Smart building equipment predictive maintenance and energy consumption optimization cloud platform

Abstract

The application discloses a cloud platform for predictive maintenance and energy consumption optimization of a smart building device, relates to the technical field of device maintenance, and comprises a measurement value construction module, a pairing module and a simulation environment construction module.The measurement value construction module calculates potential safety risk quantitative indicators of each piece of real-time device state inference result through risk assessment rules, constructs comprehensive uncertainty measurement values, and screens out high-reliability inference conclusions based on joint analysis of the safety risk quantitative indicators and the comprehensive uncertainty measurement values.The pairing module performs feature alignment and data pairing on the screened high-reliability inference conclusions and reference inference results obtained in a simulation environment, forms a training sample set, optimizes original device state analysis by adopting a joint optimization training strategy, captures dynamic factors in a real running environment, and dynamically generates intelligent decision suggestions relying on the optimized output.The cloud platform significantly improves the operation and maintenance efficiency and the energy utilization level of the smart building through a modular architecture and a multi-stage data processing mechanism.

Description

Technical Field

[0001] This invention relates to the field of equipment maintenance technology, specifically to a cloud platform for predictive maintenance and energy optimization of smart building equipment. Background Technology

[0002] In current smart building management practices, in order to ensure the stable operation of key equipment and reduce overall energy consumption, the focus is usually on regular manual inspections and experience-driven operation and maintenance strategies. However, these methods often suffer from problems such as delayed response, strong subjectivity, insufficient fault prediction, and extensive energy regulation, making them difficult to adapt to modern building scenarios with diverse equipment, complex operating environments, and frequent human-computer interactions.

[0003] Meanwhile, traditional energy consumption optimization schemes are mostly based on fixed rules or static models, which cannot dynamically adapt to complex dynamic factors in actual operation such as environmental changes, personnel mobility and equipment aging, resulting in insignificant energy-saving effects and high maintenance costs.

[0004] Therefore, there is an urgent need for an integrated solution that can sense the operating status of equipment in real time, accurately assess the reliability of prediction results, integrate simulation and real data to optimize models, and dynamically generate intelligent decisions, so as to achieve predictive maintenance and refined energy consumption management of building equipment, and improve overall operation and maintenance efficiency and energy utilization. Summary of the Invention

[0005] The purpose of this invention is to provide a cloud platform for predictive maintenance and energy consumption optimization of smart building equipment to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a cloud platform for predictive maintenance and energy consumption optimization of intelligent building equipment, comprising a data acquisition module, a measurement value construction module, and a pairing module: Data acquisition module: Collects multi-dimensional real-time data reflecting the operating status of equipment, derives equipment status inference results for various types of equipment in real-time under real operating scenarios, and deduces the state evolution process of the same equipment under similar initial conditions to obtain reference inference results; Metrics construction module: For each device status inference result in a real-time environment, calculate the potential associated security risk quantification index through risk assessment rules, construct a comprehensive uncertainty metric, and based on the joint analysis of the security risk quantification index and the comprehensive uncertainty metric, screen out device status inference results with security risks within a controllable range and low comprehensive uncertainty, and define them as high reliability inference conclusions; The pairing module aligns the highly reliable inferences obtained from the screening with the reference inferences obtained in the simulation environment by feature alignment and data pairing to form a training sample set. It adopts a joint optimization training strategy to optimize the original equipment state analysis, capture dynamic factors in the real operating environment, and dynamically generate intelligent decision suggestions based on the optimized output.

[0007] In a preferred embodiment, the metric construction module constructs a comprehensive uncertainty metric, which is composed of security uncertainty and availability uncertainty sub-items.

[0008] In a preferred embodiment, the security uncertainty reflects the degree of confidence bias in the judgment of whether there are potential security risks in the device, that is, the reliability of the output security-related conclusions themselves; The availability uncertainty measures the accuracy of information about whether the current state of the device meets actual usage requirements.

[0009] In a preferred embodiment, the metric construction module calculates a potential associated security risk quantification index for each device status inference result in a real-time environment using built-in risk assessment rules. The security risk quantification index is used to characterize the probability of security risks caused by the current device operating status.

[0010] In a preferred embodiment, the training sample set includes inferred device state values ​​under real operating scenarios and ideal inferred values ​​under simulation reference environments.

[0011] In a preferred embodiment, the training sample set includes device status indicators, associated device basic attributes, environmental background parameters, timestamp markers, and historical maintenance record context information. The device status indicators include failure probability and energy consumption level.

[0012] In a preferred embodiment, the pairing module dynamically generates intelligent decision suggestions based on the optimized output, including: The predictive maintenance plan for equipment and the dynamic control strategy for improving energy efficiency include replacing vulnerable parts in advance and adjusting the frequency of preventive inspections. The dynamic control strategy for improving energy efficiency includes adaptively adjusting air conditioning operating parameters according to changes in indoor and outdoor environments, controlling the intensity of public lighting based on the distribution of pedestrian traffic, and optimizing elevator group control scheduling logic.

[0013] In a preferred embodiment, the data acquisition module collects multi-dimensional real-time data reflecting the operating status of the equipment. The multi-dimensional real-time data includes the equipment's own parameters, surrounding environmental variables, and historical operation and maintenance and fault event records.

[0014] In a preferred embodiment, the data acquisition module derives real-time status inferences for various types of equipment in real-world operating scenarios, including the probability of potential failures in the future, the remaining effective usage period, abnormal behavior tendencies, and current energy consumption performance indicators.

[0015] In a preferred embodiment, the device's own parameters include current, voltage, rotational speed, vibration frequency, temperature, and pressure, while the surrounding environmental variables include indoor temperature and humidity, light intensity, and the density of people in the space.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention achieves comprehensive, real-time perception of equipment operating status through a data acquisition module. It not only collects key operating parameters of the equipment itself (such as current, voltage, and vibration frequency), but also integrates environmental variables (such as temperature, humidity, light intensity, and personnel density) as well as historical maintenance and fault records to construct a multi-dimensional profile of equipment operation. Based on this, the module can derive accurate equipment status inference results in real-time within realistic operating scenarios, including core indicators such as failure probability, remaining lifespan, anomaly tendencies, and energy consumption performance. Simultaneously, by simulating the state evolution process of the same equipment under similar initial conditions in a simulation environment, it obtains standardized reference inference results, providing a comparative basis of real and ideal benchmarks for subsequent data quality assessment and model optimization. This significantly enhances the comprehensiveness and reference value of the system's understanding of equipment status.

[0017] This invention performs in-depth reliability analysis on each real-time device status inference result, introducing a safety risk quantification index and a comprehensive uncertainty metric. The safety risk quantification index assesses the probability of potential safety hazards (such as electrical overheating, system downtime, etc.) arising from device operation, while the comprehensive uncertainty metric further quantifies the unreliability of the prediction results from both safety and availability perspectives. This includes the model's confidence bias in judging safety hazards (safety uncertainty) and the deviation between the prediction results and actual operating performance (availability uncertainty). Based on the joint analysis of these multi-dimensional indicators, the module can effectively screen out device status inference results with controllable safety risks and low prediction uncertainty, defining them as high-reliability inference conclusions. This significantly improves the input quality for subsequent model training and decision generation from the data source, ensuring that the optimization process is based on highly reliable data and avoiding interference from low-quality data on model inference and strategy output.

[0018] This invention constructs a composite training sample set that combines real-world operational status information with ideal reference information by precisely aligning and pairing highly reliable inference conclusions with reference inference results in a simulation environment. This set not only covers core equipment status indicators but also integrates rich contextual information such as basic equipment attributes, environmental parameters, timestamps, and historical maintenance records, providing the model with multi-dimensional and multi-scenario learning samples. Based on this training sample set, a joint optimization training strategy is used to specifically optimize the original equipment status analysis model, enabling the model to more accurately capture complex dynamic factors in the real-world operating environment (such as sudden changes in pedestrian flow, environmental fluctuations, and equipment aging), thereby significantly improving the model's prediction accuracy, adaptability, and generalization ability in real-world scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a framework diagram of the cloud platform of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0022] Example: This example provides a cloud platform for predictive maintenance and energy optimization of smart building equipment. Please refer to [link / reference]. Figure 1 As shown, it includes a data acquisition module, a metric construction module, and a pairing module: Data Acquisition Module: Collects multi-dimensional real-time data reflecting the operating status of equipment, specifically covering equipment parameters (such as current, voltage, speed, vibration frequency, temperature, and pressure), surrounding environmental variables (such as indoor temperature and humidity, light intensity, and personnel density), and historical maintenance and fault event records. Based on the above dynamic inputs, it derives real-time status inference conclusions for various types of equipment in real-world operating scenarios. These conclusions include, but are not limited to, the probability of potential faults occurring in the future, the remaining effective service life, abnormal behavior tendencies, and current energy consumption performance. These conclusions are collectively referred to as equipment status inference results in the real-time environment. For example, by reproducing historical operating data, simulating typical operating conditions, or constructing a virtual operating environment based on digital twin technology, it extrapolates the state evolution process of the same equipment under similar initial conditions, thereby obtaining a set of reference inference conclusions under ideal or standardized settings, i.e., reference inference results in the simulation environment. The equipment status inference results are sent to the metric construction module, and the reference inference results are sent to the pairing module.

[0023] The metric construction module calculates a quantitative indicator of potential associated safety risks for each real-time equipment status inference result using built-in risk assessment rules. This indicator characterizes the probability of safety hazards (such as fire risks caused by electrical system overheating) that may be caused by the current equipment operating state. It also constructs a comprehensive uncertainty metric, which consists of two core sub-items: First, there is the uncertainty of safety, which reflects the degree of confidence bias in the judgment of whether there are potential safety hazards in the equipment, that is, the reliability of the output safety-related conclusions (such as high-risk anomaly warnings). Second, availability uncertainty measures the accuracy of predictions about whether the current state of equipment truly meets actual usage needs. Examples include the discrepancy between predicted energy consumption and actual metering values, and the inconsistency between predicted equipment availability and time-series maintenance feedback.

[0024] By integrating the two uncertainty components mentioned above, a comprehensive uncertainty metric describing the credibility of the overall prediction results is formed. Based on the joint analysis of the safety risk quantification index and the comprehensive uncertainty metric, equipment status inferences with controllable safety risks and low comprehensive uncertainty are selected and defined as high-reliability inferences. The selection results are then sent to the pairing module.

[0025] The pairing module performs feature alignment and data pairing between the selected high-reliability inferences and the obtained reference inferences in the simulation environment, forming a training sample set that simultaneously contains equipment status inferences from real-world operating scenarios and ideal inferences from the simulation reference environment. This set not only covers core equipment status indicators (such as failure probability and energy consumption levels) but also integrates related basic equipment attributes, environmental background parameters, timestamps, and historical maintenance records. Based on this training sample set, a joint optimization training strategy is employed to specifically optimize the original equipment status analysis, enabling it to more accurately capture complex dynamic factors in the real-world operating environment (such as sudden changes in personnel flow, environmental parameter fluctuations, and natural equipment aging). Based on the optimized output, intelligent decision-making suggestions for actual operation and maintenance and energy consumption management are dynamically generated, including but not limited to: Predictive maintenance plans for critical equipment (such as replacing vulnerable parts in advance and adjusting the frequency of preventive inspections), and dynamic control strategies aimed at improving energy efficiency (such as adaptively adjusting air conditioning operating parameters according to changes in indoor and outdoor environments, controlling the intensity of public lighting based on the distribution of pedestrian traffic, and optimizing elevator group control scheduling logic to reduce ineffective energy consumption).

[0026] The cloud platform's operation process is as follows: In the actual operation environment of smart buildings, multi-dimensional real-time data reflecting the operating status of equipment is continuously collected through various types of sensors deployed on key equipment (such as central air conditioning units, vertical transportation systems, public lighting facilities, and security monitoring terminals). This data specifically covers equipment parameters (such as current, voltage, speed, vibration frequency, temperature, and pressure), surrounding environmental variables (such as indoor temperature and humidity, light intensity, and personnel density), and historical maintenance and fault event records. Based on these dynamic inputs, real-time state inference conclusions for various types of equipment are derived in real-time under actual operating scenarios. These conclusions include, but are not limited to, the probability of potential faults occurring in the future, the remaining effective service life, abnormal behavior tendencies, and current energy consumption performance. These conclusions are collectively referred to as equipment state inference results in the real-time environment. For example, by reproducing historical operating data, simulating typical operating conditions, or constructing a virtual operating environment based on digital twin technology, the state evolution process of the same equipment under similar initial conditions is extrapolated, thereby obtaining a set of reference inference conclusions under ideal or standardized settings—that is, reference inference results in the simulation environment.

[0027] For each real-time equipment status inference result, a potential associated safety risk quantification index is calculated using built-in risk assessment rules. This index characterizes the probability of safety hazards (such as fire risk caused by electrical system overheating) that may be caused by the current equipment operating state. A comprehensive uncertainty metric is constructed, consisting of two core sub-items: First, there is the uncertainty of safety, which reflects the degree of confidence bias in the judgment of whether there are potential safety hazards in the equipment, that is, the reliability of the output safety-related conclusions (such as high-risk anomaly warnings). Second, availability uncertainty measures the accuracy of predictions about whether the current state of equipment truly meets actual usage needs. Examples include the discrepancy between predicted energy consumption and actual metering values, and the inconsistency between predicted equipment availability and time-series maintenance feedback.

[0028] By integrating the two uncertainty components mentioned above, a comprehensive uncertainty metric describing the credibility of the overall prediction results is formed. Based on the joint analysis of the safety risk quantification index and the comprehensive uncertainty metric, equipment status inferences with controllable safety risks and low comprehensive uncertainty are selected and defined as high-reliability inference conclusions.

[0029] The highly reliable inferences obtained through screening are matched with the reference inferences obtained in the simulation environment through feature alignment and data pairing to form a training sample set that simultaneously contains inferred equipment status values ​​under real operating scenarios and ideal inference values ​​under the simulation reference environment. This set not only covers core equipment status indicators (such as failure probability and energy consumption level) but also integrates contextual information such as associated basic equipment attributes, environmental background parameters, timestamps, and historical maintenance records. Based on this training sample set, a joint optimization training strategy is adopted to specifically optimize the original equipment status analysis, enabling it to more accurately capture complex dynamic factors in the real operating environment (such as sudden changes in personnel flow, fluctuations in environmental parameters, and natural aging of equipment). Based on the optimized output, intelligent decision-making suggestions for actual operation and maintenance and energy consumption management are dynamically generated, including but not limited to: Predictive maintenance plans for critical equipment (such as replacing vulnerable parts in advance and adjusting the frequency of preventive inspections), and dynamic control strategies aimed at improving energy efficiency (such as adaptively adjusting air conditioning operating parameters according to changes in indoor and outdoor environments, controlling the intensity of public lighting based on the distribution of pedestrian traffic, and optimizing elevator group control scheduling logic to reduce ineffective energy consumption).

[0030] The functions of each module in this application are described in detail below: The data acquisition module continuously collects diverse data from various physical devices and their operating environment within the building. This data effectively characterizes the current operating status, usage environment, and historical behavioral features of the devices, supporting subsequent status deduction and analysis. Based on their source and nature, the collected data can be categorized into the following three types: Device's own parameters (native monitoring data during device operation) This data comes directly from the equipment's built-in sensors or connected monitoring units, reflecting the core physical and electrical characteristics of the equipment during operation. It is the most fundamental basis for judging the equipment's health status and operational efficiency. Specifically, it covers, but is not limited to: Current and voltage reflect the electrical load and power supply stability of equipment; abnormal fluctuations may indicate circuit faults or load abnormalities. Rotational speed is applicable to rotating equipment (such as fans, pumps, and motors) and is used to determine whether the mechanical operation is within the rated or abnormal range. Vibration frequency is particularly important for mechanical equipment; frequency domain characteristics can reveal hidden faults such as bearing wear, rotor imbalance, and poor gear meshing. Temperature monitoring of temperature changes in key components (such as motor windings, bearings, and electrical contacts) is a core indicator for assessing overheating risk and heat accumulation effects. For equipment involving fluid transmission or compression (such as air conditioning compressors and hydraulic systems), pressure parameters are an important basis for evaluating their operating status and safety boundaries.

[0031] Surrounding environmental variables (external condition data for equipment operation) This type of data describes the real-time state of the physical environment in which the equipment exists. These environmental factors often have a significant impact on equipment performance, energy consumption, and fault triggering, and are indispensable inputs for building environment-equipment coupling analysis models. Specifically, this includes: Indoor temperature and humidity affect the load and energy efficiency of environmental control equipment such as air conditioners and fresh air systems, and are also important factors for personnel comfort and equipment condensation risk; light intensity has direct guiding significance for the adjustment strategy of public area lighting systems, and is also the basis for judging the utilization of natural light and the need for artificial lighting; spatial personnel density is obtained through human body sensors or video analysis modules deployed in buildings, and is used to dynamically assess the operation strategies of equipment related to personnel flow, such as air conditioning load, lighting demand, and elevator scheduling; Historical maintenance and fault event records (time-series and event-based data) This type of data is time-series in nature, recording key information such as maintenance operations, failure times, failure types, repair measures, and replaced parts during past operating cycles. It forms the basis for building equipment health records and achieving failure mode identification and trend prediction based on historical experience. Specific content includes: The time, content, and results of each preventive maintenance; the specific time of the failure, the fault code, the fault location information, and the handling process; the model, supplier, installation time, and expected lifespan of the replaced parts, etc.

[0032] Based on these dynamic inputs, and with the help of built-in data processing and analysis logic, the state of each type of equipment under current operating conditions is comprehensively extrapolated, thereby outputting a set of key inference indicators in real time to describe the equipment's health, risk level, and energy efficiency performance. These indicators are collectively referred to as the equipment state inference results under real-time conditions. Their specific content includes, but is not limited to: By analyzing the temporal variation trends, statistical distribution characteristics, and threshold deviations of key equipment parameters (such as vibration, temperature, and current), and combining this with a database of failure modes of similar historical equipment, the probability of a specific type of failure (such as bearing failure, electrical short circuit, or mechanical jamming) occurring within a certain time window in the future (such as the next 24 to 72 hours) is assessed. The output is a probability value between 0 and 1, which is used to characterize the risk level of the failure.

[0033] Based on the current performance degradation indicators of the equipment (such as efficiency decline, parameter drift, cumulative damage, etc.), combined with the equipment's design life, operating time, historical maintenance records and the degradation trajectory of similar equipment, the remaining time (usually measured in hours, days or usage cycles) that the equipment is expected to maintain normal and reliable operation under the current operating conditions is estimated, providing a quantitative basis for developing preventive replacement or overhaul plans.

[0034] By constructing a baseline for the normal operation of equipment (such as parameter distribution range, timing pattern, start-stop pattern, etc. under normal operating conditions), the deviation of the current observation data from the baseline is compared in real time to identify operating modes or parameter mutations that may indicate potential anomalies, such as unplanned start-stop, disappearance of periodic fluctuations, abnormal parameter coupling relationship, etc., and then output qualitative or semi-quantitative assessment conclusions of abnormal tendencies.

[0035] By combining real-time power data, runtime, load rate, and environmental regulation requirements of the integrated equipment, the energy consumption level under current operating conditions can be assessed. It can also be compared with historical data, similar equipment, or preset energy efficiency benchmarks to identify high energy consumption behaviors or energy-saving potential points, providing input for subsequent energy consumption optimization strategies.

[0036] The above inference process can be implemented using various data processing methods, such as threshold and trend analysis based on statistical process control (SPC) and anomaly detection logic based on rule engines. These methods can be flexibly combined according to device type, data quality, and business needs to achieve accurate adaptation to different devices and scenarios. This method belongs to the prior art and will not be elaborated further in this application.

[0037] To provide a benchmark and optimization reference for the equipment state inference results in a real-world environment, this module also simultaneously constructs a set of equipment state inference conclusions derived under ideal or standardized settings, referred to as the reference inference results in the simulation environment. These results are typically generated in the following way: Select a period of time in the history when the equipment is in good operating condition and free from faults, extract its parameter sequence and operating environment, and use it as a health baseline for playback and analysis; based on the equipment design parameters and typical usage scenarios (such as standard load, rated temperature and humidity, unmanned environment, etc.), simulate the equipment's operating performance under ideal working conditions through digital modeling; use digital twin technology to perform high-fidelity modeling of the equipment and its operating environment, input initial conditions and boundary parameters similar to the real situation, and deduce the state evolution process of the equipment in the future in virtual space, thereby obtaining a set of expected state indicators, that is, the expected performance of the equipment in theory under no abnormal interference, standard configuration and ideal environment.

[0038] The metric construction module quantifies and assesses the potential safety hazards implied in each real-time device status inference result to clarify the probability of the device triggering specific types of safety events (such as electrical fires, mechanical failures, overheating damage, etc.) under the current operating state, thereby providing the system with an intuitive basis for risk warning.

[0039] Specifically, the platform incorporates a risk assessment rule engine. Based on industry safety standards, equipment operating mechanisms, historical failure cases, and an expert knowledge base, this engine pre-defines several risk association rules and risk level mapping logics for different types of equipment (such as HVAC equipment, elevators, power distribution facilities, and lighting systems) and key operating parameters (such as temperature, current, vibration, and pressure). These rules are not simply threshold comparisons, but rather a comprehensive judgment logic that considers the current value of the parameter, its changing trend, duration, multi-parameter coupling relationships, and the environmental context of the equipment (such as personnel density and space usage). Key parameters strongly correlated with safety are extracted from real-time equipment status inference results, such as temperature, current density, and insulation status of electrical equipment; vibration amplitude and bearing temperature of mechanical equipment; and environmental parameters such as personnel density and space airtightness. The extracted risk factors are input into a predefined set of risk assessment rules, and each rule is matched according to a logical matching principle (e.g., if the equipment temperature exceeds a set threshold and persists for a certain period, a high-temperature risk is triggered). Based on the level of the successfully matched rule, the type of safety hazard currently present in the equipment is preliminarily determined (e.g., electrical overheating, mechanical fatigue, structural overload, etc.). Based on historical risk occurrence statistics corresponding to the matching rules, the inherent risk coefficient of the equipment type, the degree of deviation of current parameters from the baseline, and environmental synergistic influence factors, a logic combining rules and statistics is used to estimate the probability of this type of safety hazard occurring under current operating conditions, and quantify it into a safety risk quantification index ranging from 0 to 1 (the higher the value, the greater the likelihood of a potential safety event). This safety risk quantification index is not only used to determine whether the equipment is currently in a high-risk state, but also provides a numerical basis for subsequent risk controllability assessment and maintenance priority ranking.

[0040] Besides safety risks, another key quality dimension of equipment status inference results is the reliability and accuracy of its predictions, i.e., the extent to which the system's output conclusions (such as the equipment currently having no abnormalities, predicting no failure within the next 24 hours, and estimating energy consumption of X kWh) are credible and consistent with reality. To this end, this module further constructs an evaluation index called the Comprehensive Uncertainty Metric, which consists of two independent but jointly influential sub-items affecting the overall prediction quality: 1) Security uncertainty Security uncertainty measures the confidence bias of a system's judgment regarding the core question of whether a device currently possesses potential security vulnerabilities. In other words, it reflects the reliability of the system's safety-related conclusions, such as high-risk anomaly warnings and potential fault indications. The processing logic is described below: This analysis examines the stability and consistency of safety-related sub-indicators in current equipment status inference (such as temperature deviation, vibration anomaly index, and current fluctuation rate). For example, it assesses whether there are phenomena like drastic parameter fluctuations within a short period, inconsistencies in multi-sensor data, or significant deviations compared to historical data. It also evaluates whether key parameters used for safety judgment fall within a known ambiguity or gray area (i.e., close to but not exceeding a clear threshold, making direct anomaly determination difficult). Combining this with the accuracy of the system's judgments under similar historical conditions (indirectly reflected by statistics such as confusion matrix, false alarm rate, and false negative rate), the reliability deviation of current safety-related conclusions is comprehensively derived and quantified into a sub-indicator representing the degree of uncertainty in safety judgment: safety uncertainty. A higher sub-indicator indicates a lack of confidence in the system's current judgment regarding the existence of safety hazards or a greater risk of misjudgment.

[0041] 2) Availability uncertainty Availability uncertainty measures the accuracy of system-generated predictions about whether the current state of equipment truly meets actual usage requirements. Specifically, it measures the degree of agreement between the system's predicted equipment performance, energy consumption, or availability and actual operational feedback. Examine the degree of deviation between key forecast indicators (such as predicted energy consumption, predicted equipment availability, and predicted remaining lifespan) and actual historical metering data or recent maintenance feedback. Examples include the absolute or relative error between predicted energy consumption and actual meter readings, and the difference between predicted equipment availability and actual downtime. If multiple sources of feedback exist (such as manual records from maintenance personnel, user feedback, and data from automatic metering systems), aggregate and analyze this feedback to assess the consistency between the forecast conclusions and the actual data from multiple sources. Based on the analysis results of the above deviations and consistency, derive a quantitative indicator reflecting the accuracy and reliability of the forecast conclusions in actual usage scenarios: availability uncertainty. The higher this sub-indicator, the more likely the system's current prediction of whether the equipment status truly meets usage requirements is to deviate from reality, thus reducing its decision-making guidance value.

[0042] The two sub-indicators, security uncertainty and availability uncertainty, are synthesized into a unified indicator for comprehensively evaluating the overall credibility of prediction results. This unified indicator is achieved through a predefined logical fusion strategy (such as weighted average, analytic hierarchy process, expert scoring mapping, etc., with weights determined based on actual business needs and data distribution characteristics). This comprehensive indicator reflects the system's overall grasp of the inference results regarding the state of a certain device in both the security judgment and usability prediction dimensions. The lower the value, the higher the overall credibility of the inference result, and the more suitable it is as input for subsequent model training and decision support.

[0043] Based on the aforementioned analysis results—namely, the safety risk quantification index and comprehensive uncertainty measure corresponding to each equipment status inference result—this module further implements joint analysis and threshold screening logic. Its core objective is to identify equipment status inference results that are both safe and controllable (i.e., safety risks are within an acceptable range, such as safety risk quantification indexes below a preset threshold) and highly reliable (i.e., low comprehensive uncertainty measure, indicating strong reliability and accuracy of the prediction conclusion). Screening logic: Inferences whose quantifiable safety risk indicators exceed the system's preset safety thresholds (e.g., above 0.7 or 0.8, depending on the industry or equipment type) are discarded to ensure that data entering subsequent processes does not introduce high security risks. Based on controllable safety risks, inferences whose comprehensive uncertainty metric value is below a certain reliability threshold (e.g., below 0.3 or 0.4) are further screened to ensure that the retained inferences are not only reliable in terms of safety but also possess high levels of predictive accuracy and practical applicability. After this dual screening, the remaining inferences are defined as high-reliability inferences. These conclusions represent the most credible, safest, and most realistic assessments and predictions of the equipment status made by the system under the current operating environment.

[0044] The foundational work of the pairing module is to perform multi-dimensional feature alignment and data association between the rigorously screened, highly reliable inferences from the metric construction module (i.e., inferences obtained in real operating environments with controllable safety risks and low overall uncertainty) and the reference inferences from the simulation environment synchronously generated by the data acquisition module (i.e., equipment expected state indicators derived under ideal or standardized settings through methods such as digital twins, historical reproduction, or typical operating condition simulation). This results in a set of high-quality training samples that combine realism and reference.

[0045] The core state indicators contained in the actual and simulated inference results are uniformly identified and mapped to ensure consistency in key parameters. These core indicators typically include, but are not limited to: Failure probability (reflecting the likelihood of equipment failure within a certain future timeframe); energy consumption levels (such as instantaneous power, cumulative energy consumption, and energy efficiency per unit time); equipment operating status classification labels (such as normal, sub-healthy, and high-risk); and derived indicators such as predicted remaining useful life (RUL) or abnormal tendency scores. Precise matching is performed using field names, data types, timestamps, and unique equipment identifiers (such as equipment ID and location codes) to establish a one-to-one or one-to-many relationship between real and simulated data, ensuring that subsequent paired data remains consistent across time windows, equipment objects, and operating contexts.

[0046] Multidimensional contextual information associated with each inference result is integrated to enhance the environmental description capability of training samples and the model's generalization ability. This contextual information mainly includes: Static characteristics of equipment include basic attributes such as equipment model, specifications, installation location, commissioning time, rated parameters, and manufacturer information; dynamic environmental variables such as indoor temperature and humidity, light intensity, personnel density, and season / time period labels at the time of inference; accurate recording of the time information (e.g., year / month / day / hour / minute) corresponding to each inference result to depict the time-series evolution pattern and periodic characteristics; historical maintenance records include the time of the most recent maintenance, maintenance type (preventive / fault-related), replacement parts list, and maintenance personnel feedback to reflect the historical health status of the equipment and the effectiveness of operation and maintenance interventions.

[0047] Based on obtaining a high-quality training sample set, this module introduces and implements a joint optimization training strategy to conduct targeted, multi-objective collaborative optimization training on the platform's existing equipment status analysis models (which may include fault prediction models, energy consumption prediction models, status classification models, etc.) to improve the models' performance in real-world operating environments, particularly their adaptability to complex, dynamic, and non-steady-state factors. The optimization objectives and processing are as follows: The joint optimization strategy sets multi-dimensional optimization objectives, including but not limited to: Improve the model's prediction accuracy for core indicators such as real-world failure probability and energy consumption level (e.g., reduce prediction error and improve classification F1 score); enhance the model's robustness and generalization ability to common disturbances in real-world environments (e.g., sudden changes in pedestrian flow, fluctuations in environmental parameters, and equipment aging trends); reduce the prediction bias between real-world inference and simulation reference, so that the model output gradually approaches the expected performance under ideal conditions.

[0048] Model training: The training sample set is input into the training process of the device state analysis model. A supervised learning mechanism is adopted, using core indicators from real inference results as labels and simulation reference results as benchmarks to guide the model to learn how to more accurately fit complex patterns in real data. During training, advanced training techniques such as multi-task learning, adversarial training, attention mechanisms, or dynamic weight adjustment are introduced (the specific technique selection depends on the model architecture and business needs) to enhance the model's ability to capture key features and adapt to complex factors such as noise, anomalies, and temporal dependencies. Training optimization methods such as cross-validation, early stopping mechanisms, and learning rate scheduling are used to prevent overfitting and improve the model's generalization performance in new scenarios. This optimization process is essentially driven by dual-source data of real and simulation data, enabling the model to gradually learn to maintain high-precision state inference and prediction capabilities under imperfect, dynamic, and multi-interference conditions in the real world. The specific model training is existing technology and will not be described in detail in this application.

[0049] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0050] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A cloud platform for predictive maintenance and energy optimization of smart building equipment, characterized by: It includes a data acquisition module, a metric construction module, and a pairing module: Data acquisition module: Collects multi-dimensional real-time data reflecting the operating status of equipment, derives equipment status inference results for various types of equipment in real-time under real operating scenarios, and deduces the state evolution process of the same equipment under similar initial conditions to obtain reference inference results; Metrics construction module: For each device status inference result in a real-time environment, calculate the potential associated security risk quantification index through risk assessment rules, construct a comprehensive uncertainty metric, and based on the joint analysis of the security risk quantification index and the comprehensive uncertainty metric, screen out device status inference results with security risks within a controllable range and low comprehensive uncertainty, and define them as high reliability inference conclusions; The pairing module aligns the highly reliable inferences obtained from the screening with the reference inferences obtained in the simulation environment by feature alignment and data pairing to form a training sample set. It adopts a joint optimization training strategy to optimize the original equipment state analysis, capture dynamic factors in the real operating environment, and dynamically generate intelligent decision suggestions based on the optimized output.

2. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 1, characterized in that: The metric construction module constructs a comprehensive uncertainty metric, which is composed of security uncertainty and availability uncertainty sub-items.

3. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 2, characterized in that: The aforementioned security uncertainty reflects the degree of confidence bias in judging whether there are potential security risks in the equipment, that is, the reliability of the output security-related conclusions themselves; The availability uncertainty measures the accuracy of information about whether the current state of the device meets actual usage requirements.

4. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 3, characterized in that: The metric construction module calculates the potential associated security risk quantification index for each device status inference result in a real-time environment through built-in risk assessment rules. The security risk quantification index is used to characterize the probability of security risks caused by the current device operating status.

5. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 1, characterized in that: The training sample set includes inferred device state values ​​under real operating scenarios and ideal inferred values ​​under simulation reference environments.

6. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 5, characterized in that: The training sample set includes equipment status indicators, associated basic equipment attributes, environmental background parameters, timestamp markers, and historical maintenance record context information. The equipment status indicators include failure probability and energy consumption level.

7. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 6, characterized in that: The pairing module dynamically generates intelligent decision suggestions based on the optimized output, including: The predictive maintenance plan for equipment and the dynamic control strategy for improving energy efficiency include replacing vulnerable parts in advance and adjusting the frequency of preventive inspections. The dynamic control strategy for improving energy efficiency includes adaptively adjusting air conditioning operating parameters according to changes in indoor and outdoor environments, controlling the intensity of public lighting based on the distribution of pedestrian traffic, and optimizing elevator group control scheduling logic.

8. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 1, characterized in that: The data acquisition module collects multi-dimensional real-time data reflecting the operating status of the equipment. The multi-dimensional real-time data includes the equipment's own parameters, surrounding environmental variables, and historical operation and maintenance and fault event records.

9. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 8, characterized in that: The data acquisition module derives real-time status inferences for various types of equipment in real-world operating scenarios, including the probability of potential failures in the near future, remaining effective usage period, abnormal behavior tendencies, and current energy consumption performance indicators.

10. The cloud platform for predictive maintenance and energy optimization of smart building equipment according to claim 8, characterized in that: The device's own parameters include current, voltage, rotation speed, vibration frequency, temperature, and pressure, while the surrounding environmental variables include indoor temperature and humidity, light intensity, and the density of people in the space.

Images