Building operation and maintenance intelligent decision-making system and method based on digital twinning

Abstract

The invention discloses a building operation and maintenance intelligent decision-making system and method based on digital twinning, and relates to the technical field of intelligent decision-making. The system comprises a data acquisition and historical database construction module, a correlation analysis and dynamic response rule construction module, a sewage data trend pre-judgment module and an operation regulation and data updating module. According to the method, a hierarchical association factor library is constructed, bidirectional causal tracing and cross-scene conduction analysis are carried out during abnormity, an initial operation and maintenance scheme is generated, and the initial operation and maintenance scheme is output, executed and dynamically fed back after collaborative verification and optimization of a digital twin model. According to the method, the problems of cross-scene data splitting and poor decision collaboration in the prior art are solved, and the accuracy, systematicness and practicability of building operation and maintenance decisions are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology, specifically to an intelligent decision-making system and method for building operation and maintenance based on digital twins. Background Technology

[0002] With the improvement of building intelligence, digital twin technology is gradually being applied to the field of building operation and maintenance. By constructing a virtual mapping model of the physical building, it enables visualized management and decision support of operation and maintenance data. Existing intelligent decision-making systems for building operation and maintenance mostly conduct independent analysis for a single operation and maintenance scenario. For example, data collection and analysis modules are established for scenarios such as energy consumption management, equipment fault diagnosis, and space utilization optimization. The data between these modules is fragmented, and there is a lack of an effective cross-scenario correlation analysis mechanism. Some improvement solutions attempt to uncover cross-scenario data correlations, but these are limited to simple factor matching and lack a hierarchical system of correlation factors. This makes it difficult to distinguish between direct driving factors and implicit influencing factors, resulting in low accuracy and efficiency in correlation analysis. Furthermore, existing technologies can only achieve correlation mining in a single direction, failing to trace the chain reaction pathways between multiple scenarios, easily leading to the drawback of "solving a single problem causing anomalies in other scenarios." In addition, the lack of multi-scenario collaborative verification before execution of maintenance actions means that decision-making is often based on optimization of a single scenario, ignoring the potential impact on related scenarios. This results in insufficient practicality and poor coordination in maintenance decisions, failing to meet the needs of collaborative maintenance across all scenarios in complex buildings. Therefore, there is an urgent need for an analytical logic with dynamic correlation tracing and collaborative optimization capabilities to improve the accuracy and systematic nature of building maintenance decisions. Summary of the Invention

[0003] The purpose of this invention is to provide a building operation and maintenance intelligent decision-making system and method based on digital twins to solve the problems mentioned in the background art.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A building operation and maintenance intelligent decision-making method based on digital twins includes the following steps: S1. Construct a hierarchical library of multi-scenario related factors for building operation and maintenance. Store the related factors corresponding to the core scenarios of building operation and maintenance in layers: core driving layer, scenario linkage layer, and implicit influence layer. Configure linkage trigger conditions for each layer of factors. When the core driving layer factor exceeds the corresponding preset threshold, the data collection of the scenario linkage layer factor is automatically activated. When the scenario linkage layer factor exceeds the corresponding preset threshold, the data collection of the implicit influence layer factor is automatically activated. S2. Based on the building digital twin model, real-time operation and maintenance data is obtained. When any scene anomaly is detected, bidirectional causal tracing and multi-scene chain transmission analysis are initiated. The direct related factors and cross-scene chain transmission paths corresponding to the abnormal scene are traced in the forward direction, while the multi-scene chain anomalies and transmission paths caused by the change of hidden influence layer factors are deduced in the reverse direction. S3. Generate an initial operation and maintenance action plan based on the traced chain transmission path, conduct multi-scenario collaborative verification in the digital twin model, and simultaneously simulate the improvement effect of the initial plan on abnormal scenarios, the potential impact on related scenarios, and the dynamic changes of the chain transmission path. S4. Optimize the initial operation and maintenance action plan based on the collaborative verification results, supplement the adaptation measures of related scenarios, form the final decision plan and output it to the building operation and maintenance execution system for execution, so as to complete the anomaly handling and multi-scenario collaborative control.

[0005] Furthermore, S1 includes the following: Based on the building's functional positioning and core operation and maintenance objectives, core scenarios are defined. Scenario boundaries are defined using a three-dimensional quantitative parameter set Ωi, where Ωi = {geographic coordinate range Ai, associated device identifier set Di, core indicator set Mi}, where i is the scenario number, Ai is represented by the building's rectangular coordinates and floor intervals, Di is the unique code set of the physical devices associated with the scenario, and Mi is the core operation and maintenance indicator for the scenario, determined by a variance contribution rate ≥ η, where η is a screening threshold determined by industry statistics. An intersection verification formula ensures that the boundaries of each scenario do not overlap; that is, for any i ≠ j, [Ai ∩ Aj = ∅] ∧ [Di ∩ Dj = ∅] ∧ [the proportion of indicators in Mi ∩ Mj ≤ γ], where γ is the indicator overlap rate constraint threshold. If this condition is not met, Ai or Di is adjusted until the requirement is met. Based on historical building operation and maintenance data, mutual information analysis combined with significance test is used to screen candidate factors; the correlation strength between candidate factors and core indicators of the scenario is calculated by mutual information formula, and after significance test, factors that meet the correlation strength threshold are screened to form a candidate set of correlation factors Fi for each scenario. A two-dimensional quantitative model of factor influence and response speed is constructed. The factor influence If is calculated by combining the partial correlation coefficient with the variance contribution, and the factor response speed Vf is calculated based on the cross-correlation function. A quantitative threshold range is set, and according to the numerical distribution of If and Vf, the candidate factors are respectively assigned to the core driving layer F1_i, the scene linkage layer F2_i, and the latent influence layer F3_i, forming a three-layer set of related factors for each scene. Based on historical data statistical characteristics and operational safety constraints, the 3σ principle combined with dynamic safety margin is adopted. The formula is T1_j=μ1_j+3σ1_j+α·CV1_j, where μ1_j is the historical data mean of the j-th core driving layer factor, σ1_j is the standard deviation, CV1_j=σ1_j / μ1_j is the coefficient of variation, and α is the safety margin coefficient, which is determined based on the equipment rated parameter range. α=α0+α1×(1-equipment rated parameter utilization rate), where α0 and α1 are constant coefficients. The quantile method combined with cross-scenario coupling is adopted, and the formula is T2_m=Q3_m+k1×IQR_m+β·C_m, where Q3_m is the upper quartile, Q1_m is the lower quartile, IQR_m=Q3_m-Q1_m is the interquartile range, k1 is the quantile adjustment coefficient, C_m is the coupling degree of the factor with other scenarios, which is calculated by the scenario correlation matrix C_m=∑|r(Xm,Yn)| / n, where n is the number of other scenarios, r(Xm,Yn) is the correlation coefficient of factor Xm with the core indicator Yn of other scenarios, β is the coupling coefficient, and β=β0+β1×C_m, where β0 and β1 are constant coefficients; When the real-time monitoring value x1_j of the core driving layer factor f1_j is greater than T1_j, the factor data acquisition of the corresponding scene linkage layer F2_i is automatically activated, and the acquisition frequency is adjusted from the basic frequency f0 to f2, and f2=f0×(1+a×(x1_j-T1_j) / T1_j), where a is the frequency adjustment coefficient, and the value range of f2 is f0×b~f0×c, where b and c are frequency constraint thresholds; when the real-time monitoring value x2_m of the scene linkage layer factor f2_m is greater than T2_m, the factor data acquisition of the corresponding latent influence layer F3_i is automatically activated, and the acquisition frequency is adjusted to f3, and f3=f0×(1+d×(x2_m-T2_m) / T2_m), where d is the frequency adjustment coefficient, and the value range of f3 is f0×e~f0×f, where e and f are frequency constraint thresholds; Set a data freshness threshold θ0, and set data freshness θ = proportion of data in the last Δt1 period × w1 + data integrity × w2, where w1 and w2 are weighting coefficients, and w1 + w2 = 1, determined based on the timeliness and reliability requirements of operation and maintenance data; proportion of data in the last Δt1 period = actual data volume collected in the last Δt1 period / data volume to be collected in the last Δt1 period, where Δt1 is the statistical period for data freshness; data integrity = effective data volume / total data volume to be collected, where effective data volume is the amount of data collected that meets the requirement of "data values ​​within a preset reasonable range and without logical conflicts", and total data volume to be collected = collection frequency × statistical period, with data integrity values ​​ranging from [0,1]; trigger threshold update when θ < θ0, with threshold update period Δt = Δt0 × (1 + CV_total), where Δt0 is the basic update period, and CV_total is the mean of the coefficients of variation of all core driving layer factors; recalculate T1_j and T2_m based on historical data in the last Δt period, and update the trigger condition parameters; Data is stored using a distributed time-series database. The database structure includes a table of basic scene information, a table of factor attributes, and a table of trigger conditions. It supports dynamic maintenance of factors for addition and deletion, threshold modification, and scene relationship updates. Furthermore, the data interface between the database and the building digital twin model meets the requirements for real-time synchronization.

[0006] Furthermore, S2 includes the following: Based on the real-time data interface of the building digital twin model, operation and maintenance data of each core scenario are accessed. The data types include real-time monitoring values ​​of activated factors in the core driving layer and scenario linkage layer. The data sampling frequency is consistent with the collection frequency set in S1. For any core scenario i, based on the corresponding core indicator set Mi, the abnormality deviation formula is used to determine whether the scenario is abnormal. The formula is: Li(t)=|xi_t-μ_Mi| / σ_Mi, where xi_t is the real-time monitoring value of the core indicator of scenario i at time t, μ_Mi is the historical average of the core indicator of scenario i, and σ_Mi is the historical standard deviation of the core indicator of scenario i. An abnormality judgment threshold L0 is set. L0 is determined based on the abnormal statistical data of similar building scenarios in the industry (satisfying the abnormality identification accuracy at a confidence level of 95%). When Li(t)≥L0, scenario i is determined to be abnormal at time t, and the abnormal scenario identifier and the abnormality occurrence time t0 are recorded. Based on the three-layer correlation factor set constructed by S1, the correlation factors of the core driving layer F1_i and the scene linkage layer F2_i corresponding to the abnormal scene i are extracted. The real-time correlation strength between each correlation factor and the core abnormal indicator is calculated. The calculation formula is: RI(f,Yi)=MI(f,Yi)×Vf, where MI(f,Yi) is the mutual information value between the correlation factor f and the core indicator Yi of scene i, and Vf is the response speed of the correlation factor f. Factors with RI(f,Yi)≥RI_th are selected as direct correlation factors, where RI_th is the correlation strength screening threshold, which is determined based on the factor contribution statistics of historical abnormal data. Starting with the abnormal scenario i, extract the three-layer correlation factors of all core scenarios, construct the cross-scenario factor correlation strength matrix R, and R=[r_pq] n×n Where n is the total number of correlation factors across all scenarios, and r_pq is the cross-scenario correlation strength between factor fp and factor fq, calculated as: r_pq=|r_corr(fp,fq)|×e^(-τ_pq×δ), where r_corr(fp,fq) is the Pearson correlation coefficient between factor fp and fq, τ_pq is the lag time from fp change to fq response, calculated based on the cross-correlation function; δ is the time decay coefficient, determined based on the timeliness requirements of operation and maintenance data; and the value range of r_pq is [0,1], the larger the value, the more significant the cross-scenario correlation between factors; Based on the correlation strength matrix R, a path search algorithm is used to mine cross-scene transmission paths. The path weight calculation formula is: W_path=∑(r_pq×I_fp), where I_fp is the influence degree of factor fp in the path, and ∑ is the sum of the products of the correlation strength of all adjacent factor pairs in the path and the influence degree of the preceding factor. Paths with W_path≥W1_th are selected as effective positive chain transmission paths, where W1_th is the positive path weight threshold, determined based on the statistical analysis of the abnormal contribution of historical transmission paths. The path form is represented as: abnormal scenario i → directly related factor → cross-scene factor 1 → cross-scene factor 2 → ... → related scenario j. The effective positive chain transmission paths are mapped to the building digital twin model. By verifying the correlation between the three-dimensional spatial coordinates of the model and the equipment association relationship, the correlation of the physical equipment and spatial location corresponding to the factors in the path is verified, ensuring that the transmission path conforms to the physical logic of building operation and maintenance. Extract the real-time monitoring values ​​of the latent influence layer F3_i factor from all scenarios, and calculate the degree of factor anomaly using the formula: Δfk(t)=|x_k(t)-μ_fk| / μ_fk, where x_k(t) is the real-time monitoring value of the latent influence layer factor fk at time t, and μ_fk is the historical average value of the latent influence layer factor fk. Set an anomaly judgment threshold Δ_th, which is determined based on the statistical analysis of historical anomaly data of latent factors. When Δfk(t)≥Δ_th, it is determined that factor fk has an anomaly, and the anomaly factor identifier is recorded. The influence of the latent factor fk on each core scenario is calculated using the following formula: I(fk,Sj)=I_fk×C_kj×e^(-τ_kj×ε), where I_fk is the influence of factor fk, C_kj is the coupling degree between fk and scenario Sj, τ_kj is the lag time from the change of fk to the change of indicators in scenario Sj, and ε is the influence attenuation coefficient, determined based on the lag characteristics of the latent factor influence; the value of I(fk,Sj) ranges from [0,1], and the larger the value, the more significant the influence of fk on scenario Sj; Based on the influence I(fk,Sj), a logistic regression model is used to predict the probability of scenario Sj becoming abnormal. The model formula is: P(Sj abnormal|fk abnormal)=1 / [1+e^(-(a1×I(fk,Sj)+b1))], where a1 is the influence coefficient and b1 is the bias term, both of which are trained based on the corresponding data of historical latent factor abnormalities and scenario abnormalities. An abnormality probability threshold P0 is set, and scenarios with P(Sj abnormal|fk abnormal)≥P0 are selected as potential abnormal scenarios. Starting with the latent abnormal factor fk, based on the cross-scene association strength matrix R, a reverse path search algorithm is used to mine the transmission path from fk to each potential abnormal scene, calculate the corresponding path weight W_path, and select the path with path weight W_path≥W2_th as the effective reverse chain transmission path, where W2_th is the reverse path weight threshold. The path form is represented as: latent factor fk → intermediate association factor → ... → potential abnormal scene Sj; By integrating forward and reverse chain propagation paths, a multi-scenario chain propagation directed graph G=(V,E) is constructed, where V is the set of nodes (including abnormal scenarios, potential abnormal scenarios, and related factors), and E is the set of directed edges (corresponding to the propagation relationship between factors and scenarios, and between factors). The propagation strength of each directed edge is calculated using the formula: S(e_pq)=W_path×(1-τ_pq / τ_max), where τ_max is the maximum lag time of all propagation paths. The quantified chain propagation paths are output, including key information such as path nodes, directed edges, propagation strength, and lag time, providing data support for S3 to generate initial operation and maintenance action plans.

[0007] Furthermore, S3 includes the following: Based on the constructed multi-scenario chain propagation directed graph G=(V,E) and the effective forward and reverse chain propagation paths, the initial operation and maintenance action plan is generated according to the following logic: For each effective transmission path, the preceding factors corresponding to the directed edges with transmission strength S(e_pq) ≥ S0 are extracted as key control nodes. S0 is the control node screening threshold, determined based on the control efficiency statistics of historical operation and maintenance actions. Key control nodes belong to the core driving layer or the scenario linkage layer and their contribution to the path weight W_path is ≥ 30%. The contribution percentage is calculated as follows: Contrib(fp) = (r_pq × I_fp) / W_path × 100%, where r_pq is the correlation strength between the key node fp and the subsequent factor fq, and I_fp is the influence of fp. A standardized operation and maintenance action library is constructed, which includes four basic actions: equipment parameter control, operation strategy adjustment, environmental intervention, and spatial resource scheduling. Each action is associated with attributes such as "control factor type, control range, execution cost, and historical execution effect." For each key control node fp, candidate actions in the action library that match the control object of fp are matched, and the action adaptability score is calculated. The formula is: S_A(fp,A)=λ1×R_reg(A,fp)×λ2×(1-Cost(A) / Cost_max)×λ3×I_fp, where A is the candidate action, and R_reg(A,fp) is the control coefficient of action A on factor fp, which is obtained by training based on historical action execution data. The value range is [0,1]; Cost(A) is the single execution cost of action A, and Cost_max is the maximum execution cost of similar actions; λ1, λ2, and λ3 are weight coefficients that satisfy λ1+λ2+λ3=1 and are dynamically determined based on the operation and maintenance goals; candidate actions with S_A(fp,A)≥A0 are selected, where A0 is the action adaptability threshold and is determined based on the historical action success rate statistics; according to the principle that each path matches at least one key node action and cross-path actions do not conflict, an initial operation and maintenance action scheme set {U1,U2,...,Us} is formed, where s is the number of initial schemes, and each scheme Ug contains one set of related actions and the corresponding execution order and parameter settings, and g takes values ​​from 1 to s; The initial set of operation and maintenance action plans {U1,U2,...,Us} is input into the building digital twin model one by one. Simultaneously, a three-dimensional collaborative verification is conducted, assessing the improvement effect of anomalies, the impact of related scenarios, and the dynamic changes in the transmission path. The verification period is consistent with the maximum lag time τ_max of the transmission path. After simulating the execution of plan Ug, the improvement rate of core indicators for anomalies (including both existing and potential anomalies) is calculated using the formula: R_imp(Sj,Ug)=|hj_pre-hj_sim| / hj_pre×100%, where Sj is an anomaly or potential anomaly scenario, hj_pre is the anomaly value of the core indicator of scenario Sj before plan execution (real-time monitoring value for existing anomalies, predicted anomaly value for potential anomalies), and hj_sim is the steady-state value of the core indicator of scenario Sj after the digital twin model simulates the execution of Ug. When R_imp(Sj,Ug)≥R_imp_th, the improvement effect of anomaly scenario Sj is deemed to have met the standard. R_imp_th is the minimum improvement effect threshold, determined based on industry operation and maintenance standards. Extract non-abnormal scenarios associated with effective transmission paths, and calculate the impact coefficient of scheme Ug on the core indicators of the associated scenarios. The formula is: Influ(Sc,Ug)=|hc_ori-hc_sim| / σc, where Sc is the associated scenario, hc_ori is the normal mean value of the core indicator of scenario Sc before scheme execution, hc_sim is the core indicator value after simulated execution of Ug, and σc is the historical data standard deviation of the core indicator of scenario Sc. When Influ(Sc,Ug)≤Influ_th, the impact is considered acceptable, where Influ_th is the impact threshold of the associated scenario, and the allowable fluctuation range of the operation and maintenance indicators based on the associated scenario is determined. After simulating the execution scheme Ug, the directed edge propagation strength S'(e_pq) of the multi-scenario chain propagation directed graph G is recalculated. The formula is: S'(e_pq)=W'_path×(1-τ_pq / τ_max), where W'_path is the updated weight of the path after simulation, recalculated based on the cross-scenario factor association strength matrix R' after simulation, and the calculation logic of R' is the same as the construction logic of R in S2. The path decay rate is calculated. The formula is: Decay(path,Ug)=|W_path-W'_path| / W_path×100%. When Decay(path,Ug)≥Decay_th, the path is considered to be effectively suppressed. Decay_th is the path decay threshold, which is statistically determined based on the operational risk control requirements. For each initial operational action scheme Ug, the three-dimensional verification results are integrated to form a verification report. The verification report includes: R_imp(Sj,Ug), Influ(Sc,Ug), and Decay(path,Ug) and the corresponding judgment results.

[0008] Furthermore, S4 includes the following: Based on the S3 verification report, a three-dimensional quantitative evaluation is performed on the initial set of operation and maintenance action plans {U1, U2, ..., Us}. The specific formula is as follows: Score(Ug)=r1×(N_imp_pass / N_imp_total)+r2×(N_influ_accept / N_influ_total)+r3×(N_decay_eff / N_decay_total), Wherein, N_imp_pass is the number of abnormal or potential abnormal scenarios that achieve the improvement effect, N_imp_total is the total number of abnormal or potential abnormal scenarios, N_influ_accept is the number of related scenarios with acceptable impact, N_influ_total is the total number of related scenarios, N_decay_eff is the number of transmission paths that are effectively suppressed, N_decay_total is the total number of effective transmission paths, r1, r2, and r3 are weight coefficients, and r1+r2+r3=1, dynamically determined based on operation and maintenance priority; scoring thresholds Score1 and Score2 are set. When Score(Ug)≥Score1, it is marked as the optimal initial solution; when Score2<Score(Ug)<Score1, it is marked as the initial solution to be optimized; when Score(Ug)≤Score2, it is marked as the eliminated solution; For the initial solution to be optimized, optimizations were made based on the non-compliance items in the validation report: For scenarios where R_imp(Sj,Ug) < R_imp_th, calculate the improvement gap ΔR_imp = R_imp_th - R_imp(Sj,Ug), and prioritize adjusting the control parameters of the corresponding key node actions; if the parameters still do not meet the requirements after adjustment, replace them with candidate actions with higher adaptability scores in the action library. For paths where Decay(path,Ug) < Decay_th, identify secondary key nodes in the path that do not match actions, supplement the action library with targeted actions that have an adaptability score ≥ A0 × 0.9, incorporate them into the plan, and set a reasonable execution order. For related scenarios Sc where Influ_th×0.8≤Influ(Sc,Ug)≤Influ_th, supplementary adaptation measures are implemented; the adaptation necessity Nec(Sc,Ug)=Influ(Sc,Ug) / Influ_th×I_Sc is calculated, where I_Sc is the importance of the scenario, and scenarios where Nec(Sc,Ug)≥Nec_th are selected, where Nec_th is the adaptation necessity threshold; measures in the adaptation measure library with an impact factor and Sc core indicator correlation degree ≥0.6 and an adaptation effect coefficient ≥0.9 are matched and included in the plan to avoid the impact of related scenarios exceeding the limit; The optimized core actions, supplementary suppression actions, and related scenario adaptation measures are integrated into a candidate final solution U_final according to their priority. For U_final, parameter conflicts (adjustment parameters do not exceed the equipment's rated range), resource conflicts (total resource consumption ≤ supply limit), and timing conflicts (compliant with operational physical logic) are verified. If conflicts exist, adjustments are made according to the core action priority principle until the verification passes. The verified U_final is output to the building operation and maintenance execution system. The solution includes an action execution list, triggering conditions, and execution time limits. The system provides real-time feedback on the execution effect. If the actual improvement rate does not reach R_imp_th×0.95, a secondary optimization process is triggered.

[0009] A building operation and maintenance intelligent decision-making system based on digital twins includes: a correlation factor hierarchical library construction module, a two-way causal tracing and transmission analysis module, an initial operation and maintenance plan generation and collaborative verification module, a decision plan optimization and execution module, and a data storage and synchronization module; The associated factor hierarchy library construction module is used to define the core scenarios of building operation and maintenance, filter associated factors, divide factor hierarchy and configure linkage trigger conditions to ensure that the scenario boundaries do not overlap and the factor hierarchy adapts to the operation and maintenance needs. The two-way causal tracing and transmission analysis module is used to access real-time operation and maintenance data, determine scene anomalies, trace the direct related factors of the abnormal scene and the cross-scene transmission path in the forward direction, and deduce the potential anomalies and transmission paths caused by the movement of hidden factors in the reverse direction. The initial operation and maintenance plan generation and collaborative verification module is used to extract key control nodes based on the transmission path, match candidate actions to form an initial plan, and carry out multi-dimensional collaborative verification in the digital twin model; The decision optimization and execution module is used to quantitatively evaluate the initial plan based on the verification results, optimize non-compliant items, supplement adaptation measures, output the final plan after conflict verification, and provide feedback on the execution effect. The data storage and synchronization module is used to store relevant data, support dynamic data maintenance, and achieve real-time data synchronization with the digital twin model.

[0010] Furthermore, the associated factor hierarchy library construction module includes a scenario definition and factor screening unit and a hierarchy division and threshold configuration unit; The scenario definition and factor screening unit defines core scenarios based on building functions and operation and maintenance goals, clarifies scenario boundaries through quantitative parameter sets, and uses professional analysis methods to screen related factors that are associated with the core indicators of the scenario. The hierarchical division and threshold configuration unit constructs a factor quantification model, classifies candidate factors into corresponding levels, configures the trigger thresholds and collection frequency adjustment rules for factors at each level, and establishes a dynamic threshold update mechanism.

[0011] Furthermore, the two-way causal tracing and transmission analysis module includes an anomaly detection and forward tracing unit and a latent factor anomaly analysis and reverse tracing unit; The anomaly detection and forward tracing unit accesses real-time operation and maintenance data, uses anomaly detection algorithms to identify abnormal scenarios, calculates the correlation strength between correlation factors and abnormal indicators, and mines and verifies cross-scenario positive transmission paths. The latent factor anomaly analysis and reverse tracing unit monitors the status of latent influence layer factors, determines factor anomalies, calculates the impact of anomaly factors on each scenario, predicts potential abnormal scenarios, and uncovers reverse transmission paths.

[0012] Furthermore, the initial operation and maintenance plan generation and collaborative verification module includes an initial plan generation unit and a multi-dimensional collaborative verification unit; The initial scheme generation unit extracts key control nodes based on the transmission path, matches suitable candidate actions from the standardized action library, and combines them into an initial set of operation and maintenance action schemes after scoring and screening. The multi-dimensional collaborative verification unit inputs the initial solution into the digital twin model, simulates the solution execution process, verifies the effect of anomaly improvement, the impact of related scenarios, and the suppression of transmission paths, and integrates the verification results to form a report.

[0013] Furthermore, the decision optimization and execution module includes a scheme optimization unit and a scheme verification and execution unit; The solution optimization unit performs quantitative evaluation and classification of the initial solution, adjusts action parameters or adds targeted actions for non-compliant items, and supplements adaptation measures for related scenarios. The scheme verification and execution unit integrates and optimizes the actions according to priority, performs parameter, resource and timing conflict verification, outputs the verified scheme to the execution system, provides real-time feedback on the execution effect and triggers secondary optimization.

[0014] Compared with existing technologies, the beneficial effects of this invention are as follows: By establishing a three-layer hierarchical library of related factors and a linked data collection logic, factors are categorized hierarchically according to core drivers, scenario linkages, and latent influences. Dynamic thresholds trigger hierarchical data collection, clarifying the influence priority of different factors and achieving targeted and efficient data collection. This solves the problems of missing related factor systems and low analytical accuracy in existing technologies, laying a precise data foundation for cross-scenario correlation analysis. The innovative bidirectional causal tracing and cross-scenario transmission path analysis logic traces directly related factors and chain transmission paths in abnormal scenarios, and reversely derives potential anomalies and transmission paths caused by latent factor anomalies. Combined with directed graph quantification to represent the entire link relationship, this invention breaks through the limitations of single-direction correlation mining in existing technologies, achieving full coverage of anomaly tracing and potential risk prediction, and avoiding the drawback of "solving a single problem leading to secondary anomalies." Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a building operation and maintenance intelligent decision-making system based on digital twins according to the present invention. Detailed Implementation

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

[0017] Please see Figure 1 The present invention provides the following technical solution: A building operation and maintenance intelligent decision-making system based on digital twins includes: a correlation factor hierarchical library construction module, a two-way causal tracing and transmission analysis module, an initial operation and maintenance plan generation and collaborative verification module, a decision plan optimization and execution module, and a data storage and synchronization module; The associated factor hierarchy library construction module is used to define the core scenarios of building operation and maintenance, filter associated factors, divide factor hierarchy and configure linkage trigger conditions to ensure that the scenario boundaries do not overlap and the factor hierarchy adapts to the operation and maintenance needs. The two-way causal tracing and transmission analysis module is used to access real-time operation and maintenance data, determine scene anomalies, trace the direct related factors of the abnormal scene and the cross-scene transmission path in the forward direction, and deduce the potential anomalies and transmission paths caused by the movement of hidden factors in the reverse direction. The initial operation and maintenance plan generation and collaborative verification module is used to extract key control nodes based on the transmission path, match candidate actions to form an initial plan, and carry out multi-dimensional collaborative verification in the digital twin model; The decision optimization and execution module is used to quantitatively evaluate the initial plan based on the verification results, optimize non-compliant items, supplement adaptation measures, output the final plan after conflict verification, and provide feedback on the execution effect. The data storage and synchronization module is used to store relevant data, support dynamic data maintenance, and achieve real-time data synchronization with the digital twin model.

[0018] The module for building a hierarchical library of related factors includes a scenario definition and factor selection unit and a hierarchy division and threshold configuration unit. The scenario definition and factor screening unit defines core scenarios based on building functions and operation and maintenance goals, clarifies scenario boundaries through quantitative parameter sets, and uses professional analysis methods to screen related factors that are associated with the core indicators of the scenario. The hierarchical division and threshold configuration unit constructs a factor quantification model, classifies candidate factors into corresponding levels, configures the trigger thresholds and collection frequency adjustment rules for factors at each level, and establishes a dynamic threshold update mechanism.

[0019] The two-way causal tracing and transmission analysis module includes an anomaly detection and forward tracing unit and a latent factor anomaly analysis and reverse tracing unit; The anomaly detection and forward tracing unit accesses real-time operation and maintenance data, uses anomaly detection algorithms to identify abnormal scenarios, calculates the correlation strength between correlation factors and abnormal indicators, and mines and verifies cross-scenario positive transmission paths. The latent factor anomaly analysis and reverse tracing unit monitors the status of latent influence layer factors, determines factor anomalies, calculates the impact of anomaly factors on each scenario, predicts potential abnormal scenarios, and uncovers reverse transmission paths.

[0020] The initial operation and maintenance plan generation and collaborative verification module includes an initial plan generation unit and a multi-dimensional collaborative verification unit; The initial scheme generation unit extracts key control nodes based on the transmission path, matches suitable candidate actions from the standardized action library, and combines them into an initial set of operation and maintenance action schemes after scoring and screening. The multi-dimensional collaborative verification unit inputs the initial solution into the digital twin model, simulates the solution execution process, verifies the effect of anomaly improvement, the impact of related scenarios, and the suppression of transmission paths, and integrates the verification results to form a report.

[0021] The decision optimization and execution module includes a scheme optimization unit and a scheme verification and execution unit; The solution optimization unit performs quantitative evaluation and classification of the initial solution, adjusts action parameters or adds targeted actions for non-compliant items, and supplements adaptation measures for related scenarios. The scheme verification and execution unit integrates and optimizes the actions according to priority, performs parameter, resource and timing conflict verification, outputs the verified scheme to the execution system, provides real-time feedback on the execution effect and triggers secondary optimization.

[0022] A building operation and maintenance intelligent decision-making method based on digital twins includes the following steps: S1. Construct a hierarchical library of multi-scenario related factors for building operation and maintenance. Store the related factors corresponding to the core scenarios of building operation and maintenance in layers: core driving layer, scenario linkage layer, and implicit influence layer. Configure linkage trigger conditions for each layer of factors. When the core driving layer factor exceeds the corresponding preset threshold, the data collection of the scenario linkage layer factor is automatically activated. When the scenario linkage layer factor exceeds the corresponding preset threshold, the data collection of the implicit influence layer factor is automatically activated. S2. Based on the building digital twin model, real-time operation and maintenance data is obtained. When any scene anomaly is detected, bidirectional causal tracing and multi-scene chain transmission analysis are initiated. The direct related factors and cross-scene chain transmission paths corresponding to the abnormal scene are traced in the forward direction, while the multi-scene chain anomalies and transmission paths caused by the change of hidden influence layer factors are deduced in the reverse direction. S3. Generate an initial operation and maintenance action plan based on the traced chain transmission path, conduct multi-scenario collaborative verification in the digital twin model, and simultaneously simulate the improvement effect of the initial plan on abnormal scenarios, the potential impact on related scenarios, and the dynamic changes of the chain transmission path. S4. Optimize the initial operation and maintenance action plan based on the collaborative verification results, supplement the adaptation measures of related scenarios, form the final decision plan and output it to the building operation and maintenance execution system for execution, so as to complete the anomaly handling and multi-scenario collaborative control.

[0023] S1 includes the following: Based on the building's functional positioning and core operation and maintenance objectives, core scenarios are defined. Scenario boundaries are defined using a three-dimensional quantitative parameter set Ωi, where Ωi = {geographic coordinate range Ai, associated device identifier set Di, core indicator set Mi}, where i is the scenario number, Ai is represented by the building's rectangular coordinates and floor intervals, Di is the unique code set of the physical devices associated with the scenario, and Mi is the core operation and maintenance indicator for the scenario, determined by a variance contribution rate ≥ η, where η is a screening threshold determined by industry statistics. An intersection verification formula ensures that the boundaries of each scenario do not overlap; that is, for any i ≠ j, [Ai ∩ Aj = ∅] ∧ [Di ∩ Dj = ∅] ∧ [the proportion of indicators in Mi ∩ Mj ≤ γ], where γ is the indicator overlap rate constraint threshold. If this condition is not met, Ai or Di is adjusted until the requirement is met. In this embodiment, assuming an energy management scenario (i=1) is selected, its three-dimensional quantization parameter set is represented as: Ω1={A1, {A1, D1, M1}, where A1=(x1_min,x1_max,y1_min,y1_max,z1_min,z1_max), D1={equipment codes D101-D120}, M1={hourly energy consumption per unit area, daily average energy consumption fluctuation coefficient}; for the energy management scenario (i=1) and the equipment fault early warning scenario (i=2), after verification, A1∩A2 is an empty set, the proportion of equipment in D1∩D2 is ≤5%, and M1∩M2 is an empty set, which meets the preset constraints.

[0024] Based on historical building operation and maintenance data, mutual information analysis combined with significance test is used to screen candidate factors; the correlation strength between candidate factors and core indicators of the scenario is calculated by mutual information formula, and after significance test, factors that meet the correlation strength threshold are screened to form a candidate set of correlation factors Fi for each scenario. In this embodiment, it is assumed that 12 months of energy consumption and operation data of the building are used, with a sampling frequency of 1 time / hour and data integrity of 96%. 35 potential factors such as equipment operating load, ambient temperature and humidity, power supply voltage, and light intensity are selected. The correlation strength between each factor and "hourly energy consumption per unit area" is calculated using the mutual information formula MI(X,Y)=E[log(p(x,y) / (p(x)p(y)))]. The standard error is calculated by Bootstrap sampling (1000 times) and a t-test is performed (significance level tα=0.05). 18 factors with MI(X,Y)≥0.3 and t>tα / 2 are selected to form a candidate set F1={f1-f18}.

[0025] A two-dimensional quantitative model of factor influence and response speed is constructed. The factor influence If is calculated by combining the partial correlation coefficient with the variance contribution, and the factor response speed Vf is calculated based on the cross-correlation function. A quantitative threshold range is set, and according to the numerical distribution of If and Vf, the candidate factors are respectively assigned to the core driving layer F1_i, the scene linkage layer F2_i, and the latent influence layer F3_i, forming a three-layer set of related factors for each scene. In this embodiment, the partial correlation coefficient is combined with the variance contribution to calculate the value. The formula is If=|rp(X,Y)|×ω(X), where rp(X,Y) is the partial correlation coefficient between the candidate factor X and the core indicator Y of the scenario, and ω(X) is the variance contribution of X to Y (calculated by a multiple linear regression model). The lag time from factor anomaly to scene index change is calculated based on the cross-correlation function. The formula is CCF(τ)=Cov(X(t),Y(t+τ)) / (Var(X(t))Var(Y(t+τ)))^(1 / 2), where τ is the lag time, Cov() is the covariance, and Var() is the variance. The τ corresponding to the maximum value of CCF(τ) is taken as the lag time τf, and the response speed Vf=1 / τf. The specific two-dimensional threshold matrix is ​​set based on the statistical distribution of industry operation and maintenance data as follows: The core driving layer is If≥I1 and Vf≥V1, the scene linkage layer is I2≤If<I1 and V2≤Vf<V1, and the latent influence layer is If<I2 and Vf<V2, where I1>I2>0 and V1>V2>0 are quantization thresholds; based on these thresholds, a three-layer set of correlation factors for each scene is formed: core driving layer F1_i, scene linkage layer F2_i, and latent influence layer F3_i; Based on historical data statistical characteristics and operational safety constraints, the 3σ principle combined with dynamic safety margin is adopted. The formula is T1_j=μ1_j+3σ1_j+α·CV1_j, where μ1_j is the historical data mean of the j-th core driving layer factor, σ1_j is the standard deviation, CV1_j=σ1_j / μ1_j is the coefficient of variation, and α is the safety margin coefficient, which is determined based on the equipment rated parameter range. α=α0+α1×(1-equipment rated parameter utilization rate), where α0 and α1 are constant coefficients. The quantile method combined with cross-scenario coupling is adopted, and the formula is T2_m=Q3_m+k1×IQR_m+β·C_m, where Q3_m is the upper quartile, Q1_m is the lower quartile, IQR_m=Q3_m-Q1_m is the interquartile range, k1 is the quantile adjustment coefficient, C_m is the coupling degree of the factor with other scenarios, which is calculated by the scenario correlation matrix C_m=∑|r(Xm,Yn)| / n, where n is the number of other scenarios, r(Xm,Yn) is the correlation coefficient of factor Xm with the core indicator Yn of other scenarios, β is the coupling coefficient, and β=β0+β1×C_m, where β0 and β1 are constant coefficients; When the real-time monitoring value x1_j of the core driving layer factor f1_j is greater than T1_j, the factor data acquisition of the corresponding scene linkage layer F2_i is automatically activated, and the acquisition frequency is adjusted from the basic frequency f0 to f2, and f2=f0×(1+a×(x1_j-T1_j) / T1_j), where a is the frequency adjustment coefficient, and the value range of f2 is f0×b~f0×c, where b and c are frequency constraint thresholds; when the real-time monitoring value x2_m of the scene linkage layer factor f2_m is greater than T2_m, the factor data acquisition of the corresponding latent influence layer F3_i is automatically activated, and the acquisition frequency is adjusted to f3, and f3=f0×(1+d×(x2_m-T2_m) / T2_m), where d is the frequency adjustment coefficient, and the value range of f3 is f0×e~f0×f, where e and f are frequency constraint thresholds; Set a data freshness threshold θ0, and set data freshness θ = proportion of data in the last Δt1 period × w1 + data integrity × w2, where w1 and w2 are weighting coefficients, and w1 + w2 = 1, determined based on the timeliness and reliability requirements of operation and maintenance data; proportion of data in the last Δt1 period = actual data volume collected in the last Δt1 period / data volume to be collected in the last Δt1 period, where Δt1 is the statistical period for data freshness; data integrity = effective data volume / total data volume to be collected, where effective data volume is the amount of data collected that meets the requirement of "data values ​​within a preset reasonable range and without logical conflicts", and total data volume to be collected = collection frequency × statistical period, with data integrity values ​​ranging from [0,1]; trigger threshold update when θ < θ0, with threshold update period Δt = Δt0 × (1 + CV_total), where Δt0 is the basic update period, and CV_total is the mean of the coefficients of variation of all core driving layer factors; recalculate T1_j and T2_m based on historical data in the last Δt period, and update the trigger condition parameters; Data is stored using a distributed time-series database. The database structure includes a table of basic scene information, a table of factor attributes, and a table of trigger conditions. It supports dynamic maintenance of factors for addition and deletion, threshold modification, and scene relationship updates. Furthermore, the data interface between the database and the building digital twin model meets the requirements for real-time synchronization.

[0026] In this embodiment, the library structure includes a scene basic information table, a factor attribute table, and a trigger condition table, specifically: Scene basic information table: Fields include scene number (primary key), scene name, geographic coordinate range, associated device identifier set, core indicator set, creation time, and update time; Factor attribute table: Fields include factor number (primary key), factor name, scene number (foreign key), level affiliation, data type, acquisition device ID, basic acquisition frequency, and data precision; Triggering Condition Table: Fields include factor number (foreign key), threshold parameter (T1_j / T2_m), frequency adjustment coefficient, activation status (0=inactive, 1=activated), and threshold update time; the database supports dynamic maintenance of factor addition and deletion, threshold modification, and scene association updates, and the data interface with the building digital twin model meets the real-time synchronization requirements, that is, data transmission delay ≤ t_delay, and data consistency ≥ σ_consist.

[0027] S2 includes the following: Based on the real-time data interface of the building digital twin model, operation and maintenance data of each core scenario are accessed. The data types include real-time monitoring values ​​of activated factors in the core driving layer and scenario linkage layer. The data sampling frequency is consistent with the collection frequency set in S1. For any core scenario i, based on the corresponding core indicator set Mi, the abnormality deviation formula is used to determine whether the scenario is abnormal. The formula is: Li(t)=|xi_t-μ_Mi| / σ_Mi, where xi_t is the real-time monitoring value of the core indicator of scenario i at time t, μ_Mi is the historical average of the core indicator of scenario i, and σ_Mi is the historical standard deviation of the core indicator of scenario i. An abnormality judgment threshold L0 is set. L0 is determined based on the abnormal statistical data of similar building scenarios in the industry (satisfying the abnormality identification accuracy at a confidence level of 95%). When Li(t)≥L0, scenario i is determined to be abnormal at time t, and the abnormal scenario identifier and the abnormality occurrence time t0 are recorded. Based on the three-layer correlation factor set constructed by S1, the correlation factors of the core driving layer F1_i and the scene linkage layer F2_i corresponding to the abnormal scene i are extracted. The real-time correlation strength between each correlation factor and the core abnormal indicator is calculated. The calculation formula is: RI(f,Yi)=MI(f,Yi)×Vf, where MI(f,Yi) is the mutual information value between the correlation factor f and the core indicator Yi of scene i, and Vf is the response speed of the correlation factor f. Factors with RI(f,Yi)≥RI_th are selected as direct correlation factors, where RI_th is the correlation strength screening threshold, which is determined based on the factor contribution statistics of historical abnormal data. Starting with the abnormal scenario i, extract the three-layer correlation factors of all core scenarios, construct the cross-scenario factor correlation strength matrix R, and R=[r_pq] n×n Where n is the total number of correlation factors across all scenarios, and r_pq is the cross-scenario correlation strength between factor fp and factor fq, calculated as: r_pq=|r_corr(fp,fq)|×e^(-τ_pq×δ), where r_corr(fp,fq) is the Pearson correlation coefficient between factor fp and fq, τ_pq is the lag time from fp change to fq response, calculated based on the cross-correlation function; δ is the time decay coefficient, determined based on the timeliness requirements of operation and maintenance data; and the value range of r_pq is [0,1], the larger the value, the more significant the cross-scenario correlation between factors; Based on the correlation strength matrix R, a path search algorithm is used to mine cross-scene transmission paths. The path weight calculation formula is: W_path=∑(r_pq×I_fp), where I_fp is the influence degree of factor fp in the path, and ∑ is the sum of the products of the correlation strength of all adjacent factor pairs in the path and the influence degree of the preceding factor. Paths with W_path≥W1_th are selected as effective positive chain transmission paths, where W1_th is the positive path weight threshold, determined based on the statistical analysis of the abnormal contribution of historical transmission paths. The path form is represented as: abnormal scenario i → directly related factor → cross-scene factor 1 → cross-scene factor 2 → ... → related scenario j. The effective positive chain transmission paths are mapped to the building digital twin model. By verifying the correlation between the three-dimensional spatial coordinates of the model and the equipment association relationship, the correlation of the physical equipment and spatial location corresponding to the factors in the path is verified, ensuring that the transmission path conforms to the physical logic of building operation and maintenance. Extract the real-time monitoring values ​​of the latent influence layer F3_i factor from all scenarios, and calculate the degree of factor anomaly using the formula: Δfk(t)=|x_k(t)-μ_fk| / μ_fk, where x_k(t) is the real-time monitoring value of the latent influence layer factor fk at time t, and μ_fk is the historical average value of the latent influence layer factor fk. Set an anomaly judgment threshold Δ_th, which is determined based on the statistical analysis of historical anomaly data of latent factors. When Δfk(t)≥Δ_th, it is determined that factor fk has an anomaly, and the anomaly factor identifier is recorded. The influence of the latent factor fk on each core scenario is calculated using the following formula: I(fk,Sj)=I_fk×C_kj×e^(-τ_kj×ε), where I_fk is the influence of factor fk, C_kj is the coupling degree between fk and scenario Sj, τ_kj is the lag time from the change of fk to the change of indicators in scenario Sj, and ε is the influence attenuation coefficient, determined based on the lag characteristics of the latent factor influence; the value of I(fk,Sj) ranges from [0,1], and the larger the value, the more significant the influence of fk on scenario Sj; Based on the influence I(fk,Sj), a logistic regression model is used to predict the probability of scenario Sj becoming abnormal. The model formula is: P(Sj abnormal|fk abnormal)=1 / [1+e^(-(a1×I(fk,Sj)+b1))], where a1 is the influence coefficient and b1 is the bias term, both of which are trained based on the corresponding data of historical latent factor abnormalities and scenario abnormalities. An abnormality probability threshold P0 is set, and scenarios with P(Sj abnormal|fk abnormal)≥P0 are selected as potential abnormal scenarios. Starting with the latent abnormal factor fk, based on the cross-scene association strength matrix R, a reverse path search algorithm is used to mine the transmission path from fk to each potential abnormal scene, calculate the corresponding path weight W_path, and select the path with path weight W_path≥W2_th as the effective reverse chain transmission path, where W2_th is the reverse path weight threshold. The path form is represented as: latent factor fk → intermediate association factor → ... → potential abnormal scene Sj; By integrating forward and reverse chain propagation paths, a multi-scenario chain propagation directed graph G=(V,E) is constructed, where V is the set of nodes (including abnormal scenarios, potential abnormal scenarios, and related factors), and E is the set of directed edges (corresponding to the propagation relationship between factors and scenarios, and between factors). The propagation strength of each directed edge is calculated using the formula: S(e_pq)=W_path×(1-τ_pq / τ_max), where τ_max is the maximum lag time of all propagation paths. The quantified chain propagation paths are output, including key information such as path nodes, directed edges, propagation strength, and lag time, providing data support for S3 to generate initial operation and maintenance action plans.

[0028] S3 includes the following: Based on the constructed multi-scenario chain propagation directed graph G=(V,E) and the effective forward and reverse chain propagation paths, the initial operation and maintenance action plan is generated according to the following logic: For each effective transmission path, the preceding factors corresponding to the directed edges with transmission strength S(e_pq) ≥ S0 are extracted as key control nodes. S0 is the control node screening threshold, determined based on the control efficiency statistics of historical operation and maintenance actions. Key control nodes belong to the core driving layer or the scenario linkage layer and their contribution to the path weight W_path is ≥ 30%. The contribution percentage is calculated as follows: Contrib(fp) = (r_pq × I_fp) / W_path × 100%, where r_pq is the correlation strength between the key node fp and the subsequent factor fq, and I_fp is the influence of fp. A standardized operation and maintenance action library is constructed, which includes four basic actions: equipment parameter control, operation strategy adjustment, environmental intervention, and spatial resource scheduling. Each action is associated with attributes such as "control factor type, control range, execution cost, and historical execution effect." For each key control node fp, candidate actions in the action library that match the control object of fp are matched, and the action adaptability score is calculated. The formula is: S_A(fp,A)=λ1×R_reg(A,fp)×λ2×(1-Cost(A) / Cost_max)×λ3×I_fp, where A is the candidate action, and R_reg(A,fp) is the control coefficient of action A on factor fp, which is obtained by training based on historical action execution data. The value range is [0,1]; Cost(A) is the single execution cost of action A, and Cost_max is the maximum execution cost of similar actions; λ1, λ2, and λ3 are weight coefficients that satisfy λ1+λ2+λ3=1 and are dynamically determined based on the operation and maintenance goals; candidate actions with S_A(fp,A)≥A0 are selected, where A0 is the action adaptability threshold and is determined based on the historical action success rate statistics; according to the principle that each path matches at least one key node action and cross-path actions do not conflict, an initial operation and maintenance action scheme set {U1,U2,...,Us} is formed, where s is the number of initial schemes, and each scheme Ug contains one set of related actions and the corresponding execution order and parameter settings, and g takes values ​​from 1 to s; The initial set of operation and maintenance action plans {U1,U2,...,Us} is input into the building digital twin model one by one. Simultaneously, a three-dimensional collaborative verification is conducted, assessing the improvement effect of anomalies, the impact of related scenarios, and the dynamic changes in the transmission path. The verification period is consistent with the maximum lag time τ_max of the transmission path. After simulating the execution of plan Ug, the improvement rate of core indicators for anomalies (including both existing and potential anomalies) is calculated using the formula: R_imp(Sj,Ug)=|hj_pre-hj_sim| / hj_pre×100%, where Sj is an anomaly or potential anomaly scenario, hj_pre is the anomaly value of the core indicator of scenario Sj before plan execution (real-time monitoring value for existing anomalies, predicted anomaly value for potential anomalies), and hj_sim is the steady-state value of the core indicator of scenario Sj after the digital twin model simulates the execution of Ug. When R_imp(Sj,Ug)≥R_imp_th, the improvement effect of anomaly scenario Sj is deemed to have met the standard. R_imp_th is the minimum improvement effect threshold, determined based on industry operation and maintenance standards. Extract non-abnormal scenarios associated with effective transmission paths, and calculate the impact coefficient of scheme Ug on the core indicators of the associated scenarios. The formula is: Influ(Sc,Ug)=|hc_ori-hc_sim| / σc, where Sc is the associated scenario, hc_ori is the normal mean value of the core indicator of scenario Sc before scheme execution, hc_sim is the core indicator value after simulated execution of Ug, and σc is the historical data standard deviation of the core indicator of scenario Sc. When Influ(Sc,Ug)≤Influ_th, the impact is considered acceptable, where Influ_th is the impact threshold of the associated scenario, and the allowable fluctuation range of the operation and maintenance indicators based on the associated scenario is determined. After simulating the execution scheme Ug, the directed edge propagation strength S'(e_pq) of the multi-scenario chain propagation directed graph G is recalculated. The formula is: S'(e_pq)=W'_path×(1-τ_pq / τ_max), where W'_path is the updated weight of the path after simulation, recalculated based on the cross-scenario factor association strength matrix R' after simulation, and the calculation logic of R' is the same as the construction logic of R in S2. The path decay rate is calculated. The formula is: Decay(path,Ug)=|W_path-W'_path| / W_path×100%. When Decay(path,Ug)≥Decay_th, the path is considered to be effectively suppressed. Decay_th is the path decay threshold, which is statistically determined based on the operational risk control requirements. For each initial operational action scheme Ug, the three-dimensional verification results are integrated to form a verification report. The verification report includes: R_imp(Sj,Ug), Influ(Sc,Ug), and Decay(path,Ug) and the corresponding judgment results.

[0029] S4 includes the following: Based on the S3 verification report, a three-dimensional quantitative evaluation is performed on the initial set of operation and maintenance action plans {U1, U2, ..., Us}. The specific formula is as follows: Score(Ug)=r1×(N_imp_pass / N_imp_total)+r2×(N_influ_accept / N_influ_total)+r3×(N_decay_eff / N_decay_total), Wherein, N_imp_pass is the number of abnormal or potential abnormal scenarios that achieve the improvement effect, N_imp_total is the total number of abnormal or potential abnormal scenarios, N_influ_accept is the number of related scenarios with acceptable impact, N_influ_total is the total number of related scenarios, N_decay_eff is the number of transmission paths that are effectively suppressed, N_decay_total is the total number of effective transmission paths, r1, r2, and r3 are weight coefficients, and r1+r2+r3=1, dynamically determined based on operation and maintenance priority; scoring thresholds Score1 and Score2 are set. When Score(Ug)≥Score1, it is marked as the optimal initial solution; when Score2<Score(Ug)<Score1, it is marked as the initial solution to be optimized; when Score(Ug)≤Score2, it is marked as the eliminated solution; For the initial solution to be optimized, optimizations were made based on the non-compliance items in the validation report: For scenarios where R_imp(Sj,Ug) < R_imp_th, calculate the improvement gap ΔR_imp = R_imp_th - R_imp(Sj,Ug), and prioritize adjusting the control parameters of the corresponding key node actions; if the parameters still do not meet the requirements after adjustment, replace them with candidate actions with higher adaptability scores in the action library. For paths where Decay(path,Ug) < Decay_th, identify secondary key nodes in the path that do not match actions, supplement the action library with targeted actions that have an adaptability score ≥ A0 × 0.9, incorporate them into the plan, and set a reasonable execution order. For related scenarios Sc where Influ_th×0.8≤Influ(Sc,Ug)≤Influ_th, supplementary adaptation measures are implemented; the adaptation necessity Nec(Sc,Ug)=Influ(Sc,Ug) / Influ_th×I_Sc is calculated, where I_Sc is the importance of the scenario, and scenarios where Nec(Sc,Ug)≥Nec_th are selected, where Nec_th is the adaptation necessity threshold; measures in the adaptation measure library with an impact factor and Sc core indicator correlation degree ≥0.6 and an adaptation effect coefficient ≥0.9 are matched and included in the plan to avoid the impact of related scenarios exceeding the limit; The optimized core actions, supplementary suppression actions, and related scenario adaptation measures are integrated into a candidate final solution U_final according to their priority. For U_final, parameter conflicts (adjustment parameters do not exceed the equipment's rated range), resource conflicts (total resource consumption ≤ supply limit), and timing conflicts (compliant with operational physical logic) are verified. If conflicts exist, adjustments are made according to the core action priority principle until the verification passes. The verified U_final is output to the building operation and maintenance execution system. The solution includes an action execution list, triggering conditions, and execution time limits. The system provides real-time feedback on the execution effect. If the actual improvement rate does not reach R_imp_th×0.95, a secondary optimization process is triggered.

[0030] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0031] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A building operation and maintenance intelligent decision-making method based on digital twins, characterized in that: The method includes the following steps: S1. Construct a hierarchical library of multi-scenario related factors for building operation and maintenance. Store the related factors corresponding to the core scenarios of building operation and maintenance in layers: core driving layer, scenario linkage layer, and implicit influence layer. Configure linkage trigger conditions for each layer of factors. When the core driving layer factor exceeds the corresponding preset threshold, the data collection of the scenario linkage layer factor is automatically activated. When the scenario linkage layer factor exceeds the corresponding preset threshold, the data collection of the implicit influence layer factor is automatically activated. S2. Based on the building digital twin model, real-time operation and maintenance data is obtained. When any scene anomaly is detected, bidirectional causal tracing and multi-scene chain transmission analysis are initiated. The direct related factors and cross-scene chain transmission paths corresponding to the abnormal scene are traced in the forward direction, while the multi-scene chain anomalies and transmission paths caused by the change of hidden influence layer factors are deduced in the reverse direction. S3. Generate an initial operation and maintenance action plan based on the traced chain transmission path, conduct multi-scenario collaborative verification in the digital twin model, and simultaneously simulate the improvement effect of the initial plan on abnormal scenarios, the potential impact on related scenarios, and the dynamic changes of the chain transmission path. S4. Optimize the initial operation and maintenance action plan based on the collaborative verification results, supplement the adaptation measures of related scenarios, form the final decision plan and output it to the building operation and maintenance execution system for execution, so as to complete the anomaly handling and multi-scenario collaborative control.

2. The intelligent decision-making method for building operation and maintenance based on digital twins according to claim 1, characterized in that: S1 includes the following: Based on the building's functional positioning and core operation and maintenance objectives, core scenarios are defined. Scenario boundaries are defined using a three-dimensional quantitative parameter set Ωi, where Ωi = {geographic coordinate range Ai, associated device identifier set Di, core indicator set Mi}, where i is the scenario number, Ai is represented by the building's rectangular coordinates and floor intervals, Di is the unique code set of the physical devices associated with the scenario, and Mi is the core operation and maintenance indicator for the scenario, determined by a variance contribution rate ≥ η, where η is a screening threshold determined by industry statistics. An intersection verification formula ensures that the boundaries of each scenario do not overlap; that is, for any i ≠ j, [Ai ∩ Aj = ∅] ∧ [Di ∩ Dj = ∅] ∧ [the proportion of indicators in Mi ∩ Mj ≤ γ], where γ is the indicator overlap rate constraint threshold. If this condition is not met, Ai or Di is adjusted until the requirement is met. Based on historical building operation and maintenance data, mutual information analysis combined with significance test is used to screen candidate factors; the correlation strength between candidate factors and core indicators of the scenario is calculated by mutual information formula, and after significance test, factors that meet the correlation strength threshold are screened to form a candidate set of correlation factors Fi for each scenario. A two-dimensional quantitative model of factor influence and response speed is constructed. The factor influence If is calculated by combining the partial correlation coefficient with the variance contribution, and the factor response speed Vf is calculated based on the cross-correlation function. Set a quantization threshold range, and according to the numerical distribution of If and Vf, classify the candidate factors into the core driving layer F1_i, the scene linkage layer F2_i, and the implicit influence layer F3_i, respectively, to form a three-layer set of related factors for each scene. Based on historical data statistical characteristics and operational safety constraints, the 3σ principle combined with dynamic safety margin is adopted. The formula is T1_j=μ1_j+3σ1_j+α·CV1_j, where μ1_j is the historical data mean of the j-th core driving layer factor, σ1_j is the standard deviation, CV1_j=σ1_j / μ1_j is the coefficient of variation, and α is the safety margin coefficient, which is determined based on the equipment rated parameter range. α=α0+α1×(1-equipment rated parameter utilization rate), where α0 and α1 are constant coefficients. The quantile method combined with cross-scenario coupling is adopted, and the formula is T2_m=Q3_m+k1×IQR_m+β·C_m, where Q3_m is the upper quartile, Q1_m is the lower quartile, IQR_m=Q3_m-Q1_m is the interquartile range, k1 is the quantile adjustment coefficient, C_m is the coupling degree of the factor with other scenarios, which is calculated by the scenario correlation matrix C_m=∑|r(Xm,Yn)| / n, where n is the number of other scenarios, r(Xm,Yn) is the correlation coefficient of factor Xm with the core indicator Yn of other scenarios, β is the coupling coefficient, and β=β0+β1×C_m, where β0 and β1 are constant coefficients; When the real-time monitoring value x1_j of the core driving layer factor f1_j is greater than T1_j, the factor data acquisition of the corresponding scene linkage layer F2_i is automatically activated, and the acquisition frequency is adjusted from the basic frequency f0 to f2, and f2=f0×(1+a×(x1_j-T1_j) / T1_j), where a is the frequency adjustment coefficient, and the value range of f2 is f0×b~f0×c, where b and c are frequency constraint thresholds; when the real-time monitoring value x2_m of the scene linkage layer factor f2_m is greater than T2_m, the factor data acquisition of the corresponding latent influence layer F3_i is automatically activated, and the acquisition frequency is adjusted to f3, and f3=f0×(1+d×(x2_m-T2_m) / T2_m), where d is the frequency adjustment coefficient, and the value range of f3 is f0×e~f0×f, where e and f are frequency constraint thresholds; Set a data freshness threshold θ0, and data freshness θ = proportion of data in the most recent Δt1 period × w1 + data integrity × w2, where w1 and w2 are weighting coefficients, and w1 + w2 = 1; The percentage of data in the near Δt1 period = the actual amount of data collected in the near Δt1 period / the amount of data that should have been collected in the near Δt1 period, where Δt1 is the statistical period for data freshness. Data integrity = effective data volume / total data volume to be collected. The effective data volume is the amount of data collected that meets the condition that "data values ​​are within a preset reasonable range and there is no logical conflict". The total data volume to be collected = collection frequency × statistical period. The data integrity value range is [0,1]. When θ < θ0, the threshold update is triggered. The threshold update period Δt = Δt0 × (1 + CV_total), where Δt0 is the basic update period and CV_total is the mean of the coefficients of variation of all core driving layer factors. Recalculate T1_j and T2_m based on historical data from the near Δt period, and update the trigger condition parameters; Data is stored using a distributed time-series database. The database structure includes a table of basic scene information, a table of factor attributes, and a table of trigger conditions. It supports dynamic maintenance of factors for addition and deletion, threshold modification, and scene relationship updates. Furthermore, the data interface between the database and the building digital twin model meets the requirements for real-time synchronization.

3. The intelligent decision-making method for building operation and maintenance based on digital twins according to claim 2, characterized in that: S2 includes the following: Based on the real-time data interface of the building digital twin model, the operation and maintenance data of each core scenario are accessed. The data types include the real-time monitoring values ​​of activated factors in the core driving layer and the scenario linkage layer. The data sampling frequency is consistent with the collection frequency set by S1. For any core scenario i, based on the corresponding core indicator set Mi, the abnormality deviation formula is used to determine whether the scenario is abnormal. The formula is: Li(t)=|xi_t-μ_Mi| / σ_Mi, where xi_t is the real-time monitoring value of the core indicator of scenario i at time t, μ_Mi is the historical average of the core indicator of scenario i, and σ_Mi is the historical standard deviation of the core indicator of scenario i. An abnormality judgment threshold L0 is set. L0 is determined based on the abnormal statistical data of similar building scenarios in the industry. When Li(t)≥L0, it is determined that scenario i is abnormal at time t, and the abnormal scenario identifier and the time of occurrence of the abnormality t0 are recorded. Based on the three-layer correlation factor set constructed by S1, the correlation factors of the core driving layer F1_i and the scene linkage layer F2_i corresponding to the abnormal scene i are extracted. The real-time correlation strength between each correlation factor and the core abnormal indicator is calculated. The calculation formula is: RI(f,Yi)=MI(f,Yi)×Vf, where MI(f,Yi) is the mutual information value between the correlation factor f and the core indicator Yi of scene i, and Vf is the response speed of the correlation factor f. Factors with RI(f,Yi)≥RI_th are selected as direct correlation factors, where RI_th is the correlation strength screening threshold, which is determined based on the factor contribution statistics of historical abnormal data. With the abnormal scene i as the starting point, the three-layer correlation factors of all core scenes are extracted, and the cross-scene factor correlation strength matrix R is constructed, and R=[r_pq] n×n , wherein n is the total number of correlation factors of all scenes, r_pq is the cross-scene correlation strength of factor fp and factor fq, the calculation formula is: r_pq=|r_corr(fp,fq)|×e^(-τ_pq×δ), wherein r_corr(fp,fq) is the Pearson correlation coefficient of factor fp and fq, τ_pq is the lag time of fp abnormality to the response of fq; δ is the time decay coefficient, which is determined based on the timeliness requirements of operation and maintenance data; Based on the association strength matrix R, a path search algorithm is used to mine cross-scene transmission paths. The path weight calculation formula is: W_path=∑(r_pq×I_fp), where I_fp is the influence degree of factor fp in the path, and ∑ is the sum of the products of the association strength of all adjacent factor pairs in the path and the influence degree of the preceding factor. Paths with W_path≥W1_th are selected as effective positive chain transmission paths, where W1_th is the positive path weight threshold, determined based on the statistical analysis of the abnormal contribution of historical transmission paths. The path form is represented as: abnormal scene i → directly associated factor → cross-scene factor 1 → cross-scene factor 2 → ... → associated scene j. The effective positive chain transmission paths are mapped to the building digital twin model. The correlation between the model's three-dimensional spatial coordinates and the equipment association relationship is used to verify the correlation of the physical equipment and spatial location corresponding to the factors in the path. Extract the real-time monitoring values ​​of the latent influence layer F3_i factor from all scenarios, and calculate the degree of factor anomaly using the formula: Δfk(t)=|x_k(t)-μ_fk| / μ_fk, where x_k(t) is the real-time monitoring value of the latent influence layer factor fk at time t, and μ_fk is the historical average value of the latent influence layer factor fk. Set an anomaly judgment threshold Δ_th, which is determined based on the statistical analysis of historical anomaly data of latent factors. When Δfk(t)≥Δ_th, it is determined that factor fk has an anomaly, and the anomaly factor identifier is recorded. The influence of the latent factor fk on each core scenario is calculated using the following formula: I(fk,Sj)=I_fk×C_kj×e^(-τ_kj×ε), where I_fk is the influence of factor fk, C_kj is the coupling degree between fk and scenario Sj, τ_kj is the lag time from the change of fk to the change of index in scenario Sj, and ε is the influence attenuation coefficient, which is determined based on the lag characteristics of the latent factor influence. Based on the influence degree I(fk,Sj), a logistic regression model is used to predict the probability of scenario Sj becoming abnormal. The model formula is: P(Sj abnormal|fk abnormal)=1 / [1+e^(-(a1×I(fk,Sj)+b1))], where a1 is the influence degree coefficient and b1 is the bias term, both of which are trained based on the corresponding data of historical latent factor abnormalities and scenario abnormalities. Set an anomaly probability threshold P0, and filter scenarios where P(Sj abnormal|fk abnormal)≥P0 as potential anomaly scenarios; Starting with the latent abnormal factor fk, based on the cross-scene association strength matrix R, a reverse path search algorithm is used to mine the transmission path from fk to each potential abnormal scene, calculate the corresponding path weight W_path, and select the path with path weight W_path≥W2_th as the effective reverse chain transmission path, where W2_th is the reverse path weight threshold. The path form is represented as: latent factor fk → intermediate association factor → ... → potential abnormal scene Sj; Integrate the forward and reverse chain propagation paths to construct a multi-scenario chain propagation directed graph G=(V,E), where V is the set of nodes and E is the set of directed edges. Calculate the propagation strength of each directed edge using the formula: S(e_pq)=W_path×(1-τ_pq / τ_max), where τ_max is the maximum lag time of all propagation paths.

4. The intelligent decision-making method for building operation and maintenance based on digital twins according to claim 3, characterized in that: S3 includes the following: Based on the constructed multi-scenario chain propagation directed graph G=(V,E) and the effective forward and reverse chain propagation paths, the initial operation and maintenance action plan is generated according to the following logic: The preceding factors corresponding to the directed edges with conduction strength S(e_pq)≥S0 in each effective conduction path are extracted as key control nodes, and S0 is the control node screening threshold, which is determined based on the control efficiency statistics of historical operation and maintenance actions. The key control node belongs to the core driving layer or the scene linkage layer and its contribution to the path weight W_path is ≥30%. The contribution percentage is calculated as follows: Contrib(fp)=(r_pq×I_fp) / W_path×100%, where r_pq is the correlation strength between the key node fp and the subsequent factor fq, and I_fp is the influence of fp. A standardized operation and maintenance action library is constructed, which includes four basic actions: equipment parameter control, operation strategy adjustment, environmental intervention, and spatial resource scheduling. For each key control node fp, candidate actions in the action library that match the control object of fp are matched, and the action adaptability score is calculated. The formula is: S_A(fp,A)=λ1×R_reg(A,fp)×λ2×(1-Cost(A) / Cost_max)×λ3×I_fp, where A is the candidate action, R_reg(A,fp) is the control coefficient of action A on factor fp, which is obtained by training based on historical action execution data and has a value range of [0,1]. st(A) is the single execution cost of action A, and Cost_max is the maximum execution cost of similar actions; λ1, λ2, and λ3 are weight coefficients, satisfying λ1+λ2+λ3=1; candidate actions with S_A(fp,A)≥A0 are selected, where A0 is the action adaptability threshold, determined based on the historical action success rate statistics; according to the principle that each path matches at least one key node action and cross-path actions do not conflict, an initial set of operation and maintenance action schemes {U1,U2,...,Us} is formed, where s is the number of initial schemes, and each scheme Ug contains one set of related actions and their corresponding execution order and parameter settings, with g ranging from 1 to s; The initial set of operation and maintenance action plans {U1,U2,...,Us} is input into the building digital twin model one by one. Simultaneously, a three-dimensional collaborative verification is conducted on the improvement effect of anomalies, the impact of related scenarios, and the dynamic changes in the transmission path. The verification cycle is consistent with the maximum lag time τ_max of the transmission path. After simulating the execution of plan Ug, the improvement rate of the core indicators of the anomaly scenario is calculated using the formula: R_imp(Sj,Ug)=|hj_pre-hj_sim| / hj_pre×100%, where Sj is an anomaly or potential anomaly scenario, hj_pre is the anomaly value of the core indicator of scenario Sj before the plan is executed (real-time monitoring value for anomaly scenarios and predicted anomaly value for potential anomaly scenarios), and hj_sim is the steady-state value of the core indicator of scenario Sj after the digital twin model simulates the execution of Ug. When R_imp(Sj,Ug)≥R_imp_th, the improvement effect of the anomaly scenario Sj is deemed to have met the standard. R_imp_th is the minimum improvement effect threshold, determined based on industry operation and maintenance standards. Extract non-abnormal scenarios associated with effective transmission paths, and calculate the impact coefficient of scheme Ug on the core indicators of the associated scenarios. The formula is: Influ(Sc,Ug)=|hc_ori-hc_sim| / σc, where Sc is the associated scenario, hc_ori is the normal mean value of the core indicator of scenario Sc before scheme execution, hc_sim is the core indicator value after simulated execution of Ug, and σc is the historical data standard deviation of the core indicator of scenario Sc. When Influ(Sc,Ug)≤Influ_th, the impact is considered acceptable, where Influ_th is the impact threshold of the associated scenario, and the allowable fluctuation range of the operation and maintenance indicators based on the associated scenario is determined. After simulating the execution scheme Ug, the directed edge propagation strength S'(e_pq) of the multi-scenario chain propagation directed graph G is recalculated. The formula is: S'(e_pq)=W'_path×(1-τ_pq / τ_max), where W'_path is the updated weight of the path after simulation. The path decay rate is calculated. The formula is: Decay(path,Ug)=|W_path-W'_path| / W_path×100%. When Decay(path,Ug)≥Decay_th, the path is considered to be effectively suppressed. Decay_th is the path decay threshold, which is statistically determined based on the operational risk control requirements. For each initial operational action scheme Ug, the three-dimensional verification results are integrated to form a verification report. The verification report includes: R_imp(Sj,Ug), Influ(Sc,Ug), and Decay(path,Ug) and the corresponding judgment results.

5. The intelligent decision-making method for building operation and maintenance based on digital twins according to claim 4, characterized in that: S4 includes the following: Based on the S3 verification report, a three-dimensional quantitative evaluation is performed on the initial set of operation and maintenance action plans {U1, U2, ..., Us}. The specific formula is as follows: Score(Ug)=r1×(N_imp_pass / N_imp_total)+r2×(N_influ_accept / N_influ_total)+r3 Wherein, N_imp_pass is the number of abnormal or potential abnormal scenarios that achieve the improvement effect, N_imp_total is the total number of abnormal or potential abnormal scenarios, N_influ_accept is the number of related scenarios with acceptable impact, N_influ_total is the total number of related scenarios, N_decay_eff is the number of transmission paths that are effectively suppressed, N_decay_total is the total number of effective transmission paths, r1, r2, and r3 are weight coefficients, and r1+r2+r3=1, dynamically determined based on operation and maintenance priority; scoring thresholds Score1 and Score2 are set. When Score(Ug)≥Score1, it is marked as the optimal initial solution; when Score2<Score(Ug)<Score1, it is marked as the initial solution to be optimized; when Score(Ug)≤Score2, it is marked as the eliminated solution; For the initial solution to be optimized, optimizations were made based on the non-compliance items in the validation report: For scenarios where R_imp(Sj,Ug) < R_imp_th, calculate the improvement gap ΔR_imp = R_imp_th - R_imp(Sj,Ug), and prioritize adjusting the control parameters of the corresponding key node actions; if the parameters still do not meet the requirements after adjustment, replace them with candidate actions with higher adaptability scores in the action library. For paths where Decay(path,Ug) < Decay_th, identify secondary key nodes in the path that do not match actions, supplement the action library with targeted actions that have an adaptability score ≥ A0 × 0.9, incorporate them into the plan, and set a reasonable execution order. For related scenarios Sc where Influ_th×0.8≤Influ(Sc,Ug)≤Influ_th, supplementary adaptation measures are implemented; the adaptation necessity Nec(Sc,Ug)=Influ(Sc,Ug) / Influ_th×I_Sc is calculated, where I_Sc is the importance of the scenario, and scenarios where Nec(Sc,Ug)≥Nec_th are selected, where Nec_th is the adaptation necessity threshold; measures in the adaptation measure library with an impact factor and Sc core indicator correlation degree ≥0.6 and an adaptation effect coefficient ≥0.9 are matched and included in the plan to avoid the impact of related scenarios exceeding the limit; The optimized core actions, supplementary suppression actions, and related scenario adaptation measures are integrated into a candidate final solution U_final according to their priority. Parameter conflicts, resource conflicts, and timing conflicts are checked for U_final. If conflicts exist, adjustments are made according to the principle of prioritizing core actions until the checks pass. The verified U_final is output to the building operation and maintenance execution system. The solution includes an action execution list, triggering conditions, and execution time limits. The system provides real-time feedback on the execution effect. If the actual improvement rate does not reach R_imp_th×0.95, a secondary optimization process is triggered.

6. A building operation and maintenance intelligent decision-making system based on digital twins, applied to the building operation and maintenance intelligent decision-making method based on digital twins as described in any one of claims 1-5, characterized in that: The system includes: a hierarchical library construction module for correlation factors, a bidirectional causal tracing and transmission analysis module, an initial operation and maintenance plan generation and collaborative verification module, a decision-making plan optimization and execution module, and a data storage and synchronization module. The associated factor hierarchy library construction module is used to define the core scenarios of building operation and maintenance, filter associated factors, divide factor hierarchy and configure linkage triggering conditions to ensure that the scenario boundaries do not overlap and the factor hierarchy adapts to the operation and maintenance needs. The bidirectional causal tracing and transmission analysis module is used to access real-time operation and maintenance data, determine scene anomalies, trace the direct correlation factors and cross-scene transmission paths of abnormal scenes in the forward direction, and deduce the potential anomalies and transmission paths caused by the movement of latent factors in the reverse direction. The initial operation and maintenance scheme generation and collaborative verification module is used to extract key control nodes based on the transmission path, match candidate actions to form an initial scheme, and carry out multi-dimensional collaborative verification in the digital twin model. The decision-making scheme optimization and execution module is used to quantitatively evaluate the initial scheme based on the verification results, optimize non-compliant items, supplement adaptation measures, output the final scheme after conflict verification, and provide feedback on the execution effect. The data storage and synchronization module is used to store relevant data, support dynamic data maintenance, and achieve real-time data synchronization with the digital twin model.

7. The intelligent decision-making system for building operation and maintenance based on digital twins according to claim 6, characterized in that: The associated factor hierarchy library construction module includes a scenario definition and factor screening unit and a hierarchy division and threshold configuration unit; The scenario definition and factor screening unit defines core scenarios based on building functions and operation and maintenance goals, clarifies scenario boundaries through quantitative parameter sets, and uses professional analysis methods to screen related factors that are associated with the core indicators of the scenario. The hierarchical division and threshold configuration unit constructs a factor quantification model, assigns candidate factors to corresponding levels, configures the trigger thresholds and collection frequency adjustment rules for factors at each level, and establishes a dynamic threshold update mechanism.

8. A building operation and maintenance intelligent decision-making system based on digital twins according to claim 6, characterized in that: The bidirectional causal tracing and transmission analysis module includes an anomaly detection and forward tracing unit and a latent factor anomaly analysis and reverse tracing unit. The anomaly detection and forward tracing unit accesses real-time operation and maintenance data, uses anomaly detection algorithms to identify abnormal scenarios, calculates the correlation strength between correlation factors and abnormal indicators, and mines and verifies cross-scenario forward transmission paths. The latent factor anomaly analysis and reverse tracing unit monitors the status of latent influence layer factors, determines factor anomalies, calculates the impact of anomalies on each scenario, predicts potential abnormal scenarios, and uncovers reverse transmission paths.

9. A building operation and maintenance intelligent decision-making system based on digital twins according to claim 6, characterized in that: The initial operation and maintenance plan generation and collaborative verification module includes an initial plan generation unit and a multi-dimensional collaborative verification unit; The initial scheme generation unit extracts key control nodes based on the transmission path, matches suitable candidate actions from the standardized action library, and combines them into an initial operation and maintenance action scheme set after scoring and screening. The multi-dimensional collaborative verification unit inputs the initial solution into the digital twin model, simulates the solution execution process, verifies the effect of anomaly improvement, the impact of related scenarios, and the suppression of transmission paths, and integrates the verification results to form a report.

10. A building operation and maintenance intelligent decision-making system based on digital twins according to claim 6, characterized in that: The decision optimization and execution module includes a scheme optimization unit and a scheme verification and execution unit; The scheme optimization unit performs quantitative evaluation and classification of the initial scheme, adjusts action parameters or adds targeted actions for non-compliant items, and supplements adaptation measures for related scenarios. The scheme verification and execution unit integrates the optimized actions according to priority, performs parameter, resource and timing conflict verification, outputs the verified scheme to the execution system, provides real-time feedback on the execution effect and triggers secondary optimization.

Images