past year. The smoothing factor is 0.3, and it is updated monthly.

[0081] The optimal causal path from event node s to facility node j is obtained by searching the causal graph G using the improved Dijkstra algorithm described above, and corresponds to the transmission path with the largest cumulative causal effect.

[0082] The element value in the causal adjacency matrix W, from node u to node v, represents the strength of the direct causal effect of u on v, and its value ranges from -1 to 1.

[0083] This represents the sigmoid function, and the formula is: , used to The probability weights are converted to the range [0~1], representing the degree to which the probability of v occurring increases if u occurs; for Inhibitory edges with negative values A value less than 0.5 indicates a suppressive effect. In the search for risk transmission paths, this type of edge will lead to a reduction in the cumulative causal effect, and therefore will not be selected as the optimal risk transmission path.

[0084] Let represent the product of the weights of all edges along path P, and let represent the probability that the influence of event s is fully propagated to j along this path.

[0085] S22. Risk Root Cause Identification and Blockage Point Location:

[0086] Based on the results of causal inference, the system not only knows who is at risk, i.e., high risk... The system has the facilities and knows why, that is, the nodes in the risk transmission path; therefore, the system can automatically identify the interveneable nodes in the path, that is, those intermediate nodes that can be blocked by operation and maintenance actions.

[0087] Judgment rules for interventionable nodes: Each node in the cause-effect graph is pre-labeled with a node type. Node types are divided into non-interventional and interventionable categories. Non-interventional nodes include those that cannot be changed through operation and maintenance actions, such as weather events, public events, and owner feedback. Interventionable nodes include those that can be changed through operation and maintenance actions, such as facility operation status nodes and environmental status nodes. In the risk transmission path, the first interventionable node is the core blocking point, and all interventionable nodes are included in the candidate intervention scope.

[0088] Association and encapsulation of blocking measures: Pre-configure corresponding operation and maintenance blocking measures for each interventionable node. Each measure includes the measure content, required personnel skills, required spare parts, execution time, and expected risk elimination ratio. This mapping is pre-built based on FMEA documents and industry maintenance standards. This represents the proportion of risks corresponding to facility j that can be eliminated after intervention measure l is applied to facility j. The value ranges from 0 to 1; the closer the value is to 1, the better the intervention measure's effectiveness in eliminating the corresponding risk. Its initial value is determined by combining the measure effectiveness coefficient in the FMEA document, the standard effectiveness value in industry maintenance specifications, and the historical execution effectiveness statistics of similar intervention measures over the past three months. For example, this applies to measures such as cleaning drainage outlets to areas with standing water, or reducing the risk of water immersion in garage electrical distribution cabinets. The initial value is set to 0.7, and flood control barriers are installed. The initial value is set to 0.8; to ensure The accuracy is dynamically updated using the exponential moving average method, and the update formula is as follows: ,in, The actual risk elimination rate (i.e., the ratio of the difference between the risk value of facility j before and after the implementation of the measure to the risk value before implementation) after the application of the blocking measure l to facility j in the past month. The smoothing coefficient is set to 0.25, and the update frequency is once a month. For example, if the simulation path is from heavy rain to surface water to the start of the garage drainage pump to the risk of water immersion in the electrical distribution cabinet, the system can identify the surface water as an interventionable node and match the corresponding blocking measures as cleaning the drainage outlet and placing flood control barriers, while heavy rain is not an interventionable node.

[0089] Once the matched candidate blocking measures are finalized, they will be included in the subsequent optimization decision-making process along with routine facility maintenance tasks and passed on to the next module.

[0090] M3, Risk Resource Joint Optimization and Decision Execution Module: such as Figure 3 As shown, this module combines the hard constraints of property operation and maintenance, including personnel, materials, finance, and time, with the facility risk values ​​output by the risk quantification module. It is endogenously coupled with candidate blocking measures, and through a multi-objective optimization model, it directly outputs the globally optimal maintenance strategy under the current constraints and drives its execution.

[0091] S31. Mathematical expression of the full-element constraint model for property operation and maintenance:

[0092] First, the complex constraints of property operation and maintenance resources are mathematically quantified; then, basic scheduling rules are defined: the time cycle of operation and maintenance scheduling is divided into time slots of equal length, each time slot lasting 30 minutes, covering the scheduling cycle for the next 7 days. The time slot number is denoted by t. Let there be a total of... There are 10 maintenance tasks to be performed, including preventive maintenance of facilities, fault repair, and risk mitigation measures. Each task corresponds to a unique facility j, N available maintenance personnel, and K types of spare parts.

[0093] Definition of decision variables: , This means that in time slot t, person i will be assigned to execute task j. This indicates that the assignment will not be executed; each task can only be assigned to one person and will be completed within a consecutive time slot.

[0094] Mathematical expression of constraints:

[0095] 1. Single Task, Single Person Constraint: Each task j can only be assigned to one person for execution, as shown in the formula:

[0096] ;

[0097] Where T is the total number of time slots within the scheduling period. The number of consecutive time slots required to execute task j is determined by the task type, such as routine elevator maintenance. Set it to 4, which is 2 hours;

[0098] 2. Human Resources Time Constraint: Each person i can only perform one task within the same time slot t, as shown in the formula:

[0099] ;

[0100] 3. Skill Matching Constraint: Personnel i can only perform tasks for which they possess the corresponding skills, as shown in the formula:

[0101] ;

[0102] in, Assign a skill type number to task j. Match matrix elements to skills. Representative i possesses the ability to perform skill type k; otherwise, the value is 0. The skill matrix is ​​dynamically adjusted monthly based on personnel training records, certificate updates, and historical task completion evaluations.

[0103] 4. Spare parts inventory constraint: The total amount of spare parts consumed in executing task j cannot exceed the current spare parts inventory, as shown in the formula:

[0104] ;

[0105] in, The number of spare parts k required to perform task j is predetermined by the spare parts list for the maintenance task. The range of time slots that can be executed within the scheduling period; For the current moment The inventory quantity of spare part k is dynamically updated based on inbound and outbound shipments.

[0106] 5. Time Window Constraint: Maintenance work must be performed within the facility's permitted time window, as shown in the formula:

[0107] ;

[0108] in, This is the set of allowed execution time windows for the facility corresponding to task j. For example, the time window for elevator maintenance only includes the time slot corresponding to 0:00 to 4:00 AM. The time window is manually configured by the property management personnel or derived through self-learning from historical data. The self-learning method is to: count the owner complaint rate and equipment load rate of each maintenance operation of the facility in the past year, and filter out the time periods with a complaint rate of less than 5% and an equipment load rate of less than 20% as recommended time windows.

[0109] 6. Staff Scheduling Constraints: Staff member i can only perform tasks during their scheduled on-duty time slots; tasks cannot be performed outside of their on-duty time slots. The constraint is forced to be 0, which is achieved by pre-filtering off-duty time slots;

[0110] S32. Joint optimization objective function of prediction-operation inherent coupling:

[0111] This step directly uses the results of risk prediction as the primary optimization objective, and performs joint optimization with the cost objective to construct the following bi-objective optimization function:

[0112] ;

[0113] in, For all decision variables The resulting decision vector has two objective functions: minimizing the system's residual risk and minimizing the total operating cost, respectively, as defined below:

[0114] Objective function 1: Minimize the residual risk of the system.

[0115] First objective function Minimizing the weighted residual risk of the entire facility cluster does not simply involve ranking by risk level, but rather considering how to execute a set of decisions with limited resources. Afterwards, the total risk still exists in the entire system; this incentivizes the model to prioritize high-leverage tasks that can minimize the overall risk.

[0116] ;

[0117] in, Indicating in decision-making schemes The system residual risk is non-negative, and the smaller the value, the better the overall risk control effect.

[0118] This represents the comprehensive risk value of facility j corresponding to task j at the current moment, calculated using formula S21;

[0119] This represents the proportion of risk that can be eliminated after person i performs task j, with a value range of [0~1]. This parameter is related to the person's skill level and the type of task; highly skilled personnel handle tasks in their area of ​​expertise. The value can be set to 0.95; low-skilled personnel performing similar tasks, The value can be set to 0.6. The matrix values ​​are dynamically updated based on historical maintenance performance data, using an exponential moving average method. The update formula is as follows:

[0120] ;

[0121] in The percentage of facility failures that occur again within 30 days after personnel i performs maintenance tasks on facility j within the past year. The smoothing coefficient is set to 0.2, and the update frequency is once a month.

[0122] This indicates the number of time slots required to execute task j, ensuring that the corresponding risk elimination ratio will only take effect after the task is completed.

[0123] Objective function 2: Minimize total operating and maintenance costs.

[0124] Second objective function Minimize total operating costs, including labor costs, spare parts costs, and service loss costs due to downtime. The specific form can be customized according to the property management company's financial model. The basic formula is as follows:

[0125] ;

[0126] in, This is an indicator variable for whether task j is executed. If and only if ,otherwise ;

[0127] Indicating in decision-making schemes The total operation and maintenance cost is expressed in yuan; the smaller the value, the lower the operation and maintenance cost.

[0128] This represents the unit time labor cost for person i to perform task j, expressed in yuan / minute, and is determined by the person's salary level and task type. The duration of a single time slot is 30 minutes, ensuring that the unit of measurement for labor costs is yuan;

[0129] This represents the total cost of all spare parts required to perform task j, expressed in yuan, and is calculated from the spare parts list and unit price of spare parts for task j.

[0130] This represents the loss caused by the downtime of facility j per unit of time, in yuan / hour, including loss of owner satisfaction, commercial loss, etc., and is determined by the importance level of the facility.

[0131] This indicates the estimated downtime of facility j due to malfunction if task j is not performed, in hours, calculated based on historical malfunction data.

[0132] This variable indicates task execution and is used to distinguish whether a task is executed, ensuring that only executed tasks generate spare parts costs, while unexecuted tasks generate potential downtime loss costs.

[0133] S33. Constrained multi-objective optimization solution based on improved NSGA-III:

[0134] The above formula defines a multi-objective, multi-constraint, and high-dimensional combinatorial optimization problem. This problem is characterized by numerous discrete decision variables, dense hard constraints, high requirements for task execution continuity, large differences in risk priorities, and the need to adapt to dynamic rescheduling specific property scenarios. The conventional NSGA-III algorithm suffers from drawbacks such as a low percentage of initial feasible solutions, lack of priority in constraint processing, disconnect between evolutionary direction and business objectives, and slow convergence of dynamic rescheduling. Therefore, this invention adopts an improved NSGA-III algorithm adapted to the property operation and maintenance scenario for solving the problem. The four improvements and the complete solution process are as follows:

[0135] Improvement 1: Constraint-aware coding and population initialization mechanism.

[0136] To address the characteristics of the decision variables and dense constraints in this patent, the conventional one-dimensional binary encoding scheme is abandoned, and a three-dimensional integer encoding scheme for tasks, personnel, and start time slots is designed to reduce the proportion of infeasible solutions from the source of encoding. The specific implementation is as follows:

[0137] 1. Coding rule design:

[0138] Chromosome coding length equals the total number of maintenance tasks to be performed. Each gene locus on a chromosome corresponds to a task. Each gene locus is encoded by a triplet integer: ,in:

[0139] This represents the personnel number assigned to execute task j, with a value range of 0 to N, where N is the total number of maintenance personnel, and 0 indicates that the task will not be assigned at this time.

[0140] This represents the starting time slot number of task j, ranging from 1 to T. It must satisfy the requirement of consecutive time slots for task execution, i.e., the starting time slot + the number of time slots required by the task. ≤ Total number of time slots T;

[0141] This indicates a task execution flag; 0 means the task will not be executed, and 1 means the task will be executed. It is checked in conjunction with the first two parameters.

[0142] This encoding method is related to the decision variables of this patent. One-to-one correspondence: if For time slots , The remaining time slots correspond to the personnel. It fully covers all possible values ​​of the original decision variables, while satisfying the core constraints of single task and single person and continuous task execution time, thus eliminating the possibility of violating these two types of constraints at the coding level.

[0143] 2. Constraint-aware population initialization rules:

[0144] The initial population size is set to 100. The chromosome genes of each individual are generated according to the following rules to ensure that the initial individuals satisfy the core rigid constraints:

[0145] Personnel ID Filtering: For task j, select only personnel with the corresponding skill requirements (i.e., those who meet the requirements). The personnel are randomly selected from i), which naturally satisfies the skill matching constraint;

[0146] Start time slot filtering: For task j and selected personnel i, only during the time window allowed by the task. Within the intersection range of the on-duty shift slots of personnel i, randomly select those that satisfy continuous The starting point of each time slot It naturally satisfies time window constraints and personnel scheduling constraints;

[0147] Task execution marker: High-risk task ( )of The default value is 1, which forces the task to be included in the execution scope of the initial population and prioritizes the scheduling of high-risk tasks.

[0148] Improvement point 2: Hierarchical constraint and control mechanism.

[0149] To address the varying priorities of compliance with property operation and maintenance constraints, this paper abandons the conventional NSGA-III approach of uniformly calculating constraint violations. Instead, it designs a three-tiered constraint dominance mechanism: rigid hard constraints, flexible soft constraints, and an objective function. This ensures that the solved scheduling scheme prioritizes executability before optimizing risk and cost objectives. The specific implementation is as follows:

[0150] 1. Constraint hierarchy division:

[0151] Based on the business rules of property operation and maintenance, the full constraints defined in S31 are divided into two levels:

[0152] Level 1: Rigid hard constraints, including single-task, single-person constraints, human resource and time constraints, skill matching constraints, time window constraints, and personnel scheduling constraints. These constraints are legally compliant and mandatory requirements for business execution; violations render the scheduling plan completely unenforceable with zero tolerance.

[0153] The second level is flexible soft constraints, namely spare parts inventory constraints. These constraints can be temporarily resolved through emergency restocking, spare parts allocation, etc., and limited violations are allowed. The cost of violations can be included in the total cost target accounting.

[0154] 2. Calculation of constraint violation at different levels:

[0155] For each individual, calculate the violation of both levels of constraints separately:

[0156] Total violation of rigid constraints The sum of violations of all rigid constraints; for equality constraints, the violation is... For inequality constraints, the violation quantity is ; feasible individuals ;

[0157] Total violation of flexible constraints The formula for normalizing the violation rate of spare parts inventory constraints is:

[0158] ;

[0159] in To avoid smooth terms with a denominator of 0, The value range is [0, +∞);

[0160] 3. Hierarchical constraint dominating rules:

[0161] For two individuals p and q, p is considered dominant to q if any of the following conditions are met:

[0162] (1) The amount of rigid constraint violation of p Regardless of how the flexible constraints and objective function perform;

[0163] (2) The rigid constraint violations of p and q are equal (both are 0 or the same non-zero value), and the flexible constraint violation of p is... ;

[0164] (3) The rigid and flexible constraints of p and q are equal, and p is Pareto dominant on the dual objective function (i.e., the residual risk and total maintenance cost of p are not inferior to q, and at least one objective is better).

[0165] This governing rule prioritizes compliance constraints, fully adapts to the business implementation requirements of property operation and maintenance, and avoids conventional algorithms generating a large number of unexecutable theoretical optimal solutions;

[0166] Improvement point 3: Risk priority-oriented evolutionary operator optimization.

[0167] To prioritize the prevention and control of high-risk tasks in property operation and maintenance, the simulated binary crossover operator and polynomial mutation operator of the standard NSGA-III are optimized for specific scenarios to ensure that the evolution direction aligns with business needs, as detailed below:

[0168] 1. Optimization of crossover operators for risk grading:

[0169] A two-point crossover method is used, with a crossover probability set to 0.9. The crossover operation follows these rules:

[0170] First, the gene loci on the chromosomes are divided into high-risk areas according to the task risk level. ), medium-risk areas ( ), low-risk areas ( );

[0171] Crossover points are only allowed to be selected between gene loci in medium-risk and low-risk areas. Gene loci in high-risk areas do not participate in crossover operations and directly inherit the high-risk area genes of individuals with lower rigid constraint violation rates from their parents. This avoids crossover operations from disrupting existing feasible allocation schemes for high-risk tasks and ensures the execution priority of high-risk tasks.

[0172] 2. Constraint-aware mutation operator optimization:

[0173] The mutation probability is set to 1 / chromosome length (i.e. For each gene locus, the mutation operation follows the following constraint-aware rules:

[0174] If the gene locus corresponds to a high-risk task, mutations are only permitted within personnel and time slots that meet skill matching and time window constraints. It can only mutate from 1 to 1, and canceling the execution of high-risk tasks is prohibited;

[0175] If the gene locus corresponds to a low-to-medium risk task, the mutation process prioritizes selecting personnel and time slot combinations that satisfy rigid constraints. Only when there are no combinations that satisfy the constraints are mutation results that violate the constraints allowed, further reducing the proportion of infeasible solutions in the offspring.

[0176] Improvement point 4: Adapt to the population hot start mechanism for dynamic rescheduling.

[0177] To address the dynamic closed-loop scheduling requirement in section S34 of this patent, a population hot-start mechanism specifically designed for rescheduling is presented. This solves the problems of slow convergence speed and the need for random initialization from scratch during conventional NSGA-III rescheduling. The specific implementation is as follows:

[0178] 1. When fast rescheduling is triggered, the Pareto optimal solution set obtained from the previous static scheduling convergence is used as the initial population for rescheduling optimization, rather than being randomly generated;

[0179] 2. For each individual in the initial population, perform state adaptation correction:

[0180] Lock the gene loci corresponding to the issued, unchangeable tasks, and do not participate in subsequent evolutionary operations, which is completely consistent with the rescheduling rules of S34.

[0181] Remove the gene bits corresponding to completed tasks, add new risk tasks as new gene bits, and generate codes that satisfy the constraints according to the initialization rules for the new gene bits;

[0182] In response to changes in resource status (such as staff leave or spare parts shortage), the range of available personnel and time slots for the corresponding gene loci is adjusted, and infeasible coding values ​​are removed.

[0183] 3. After a warm start, the population directly enters the evolutionary iteration process. During rescheduling, the population size is reduced to 50, and the maximum number of iterations is reduced to 50 generations, which can improve the rescheduling solution speed and meet the real-time requirements of dynamic scenarios.

[0184] In summary, the complete solution process and the optimal solution output process are as follows:

[0185] 1. Population initialization: Generate an initial population according to the improved encoding rules and initialization strategy;

[0186] 2. Calculation of constraint violation and objective function: For each individual, calculate the rigid / flexible constraint violation, as well as the two objective function values ​​of residual system risk and total operation and maintenance cost;

[0187] 3. Non-dominated sorting and population selection: Based on the hierarchical constraint dominance rule mentioned above, the merged parent and child populations are non-dominated sorted to divide the Pareto front. A two-layer boundary cross construction method is used to generate 100 uniformly distributed reference points. The next generation population is selected based on the correlation degree of the reference points to maintain the diversity and convergence of solutions.

[0188] 4. Evolutionary Iteration: Generate offspring populations using optimized crossover and mutation operators, merge parent and offspring populations, and repeat steps 2-3 until the maximum number of iterations of 200 generations is reached, or the Pareto front shows no significant optimization for 20 consecutive generations, at which point the algorithm converges.

[0189] 5. Optimal Solution Output: After the algorithm converges, a complete set of Pareto optimal solutions is obtained. The system provides two solution selection strategies:

[0190] Weighted decision strategy: Assign weights to the two objective functions and select the solution with the smallest weighted sum. In the safety-first scenario, the risk objective weight is set to 0.8 and the cost objective weight is set to 0.2; in the cost-first scenario, the risk objective weight is set to 0.3 and the cost objective weight is set to 0.7; in the equilibrium scenario, both objective weights are set to 0.5.

[0191] Manual decision-making strategy: The two objective values ​​of the Pareto front are plotted as a two-dimensional scatter plot to show the residual risk of the system, total operation and maintenance cost, and constraint satisfaction status corresponding to each solution, so that property managers can manually select the execution plan that meets their needs;

[0192] 6. The final output execution plan includes the personnel to be executed for each task, the execution time window, the required spare parts, and the specific work instructions. Simultaneously, the encoded results are mapped back into a decision variable matrix. This is used for subsequent execution tracking and dynamic adjustments;

[0193] S34, Dynamic Closed-Loop Iteration Mechanism:

[0194] A dynamic closed-loop iteration mechanism is constructed to address changes in resource status, such as staff leave, spare parts shortages, or the emergence of new risk events, such as sudden pipe bursts. After a mutation or maintenance operation is completed, the system will automatically trigger the following dynamic adjustment process:

[0195] 1. Status Update:

[0196] Remove completed tasks from the to-do list to release occupied personnel and spare parts resources; insert newly generated emergency tasks and tasks with updated risk values ​​into the task pool, and update parameters such as the risk value and required resources corresponding to the tasks; when the real-time risk value of the facility exceeds the emergency threshold of 0.8, immediately trigger the emergency rescheduling process.

[0197] 2. Fast rescheduling:

[0198] Lock tasks that cannot be changed within the next 4 hours. Unchangeable tasks include those for which execution instructions have already been issued or for which maintenance personnel have already departed for the site. Fix the decision variables corresponding to these tasks and exclude them from the optimization scope of rescheduling. Rerun the optimization algorithm of S33 for the remaining tasks starting from the current state.

[0199] During rescheduling, the population size is reduced to 50 and the maximum number of iterations is reduced to 50 generations due to the significant reduction in the dimensionality of decision variables, thus quickly completing the solution and generating an updated scheduling scheme.

[0200] 3. Incremental execution:

[0201] The updated plan generated by the rescheduling is compared with the plan currently being executed. If the assigned personnel or execution time of a task that has not yet started execution changes, it is determined that there is a conflict. For conflicting tasks and new tasks, the updated execution instructions are pushed to the corresponding personnel through the operation and maintenance APP. For tasks that have not changed, instructions are not pushed repeatedly to ensure a smooth transition of on-site operations.

[0202] Through the detailed description of the above embodiments, the IoT-based intelligent predictive management system for property facilities of the present invention, through a multi-domain causal network construction module, achieves the fusion of multi-source data across all scenarios of property operation and maintenance and the accurate mining of causal relationships among all elements, establishing a dynamically updatable causal network system. This overcomes the technical limitations of traditional correlation analysis and lays a solid causal logical foundation for risk prevention and control. Through the risk quantification and causal inference module, it achieves the tracking of risk transmission paths for abnormal events, the accurate quantification of facility failure risks, and the effective identification of core intervention nodes, completing the core link from risk prediction to matching intervention measures. Through the risk resource joint optimization and decision execution module, it achieves the endogenous coupling and global optimization solution of risk control objectives and constraints of all operation and maintenance elements, as well as closed-loop scheduling and incremental execution in dynamic scenarios. This system comprehensively improves the intelligence level and risk prevention and control capabilities of property operation and maintenance, providing a complete and reliable technical solution for the refined operation and maintenance management of smart communities.

[0203] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0204] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0205] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0206] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0207] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0208] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0209] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0210] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A property facility intelligent predictive management system based on the Internet of Things, characterized in that, It includes a multi-domain causal network construction module, a risk quantification and causal inference module, and a risk resource joint optimization and decision execution module; The multi-domain causal network construction module is used to integrate prior knowledge in the property operation and maintenance field with multi-source heterogeneous IoT data to construct a dynamic directed acyclic causal network covering all elements of property facility clusters, personnel operations, environmental status, and public events. It abandons the traditional data correlation fitting method, explores the causal relationship between variables in the property operation and maintenance scenario, and provides a causal structure basis for risk inference. The risk quantification and causal deduction module is used to perform causal path deduction on real-time captured abnormal events based on the aforementioned dynamic causal network, quantify the real-time failure risk level of each property facility, identify the interventionable nodes in the risk transmission path, and match corresponding operation and maintenance blocking measures for each interventionable node. The risk resource joint optimization and decision execution module is used to couple the constraints of all elements of property operation and maintenance with the aforementioned facility risk quantification results and candidate operation and maintenance blocking measures to construct a multi-objective joint optimization model to solve the maintenance scheduling scheme, and to realize dynamic closed-loop scheduling based on the changes in operation and maintenance status and risk events.

2. The intelligent predictive management system for property facilities based on the Internet of Things according to claim 1, characterized in that, The multi-domain causal network construction module constructs an initial causal framework by extracting standardized causal rules from the property operation and maintenance field and transforms them into soft constraints for causal learning. It performs denoising, missing value imputation, and feature standardization on multi-source heterogeneous property operation and maintenance data, while performing time alignment to generate a standardized dataset adapted for causal learning. Combining the initial causal framework constraints, it infers the causal relationships between variables from the observation data and generates a directed acyclic causal graph.

3. The intelligent predictive management system for property facilities based on the Internet of Things according to claim 2, characterized in that, The multi-domain causal network construction module dynamically adjusts the structure and causal association weight parameters of the causal network through two mechanisms: periodic full updates and triggered incremental updates. This maintains the matching degree between the causal network and the actual operating status of the property system. The triggered incremental updates are triggered by facility operation anomalies, environmental changes, or public events.

4. The IoT-based intelligent predictive management system for property facilities according to claim 2, characterized in that, The multi-domain causal network construction module accesses multi-source heterogeneous IoT data, including facility operation sequence data, human operation event data, environmental and public event data, owner feedback text data, and security and fire protection event data. Corresponding preprocessing operations are performed for different types of data. Specifically, time-series data undergoes trend smoothing, text data undergoes keyword extraction, and event data undergoes classification and coding. After feature extraction, all features undergo unified time alignment.

5. The intelligent predictive management system for property facilities based on the Internet of Things according to claim 1, characterized in that, The risk quantification and causal deduction module uses real-time captured abnormal events as root cause nodes, searches for the optimal risk transmission path in the dynamic causal network, calculates the cumulative causal effect of the event on each facility node, and combines the facility's own baseline health to quantify the facility's comprehensive real-time failure risk value.

6. The IoT-based intelligent predictive management system for property facilities according to claim 5, characterized in that, The risk quantification and causal deduction module distinguishes between non-interventional and interventional nodes based on the node type of the risk transmission path, identifies core risk blocking points, and matches pre-configured operation and maintenance blocking measures for each interventional node. The operation and maintenance blocking measures are associated with corresponding operation and maintenance personnel skill requirements, spare parts types, and risk elimination efficiency parameters.

7. The intelligent predictive management system for property facilities based on the Internet of Things according to claim 1, characterized in that, The risk resource joint optimization and decision execution module mathematically quantifies all constraints of property operation and maintenance, including human resources, skill matching, spare parts inventory, and operation time windows. It constructs a multi-objective optimization model with the goals of minimizing system residual risk and minimizing total operation and maintenance cost. The Pareto optimal solution set is obtained by solving the Pareto optimal solution set through the improved NSGA-III multi-objective optimization algorithm adapted to the property operation and maintenance scenario. The improved NSGA-III includes four core improvements: constraint-aware coding mechanism, hierarchical constraint domination mechanism, risk priority-oriented evolutionary operator, and dynamic rescheduling hot start mechanism. Finally, it outputs a maintenance scheduling scheme adapted to business needs.

8. The intelligent predictive management system for property facilities based on the Internet of Things according to claim 7, characterized in that, When the status of operation and maintenance resources and facility risk status change, the risk resource joint optimization and decision execution module locks the issued unchangeable tasks, performs rapid rescheduling for the remaining tasks to be executed based on risk level priority, and incrementally updates the operation and maintenance execution instructions based on the rescheduling results, thereby realizing dynamic closed-loop adjustment of the scheduling scheme.

Images