A cloud platform-based HVAC equipment work order task intelligent scheduling system

Abstract

The application discloses a kind of based on cloud platform's HVAC equipment work order task intelligent scheduling system, it is related to operation and maintenance control technical field, collect running state data and user input unstructured repair text, and carry out feature extraction generation standard fault label and initial work order task;Build time series equipment deterioration model, and real-time calculation HVAC equipment's cumulative deterioration function, generate equipment health quantification value, calculate the predicted time consumption of HVAC equipment abnormal state repair, convert predicted time consumption into slack time constraint index;To initial work order task, equipment health quantification value and slack time constraint index are judged, and automatically insert predictive maintenance work order, based on work order task optimization algorithm to updated task set and slack time constraint index are iteratively solved, generate optimal dispatch sequence;Optimal dispatch sequence and slack time constraint index are issued to maintenance technician terminal, collect actual maintenance feedback data, to actual maintenance feedback data reverse correction model.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance control technology, and in particular to an intelligent scheduling system for HVAC equipment work orders based on a cloud platform. Background Technology

[0002] As a core infrastructure of modern buildings, the operational stability of HVAC equipment directly affects the normal operation of commercial buildings and the environmental comfort of users. With the development of IoT and cloud computing technologies, after-sales maintenance of HVAC equipment is gradually transitioning to platform-based management. However, existing work order scheduling and maintenance management systems still have the following significant shortcomings in complex real-world scenarios:

[0003] Existing repair mechanisms primarily rely on customers submitting repair requests manually via phone or text, often only responding passively when indoor temperatures are already extremely high. This lacks early detection of underlying equipment problems. Furthermore, the semantics of manual repair requests are often ambiguous, making it difficult for the dispatch system to accurately predict the true root cause of the fault. This results in maintenance technicians arriving on-site only to find missing parts or lacking the necessary skills, leading to a very high rate of repeat visits. Existing dispatch systems typically employ rigid response time standards, completely ignoring the physical properties of buildings. For large commercial complexes with excellent insulation, even if equipment is shut down, the indoor temperature may not exceed the user's comfort threshold for up to four hours, easily causing the dispatch system's capacity to be affected by pseudo-emergency tasks during peak periods. In traditional business models, emergency corrective maintenance and regular preventative maintenance are usually two separate processes. Existing scheduling algorithms struggle to deeply optimize work order paths by incorporating business logic. Most existing intelligent scheduling systems rely on static parameters, such as fixed equipment lifespan decay curves and fixed building thermal conductivity. However, as equipment wears down and buildings age, static theoretical models drift significantly from the real physical world. The lack of a closed-loop correction mechanism based on actual maintenance results leads to reduced system accuracy, requiring substantial manpower for parameter recalibration later on. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies are unable to detect potential problems at the underlying level of equipment in the early stages, blindly dispatch orders without regard to the thermal inertia of buildings, resulting in the exhaustion of pseudo-emergency tasks in the scheduling system during peak periods, and the scheduling model lacks real physical feedback.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a cloud-based intelligent scheduling system for HVAC equipment work orders, comprising an event triggering module, a state thermal inertia module, a target optimization module, and a movement execution module.

[0006] The event triggering module is used to collect the operating status data of HVAC equipment and the unstructured repair report text input by the user, extract features from the unstructured repair report text to generate standard fault labels, and perform baseline comparison and anomaly detection on the operating status data to generate an initial work order task.

[0007] The state thermal inertia module is used to construct a time-series equipment deterioration model, calculate the cumulative deterioration function of HVAC equipment in real time, generate a quantitative value of equipment health, calculate the predicted time consumption for repairing abnormal states of HVAC equipment based on the building thermal inertia physical model, and convert the predicted time consumption into a relaxation time constraint index.

[0008] The target optimization module is used to judge the initial work order tasks, equipment health quantification values ​​and relaxation time constraint indicators, and automatically insert predictive maintenance work orders. Based on the work order task optimization algorithm, iteratively solves the updated task set and relaxation time constraint indicators to generate the optimal dispatch sequence.

[0009] The mobile execution module is used to receive and send the optimal work order sequence and relaxation time constraint index to the maintenance technician terminal. After the work order is completed, it collects actual maintenance feedback data and uses the actual maintenance feedback data to reverse correct the equipment deterioration model and the building thermal inertia physical model.

[0010] Preferably, the event triggering module includes:

[0011] Obtain unstructured repair request text input by the user, and use a natural language processing algorithm model to map the unstructured repair request text into a work order feature vector;

[0012] Calculate the cosine similarity between the work order feature vector and the feature vector of the i-th standard fault template in the fault knowledge base. Traverse the fault knowledge base and perform cosine similarity judgment on the work order feature vectors:

[0013] The label corresponding to the standard fault template feature vector with the largest cosine similarity and greater than the preset semantic matching threshold is selected as the standard fault label.

[0014] Collect operating status data of HVAC equipment, including temperature, pressure, acoustic signature and vibration data, and record the data corresponding to the operating status data at the current sampling time as a real-time status vector;

[0015] Retrieve the fault-free baseline state vector, calculate the residual vector between the real-time state vector and the baseline state vector, construct the residual evaluation function, and quantify the severity of the equipment's deviation from the normal baseline.

[0016] Set a dynamic early warning threshold for the residual evaluation function and construct an event triggering logic mechanism. When an event is triggered, activate the event and generate an initial work order task. The event determination includes a first determination, a second determination, and a third determination.

[0017] Preferably, the state thermal inertia module includes a model building unit and a time consumption prediction unit;

[0018] The model building unit is used to receive operating status data, build a time-series equipment deterioration model, calculate the cumulative deterioration function of HVAC equipment in real time, and generate a quantitative value of equipment health.

[0019] Receive the running status data from the event triggering module, extract the sound and vibration data from the running status data, calculate the dynamic deterioration rate of the HVAC equipment at the current moment, and construct a time-series equipment deterioration model;

[0020] Based on historical operating loads, the nonlinear exponential parameters of equipment degradation are obtained, and the deterioration rate of HVAC equipment since the last maintenance is calculated in real time. up to the current moment The cumulative deterioration function is obtained, and the cumulative deterioration function is normalized to obtain the quantified value of equipment health.

[0021] Preferably, the time prediction unit includes:

[0022] Extract temperature data from the operating status data, including the actual indoor temperature. and outdoor ambient temperature A physical model of building thermal inertia based on a first-order equivalent thermal capacity and thermal resistance network is established. The thermodynamic differential equation of the physical model of building thermal inertia is as follows:

[0023] ;

[0024] in, The equivalent heat capacity of the building. The equivalent thermal resistance of the building shell. The internal heat load generated by indoor personnel and equipment, The actual heating or cooling capacity provided by the HVAC equipment under the current abnormal condition;

[0025] The predicted time is obtained by calculating the time it takes for the indoor temperature to rise from the current actual temperature to the critical temperature within the user's comfort range based on the thermodynamic differential equation.

[0026] Obtain the scheduling safety buffer constant, and subtract the predicted time from the scheduling safety buffer constant to obtain the relaxation time constraint index.

[0027] Preferably, the target optimization module includes a predictive maintenance decision unit, a constraint modeling unit, and a variable neighborhood optimization unit;

[0028] The predictive maintenance decision unit is used to judge the quantitative value of equipment health and insert predictive maintenance work orders;

[0029] Receive device health quantification values, set security alert thresholds, and establish a threshold-triggered comparison mechanism:

[0030] When the measured health value of the equipment is less than the safety warning threshold, the current HVAC equipment is at risk of sudden downtime. A predictive maintenance work order is generated, and the predictive maintenance work order and the initial work order task are merged to update the task queue and generate an updated task set.

[0031] Preferably, the constraint modeling unit is used to construct a scheduling optimization mathematical model based on the updated task set and relaxation time constraint index;

[0032] Obtain the set of maintenance workers to be scheduled. For the updated task set, define decision variables, including path allocation variables and time node variables:

[0033] When employee k executes two or more elements of the updated task set consecutively, the path assignment variable is 1; otherwise, it is 0.

[0034] Iterate through the updated task set, obtain the actual start time and actual completion time of each subtask, and set a latest start time constraint boundary. Current scheduling time Relaxation time constraint index;

[0035] A fitness objective function is constructed based on scheduling resources and timeout penalties. The optimized values ​​of the objective function include a first optimized value and a second optimized value.

[0036] Preferably, the neighborhood optimization unit is used to iteratively solve the problem based on the work order task optimization algorithm to generate the optimal dispatch sequence;

[0037] A genetic algorithm employing a fusion variable neighborhood search mechanism is used as the task optimization algorithm. The updated task set is encoded into an initial population chromosome, which includes task sorting and worker allocation. The fitness objective function of each individual in the population is calculated, and global iterative evolution is performed. After each generation of evolution, elite individuals in the population are extracted for variable neighborhood local search. The variable neighborhood local search includes first neighborhood actions and second neighborhood actions.

[0038] The first neighborhood action includes:

[0039] Identify the subtask with the smallest relaxation time constraint index among the current elite individuals, and place the technician to which the current subtask belongs at the first position in the task execution sequence;

[0040] The second neighborhood action includes:

[0041] Identify predictive maintenance work orders in the updated task set, traverse the initial work order tasks with the same building number, and force the initial work order tasks to be assigned to the same maintenance technician for continuous operation.

[0042] When the number of iterations reaches the preset maximum convergence algebra, the iteration is terminated, the globally optimal individual with the minimum fitness objective function is decoded and output, the globally optimal individual is transformed into a task execution order and personnel matching scheme, and the optimal dispatch sequence is output.

[0043] Preferably, the mobile execution module includes a task guidance unit, a closed-loop feedback unit, and a reverse correction unit;

[0044] The task guidance unit is used to receive and send the optimal dispatch sequence and the corresponding relaxation time constraint index to the maintenance technician's terminal to guide task execution.

[0045] Obtain the optimal dispatch sequence and the corresponding relaxation time constraint index, convert the optimal dispatch sequence into a task instruction package, and send it to the mobile terminal of the corresponding maintenance technician through the communication network. Based on the relaxation time constraint index, generate a dynamic thermal inertia timer on the mobile terminal.

[0046] The dynamic thermal inertia timing The update formula is:

[0047] ;

[0048] in, This refers to the initial absolute time at which the scheduling task is issued. The relaxation time constraint index, This refers to the current absolute moment.

[0049] Preferably, the closed-loop feedback unit is used to collect actual maintenance feedback data after the work order is completed;

[0050] After the mobile terminal receives the work order completion instruction, it triggers the data collection instruction to obtain the actual maintenance feedback data. The actual maintenance feedback data includes the actual maintenance time and equipment recovery energy efficiency. The actual maintenance time is the absolute time difference from when the maintenance technician clicks to start the work on the mobile terminal to when the work ends.

[0051] The processing logic for restoring the energy efficiency of the device is as follows:

[0052] After the maintenance work is completed and the HVAC equipment is restarted, the outlet and return water temperatures, flow rates, and actual power consumption of the HVAC equipment are collected within the duration window to calculate the actual performance coefficient of the equipment after repair.

[0053] Preferably, the reverse correction unit is used to reverse correct the equipment deterioration model and the building thermal inertia physical model using machine learning algorithms based on actual maintenance feedback data;

[0054] The actual maintenance feedback data is transmitted back to the cloud platform, where a first reverse correction and a second reverse correction are performed respectively. The first reverse correction is used to correct the timing equipment degradation model.

[0055] Calculate the mean square error loss function between the system's predicted recovery energy efficiency before maintenance and the equipment's recovery energy efficiency; use the gradient descent algorithm to perform reverse iterative correction on the basic aging constant and deterioration nonlinear exponent parameters in the time-series equipment deterioration model to obtain the updated basic aging constant and deterioration nonlinear exponent parameters.

[0056] The second inverse correction is used to correct the building's thermal inertia physical model:

[0057] The actual indoor temperature drop curve during the maintenance period is extracted and compared with the temperature curve predicted by the building's thermal inertia physical model to obtain the time-series temperature prediction residuals. A recursive least squares algorithm is then used to minimize the sum of squared cumulative temperature prediction residuals to perform inverse online identification and numerical correction of the equivalent heat capacity and equivalent thermal resistance in the building's thermal inertia physical model, resulting in updated equivalent heat capacity and equivalent thermal resistance.

[0058] The beneficial effects of this invention are as follows: This invention constructs a dual-trigger mechanism by combining text NLP extraction with a residual evaluation function based on a weighted sliding window. On the one hand, it avoids false alarms from single-point sensors due to noise; on the other hand, it can generate predictive maintenance work orders several days before catastrophic failures such as bearing seizure, based on the root mean square surge of high-frequency vibrations. This invention also innovatively establishes a first-order equivalent RC thermodynamic model, accurately calculating the predicted time it takes for indoor temperatures to exceed user tolerance limits after equipment downtime. This time is transformed into a relaxation time constraint index in the scheduling algorithm, changing the original immediate departure after receiving an order to a more flexible approach where the worker only needs to arrive before the building's insulation limits are reached. The time window releases scheduling capacity, effectively avoiding capacity congestion during peak maintenance periods in summer or winter, and achieving seamless maintenance for customers; the target optimization module of this invention deeply customizes a variable neighborhood operator in the genetic algorithm that conforms to the business logic of HVAC; the first action (crisis pre-positioning) ensures that the most critical faults with the shortest relaxation time are handled first; the second action (co-location binding) forcibly assigns emergency maintenance orders and predictive warning orders in the same physical space to the same technician for continuous processing, completely eliminating ineffective and repetitive long-distance commutes in scheduling, abandoning rigid time limits, and instead converting thermodynamic calculation results into thermal inertia time that changes dynamically with time on the technician's App. This design provides technicians with the most intuitive physical limits, effectively alleviating psychological anxiety and fatigue risks associated with maintenance personnel on the road. Simultaneously, by forcibly capturing post-repair temperature differences, flow rates, and power consumption through the underlying system to recover energy efficiency, it eliminates the possibility of false manual reporting of repair results. Addressing the pain point of model aging over time, this invention pioneers a closed-loop mechanism that uses real-world feedback data to correct theoretical parameters. For non-linear equipment degradation, it employs gradient descent to reverse-correct the equipment's fundamental aging constant and non-linear exponent. For slowly changing building environments, it uses recursive least squares to identify and update the building's equivalent heat capacity and thermal resistance online. This mechanism ensures that regardless of how aged the building's exterior walls or how worn the equipment, the AI ​​dispatch engine always closely matches the extreme physical state, reducing subsequent manual maintenance costs and improving dispatch accuracy over long lifecycles. Attached Figure Description

[0059] Figure 1 This is a basic flowchart of a cloud-based intelligent scheduling system for HVAC equipment work orders, provided as an embodiment of the present invention. Detailed Implementation

[0060] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0061] Example, refer to Figure 1This paper presents a cloud-based intelligent scheduling system for HVAC equipment work orders, including an event triggering module, a state thermal inertia module, a target optimization module, and a movement execution module.

[0062] The event triggering module is used to collect the operating status data of HVAC equipment and the unstructured repair text input by the user, extract features from the unstructured repair text to generate standard fault labels, and perform baseline comparison and anomaly detection on the operating status data to generate the initial work order task.

[0063] The thermal inertia module is used to construct a time-series equipment deterioration model and calculate the cumulative deterioration function of HVAC equipment in real time, generate quantifiable values ​​of equipment health, and calculate the predicted time consumption for repairing abnormal states of HVAC equipment based on the building thermal inertia physical model, and convert the predicted time consumption into a relaxation time constraint index.

[0064] The target optimization module is used to judge the initial work order tasks, equipment health quantification values, and relaxation time constraint indicators, and automatically insert predictive maintenance work orders. Based on the work order task optimization algorithm, it iteratively solves the updated task set and relaxation time constraint indicators to generate the optimal dispatch sequence.

[0065] The mobile execution module is used to receive and send the optimal work order sequence and relaxation time constraint index to the maintenance technician terminal. After the work order is completed, it collects actual maintenance feedback data and uses the actual maintenance feedback data to reverse correct the equipment deterioration model and the building thermal inertia physical model.

[0066] The event triggering module includes:

[0067] Obtain unstructured repair request text input by the user, and use a natural language processing algorithm model to map the unstructured repair request text into a work order feature vector;

[0068] Calculate the cosine similarity between the work order feature vector and the feature vector of the i-th standard fault template in the fault knowledge base. Traverse the fault knowledge base and perform cosine similarity judgment on the work order feature vectors:

[0069] The label corresponding to the standard fault template feature vector with the largest cosine similarity and greater than the preset semantic matching threshold is selected as the standard fault label.

[0070] Collect operating status data of HVAC equipment, including temperature, pressure, sound signature and vibration data, and record the data corresponding to the current sampling time as a real-time status vector;

[0071] Retrieve the fault-free baseline state vector corresponding to the HVAC equipment, calculate the residual vector between the real-time state vector and the baseline state vector, construct a residual evaluation function based on a weighted sliding window, and quantify the severity of the equipment's deviation from the normal baseline.

[0072] Set a dynamic early warning threshold for the residual evaluation function and construct an event triggering logic mechanism. When an event is triggered, activate the event and generate an initial work order task. The event determination includes a first determination and a second determination.

[0073] The first determination includes:

[0074] When the work order feature vector successfully passes the cosine similarity test and generates the corresponding standard fault label, the trigger event is successfully activated.

[0075] The second determination includes:

[0076] When the equipment deviation value calculated by the residual evaluation function exceeds the dynamic early warning threshold, it is determined that the data anomaly triggering condition is met.

[0077] The trigger event is successfully activated when at least one of the first or second conditions is met.

[0078] When the trigger event is successfully activated, the corresponding standard fault label and running status data are packaged and encapsulated to automatically generate the initial work order task.

[0079] In this embodiment, for text-based repair requests, the system uses a pre-trained NLP model to map fuzzy natural language into a high-dimensional work order feature vector, and matches standard tags in the knowledge base using cosine similarity. Secondly, at the device's underlying level, the system collects real-time operating status data such as temperature and voiceprint, i.e., real-time status vectors, and uses a residual evaluation function... Calculate the weighted sliding window residuals:

[0080] ;

[0081] in, The set length of the sliding window, This is a time-weighted coefficient that decays over time, used to assign higher weights to more recent sampling points. This is a positive definite diagonal weighting matrix used to adjust for dimensional differences in sensor data across different dimensions. is the transpose of the residual vector;

[0082] The above-mentioned approach, by introducing a weighted sliding window and a time-weighted coefficient that decays over time, aims to assign higher weights to the latest sensor sampling points, thus preventing historical stable data from diluting transient anomalies. The introduction of a positive definite diagonal weighted matrix is ​​intended to remove the dimensional differences between temperature and pressure.

[0083] The state thermal inertia module includes a model building unit and a time prediction unit;

[0084] The model building unit is used to receive operating status data, build a time-series equipment deterioration model, calculate the cumulative deterioration function of HVAC equipment in real time, and generate quantified values ​​of equipment health.

[0085] Receive the running status data from the event triggering module, extract the sound and vibration data from the running status data, calculate the dynamic deterioration rate of the HVAC equipment at the current moment, and construct a time-series equipment deterioration model;

[0086] Based on historical operating loads, the nonlinear exponential parameters of equipment degradation are obtained, and the deterioration rate of HVAC equipment since the last maintenance is calculated in real time. up to the current moment The cumulative deterioration function is obtained, and the cumulative deterioration function is normalized to obtain the quantified value of equipment health.

[0087] In this embodiment, the current time is extracted. The root mean square (RMS) eigenvalues ​​of the acoustic and vibration data sequences are calculated. The RMS value directly represents the energy magnitude of mechanical vibration. Taking a screw chiller unit in a shopping mall as an example, its average mean time between failures (MTBF) recorded in the digital twin ledger is 10,000 hours. The equipment's basic aging constant is set to 1 / 10,000, and the baseline RMS reference value for normal operation is 2.0 mm / s. When slight wear occurs in the bearings, the real-time RMS value spikes to 5.0 mm / s. Through the constructed exponential degradation rate formula, the current dynamic degradation rate will be... The exponential amplification reflects the accelerated wear and tear on the equipment. Different types of equipment have vastly different tolerances to overload operation. By retrieving historical fault sample sets and extracting the cumulative effective operating time and average load rate of similar equipment before downtime, for example, if a certain type of water pump operates at 120% load for a long time, its lifespan is often not linearly shortened by 20%, but may be shortened by 50%. By performing logarithmic regression fitting on the load rate and lifespan of massive historical samples, the slope can be calculated, which is the deterioration nonlinearity exponent. , The higher the value, the more sensitive the equipment is to overload;

[0088] Substituting the aforementioned dynamic degradation rate, the actual power, the rated power, and the degradation nonlinearity exponent calculated from pressure data based on the pressure ratio work principle into the cumulative degradation function... :

[0089] ;

[0090] in, The dynamic deterioration rate over time. To calculate the first based on pressure data Actual operating power at any given time Rated power, The deterioration nonlinear exponent parameter is based on the equipment type;

[0091] The dynamic deterioration rate of the equipment is directly mapped to vibration data and acoustic signature signals. The specific calculation process for the dynamic deterioration rate is as follows: extract the operating status data... A sequence of acoustic signature signals and vibration data containing M sampling points at any given time. Obtain the root mean square eigenvalues ​​under the fault-free baseline condition: The root mean square eigenvalue can be directly represented as the energy magnitude of mechanical vibration. Then, the root mean square reference value under fault-free conditions is obtained. and equipment foundation aging constant The basic aging constant of this equipment is set as the reciprocal of the mean time between failures (MTBF) for the current equipment type, and the formula for calculating the dynamic degradation rate is constructed as follows:

[0092] ;

[0093] in, It is an exponential function with the natural constant e as its base. This is a vibration sensitivity scaling factor set based on the equipment's factory specifications.

[0094] Based on the above formula and calculation principle, taking a screw chiller unit in a shopping mall as an example, its average mean time between failures (MTBF) registered in the digital twin ledger is 10,000 hours. Therefore, the equipment's basic aging constant... The baseline root mean square (RMS) reference value for normal unit operation is set to 2.0 mm / s. When slight bearing wear occurs, the real-time RMS value is... Increasing to 5.0 mm / s, substituting into the dynamic deterioration rate formula, we find the deviation term is 1.5. The current dynamic deterioration rate... The magnification accurately reflects the accelerated wear and tear on the equipment.

[0095] Furthermore, different types of equipment have completely different tolerances to overload operation. This embodiment retrieves a historical fault sample set and extracts the cumulative effective operating time and average load rate of similar equipment before it fails. For example, if a certain type of water pump operates at 120% load for a long time, its lifespan is often not linearly shortened by 20%, but may be shortened by 50%. By performing logarithmic regression fitting on the load rate and lifespan of a massive amount of historical samples, the slope of the downward curve is solved, which yields the equipment's deterioration nonlinearity index. The larger the deterioration nonlinearity index value, the more sensitive the equipment is to overload.

[0096] Rated power As a static physical parameter, it is read from the equipment digital twin ledger database, and its value corresponds to the standard full-load operating power constant specified on the HVAC equipment's factory nameplate.

[0097] The process of processing the deteriorating nonlinear exponent parameter is as follows: retrieve the historical fault sample set of the same type of equipment, extract the cumulative effective operating time before the actual fault of sample j and the average load rate in the current period, and the average load rate is the ratio of the average actual power to the rated power; in this embodiment, an extremely low health quantification value is output, which facilitates the determination of active maintenance and saves maintenance costs.

[0098] The time prediction unit includes:

[0099] Extract temperature data from the operational status data, including the actual indoor temperature. and outdoor ambient temperature A physical model of building thermal inertia based on a first-order equivalent thermal capacity and thermal resistance network is established. The thermodynamic differential equation of the physical model of building thermal inertia is as follows:

[0100] ;

[0101] in, The equivalent heat capacity of the building. The equivalent thermal resistance of the building shell. The internal heat load generated by indoor personnel and equipment, This refers to the actual net heat removed from the room by the HVAC equipment under the current abnormal condition, that is, numerically, under the support condition. Take the positive value, under heating conditions Take negative value

[0102] The predicted time is obtained by calculating the time it takes for the indoor temperature to rise from the current actual temperature to the critical temperature within the user's comfort range based on the thermodynamic differential equation.

[0103] Obtain the scheduling safety buffer constant, and subtract the predicted time from the scheduling safety buffer constant to obtain the relaxation time constraint index.

[0104] In this embodiment, the physical thermal inertia of building space is transformed into time tolerance in operations research scheduling. The specific processing and calculation are as follows:

[0105] A physical model of building thermal inertia is constructed using a first-order equivalent thermal capacity and thermal resistance network. The loaded building is abstracted as a thermodynamic RC network. The principle of the thermodynamic differential equation is: the rate of change of heat inside the building = outdoor heat conducted through the walls - heat offset by HVAC equipment + heat dissipated from the interior.

[0106] This embodiment provides the following example to address the problem of predicting the time consumption and obtaining the relaxation time constraint index from the safety buffer constant:

[0107] At 2 p.m. on a hot summer afternoon, a main chiller unit in a large commercial complex suddenly experienced a load reduction failure, meaning that the cooling capacity provided was drastically reduced. At this time, the outdoor temperature was 35 degrees Celsius, and the indoor temperature was 22 degrees Celsius. The mall was crowded with people, resulting in a high heat load inside. The user's complaint threshold (the critical temperature of the comfort range) was 26 degrees Celsius.

[0108] Solving the above thermodynamic differential equations, given the massive size of the shopping mall and the use of high-quality insulated curtain walls, it was calculated that the indoor temperature would actually take 4.5 hours to slowly rise from 22°C to 26°C. This 4.5 hours is the predicted time.

[0109] To prevent sudden surges in passenger traffic Due to the surge in uncertainties, the system deducts a preset 0.5-hour scheduling safety buffer constant and finally outputs 4 hours as the relaxation time constraint index for this work order.

[0110] This embodiment uses a first-order equivalent thermal capacity and thermal resistance network instead of a complex three-dimensional computational fluid dynamics simulation. The purpose is to reduce cloud computing power consumption, achieve millisecond-level real-time solutions, meet the low latency requirements of dynamic scheduling, and achieve higher accuracy than traditional manual experience estimation.

[0111] The introduction of a scheduling safety buffer constant aims to increase the algorithm's fault tolerance and absorb minor disturbance errors in the real environment. It breaks away from the rigid order dispatching mindset of the traditional after-sales service industry based on hard contract time. By utilizing the inherent cooling and heating storage functions of buildings, it innovatively introduces spatial environmental attributes into the human resource scheduling model, achieving technological integration.

[0112] The objective optimization module includes a predictive maintenance decision-making unit, a constraint modeling unit, and a variable neighborhood optimization unit;

[0113] The predictive maintenance decision unit is used to determine the quantitative values ​​of equipment health and insert predictive maintenance work orders;

[0114] Receive device health quantification values, set security alert thresholds, and establish a threshold-triggered comparison mechanism:

[0115] When the measured health value of the equipment is less than the safety warning threshold, there is a risk of sudden downtime of the current HVAC equipment. A predictive maintenance work order is generated, and the predictive maintenance work order and the initial work order task are merged to update the task queue and generate an updated task set.

[0116] The core task of this embodiment is to determine whether to forcibly intervene and add preventative tasks to the scheduling system based on the current health check report of the equipment. The specific processing and judgment mechanism is as follows:

[0117] The system receives the percentage-based health metrics of upstream devices. System administrators or maintenance experts set differentiated security alert thresholds for different devices based on their criticality level (e.g., critical units are set to 80 points, and ordinary units are set to 60 points). The system backend scans the health of all devices across the network frequently and performs comparisons and judgments in the form of backend microservices.

[0118] In a large office building, tenant A had just reported a "leaking fan coil unit on the 21st floor". At the same time, a background threshold scan revealed that the No. 3 chiller unit in the basement, which is responsible for the overall cooling of the building, had just dropped to 58 points due to overload caused by continuous high temperatures, triggering the condition of being below the safety warning threshold.

[0119] Therefore, it was determined that Unit 3 was highly likely to experience a sudden shutdown within the next 48 hours. At this point, without any manual intervention, a predictive maintenance work order for bearing lubrication and load checks of Unit 3 was automatically generated. Subsequently, this work order was merged and reorganized with tenant A's repair request work order, updating the task queue in the underlying database and generating an updated task set containing these two attribute tasks. By quantifying equipment risk and intervening before a shutdown, the incidence of catastrophic failures of large HVAC equipment was reduced, avoiding economic losses to commercial customers due to equipment downtime. The system backend, in the form of microservices, frequently scans the health of all network devices and performs comparative judgments.

[0120] The constraint modeling unit is used to construct a scheduling optimization mathematical model based on the updated task set and relaxation time constraint index;

[0121] Obtain the set of maintenance workers to be scheduled. For the updated task set, define decision variables, including path assignment variables and time node variables:

[0122] The path assignment variable is 1 when employee k executes two or more elements of the updated task set consecutively; otherwise, it is 0.

[0123] Iterate through the updated task set, obtain the actual start time and actual completion time of each subtask, and set a latest start time constraint boundary. Current scheduling time Relaxation time constraint index;

[0124] A fitness objective function is constructed based on scheduling resources and timeout penalties. The optimized value of the objective function includes a first optimized value and a second optimized value. The first optimized value is the movement and travel time cost of all workers in each subtask, and the second optimized value is the cumulative time penalty for all tasks violating the latest start time limit of thermal inertia.

[0125] In this embodiment, the set of maintenance technicians currently on duty is obtained. For the updated task set that includes repair work orders and predicted work orders, this embodiment defines two types of key decision variables: path assignment variables. The time node variable and path assignment variable are used to describe the spatial movement trajectory of the mechanic. When mechanic k completes task i and goes directly to perform task j, =1, otherwise set to 0; the time node variable represents the specific timeline that the system tracks and records for each task, let's set it to 1. The actual planned start time for task i. Its actual completion time;

[0126] The relaxation time constraint index calculated based on the building's physical characteristics is directly converted into the hard time window boundary in the algorithm;

[0127] Current time The relaxation time calculated for task i in a shopping mall at 9:00 AM. If the time limit is 3 hours, the algorithm forces the latest start time boundary for task i to be 12:00, meaning that the inequality constraint must be satisfied in the modeling: ;

[0128] The expression for calculating the fitness objective function in this embodiment is as follows:

[0129] ;

[0130] in, This indicates that the optimization direction of the algorithm is to find the combination of variables that minimizes the function value Z (total cost).

[0131] The first optimal value, This represents the spatial travel time cost from task location i to task location j, which only occurs when the technician actually performs tasks i and j consecutively (i.e., ...). When the distance is equal to 1, this distance will be included in the total cost. This prompts the algorithm to find a route that minimizes the total distance traveled by all technicians.

[0132] It is the second optimal value, the penalty for breach of contract, and the weight. and To balance the priorities of saving on road costs and arriving on time, it is usually set to ;

[0133] This embodiment constructs a standard simulation test set containing 100 concurrent maintenance tasks and 10 maintenance technicians, and introduces algorithm convergence time, average passage time, and relaxation time default rate as evaluation indicators. The experimental comparison data under different weight ratios are shown in the table below: Table 1

[0134] Based on the in-depth analysis of the experimental data in Table 1 above and the determination of the optimal parameters, when the low penalty ratio of group A or group B is adopted, although the algorithm can find the route with the shortest average travel time per person, the algorithm will easily sacrifice the customer's scheduled on-site arrival time in exchange for the technician to travel less, since the travel cost and the default cost are close in magnitude. This results in a default rate of up to 28.5% during the relaxation time, which is an unacceptable serious service incident in HVAC after-sales business.

[0135] When the severe penalty ratio of group C is adopted, the default rate drops precipitously to 0.5%. Only a very small number of extreme tasks that are mathematically unsolvable experience minor delays. At this time, the soft time window penalty term acts as a hard boundary blocking effect. The algorithm will optimize the path under the premise of strictly ensuring that it will never exceed the time limit. At the same time, the convergence time of the algorithm under this ratio can still meet the computing power requirements of the cloud platform for real-time scheduling.

[0136] When the extreme penalty ratio of group D is adopted, although the default rate fails to improve further, the rapid increase in the penalty term causes the gradient cliff of the genetic algorithm's fitness function in the solution space. A large number of populations are eliminated during the initial crossover and mutation, making the algorithm prone to getting stuck in a dead state. This results in a dramatic increase in the convergence time to 18.7 seconds, which seriously hinders the system's real-time response capability.

[0137] In summary, this embodiment typically sets the weight parameters to satisfy... Set the weight ratio to =1, =1000, this parameter combination achieves the minimum commuting cost of HVAC services while taking into account the convergence speed of the cloud-based solver engine.

[0138] The above content combines the vehicle routing problem with the physical time sequence problem in the HVAC field, establishing a unique benchmark for the entire intelligent dispatch engine to solve the problem. By quantitatively constructing an objective function that includes the first and second optimization values, the system solves the problem of chaotic scheduling by manual dispatchers in traditional dispatching. It can not only generate a solution that takes into account the overall commuting efficiency in an instant, but also strictly adhere to the bottom line of maintenance service quality through timeliness penalty items, achieving dual optimization of operating costs and customer satisfaction.

[0139] The neighborhood optimization unit is used to iteratively solve the work order task optimization algorithm to generate the optimal dispatch sequence.

[0140] A genetic algorithm with a fusion variable neighborhood search mechanism is adopted as the task optimization algorithm. The updated task set is encoded as an initial population chromosome, which includes task sorting and worker allocation. The fitness objective function of each individual in the population is calculated, and global iterative evolution is performed. After each generation of evolution, elite individuals in the population are extracted for variable neighborhood local search. The variable neighborhood local search includes first neighborhood action and second neighborhood action.

[0141] The first neighborhood actions include:

[0142] Identify the subtask with the smallest relaxation time constraint index among the current elite individuals, and place the technician to which the current subtask belongs at the first position in the task execution sequence;

[0143] The second neighborhood actions include:

[0144] Identify predictive maintenance work orders in the updated task set, traverse the initial work order tasks with the same building number, and force the initial work order tasks to be assigned to the same maintenance technician for continuous operation.

[0145] When the number of iterations reaches the preset maximum convergence algebra, the iteration is terminated, the globally optimal individual with the minimum fitness objective function is decoded and output, the globally optimal individual is transformed into a task execution order and personnel matching scheme, and the optimal dispatch sequence is output.

[0146] This embodiment mainly involves the solution engine of the target optimization module. It adopts a genetic algorithm with a fusion variable neighborhood search mechanism to find the optimal task assignment sequence and convert the updated task set into a mathematical code that the genetic algorithm can recognize. It adopts a two-segment integer encoding mechanism. Each chromosome individual contains two gene segments: the sorting gene segment and the allocation gene segment. The sorting gene segment represents the absolute order in which the tasks are executed; the allocation gene segment represents the technician number assigned to the task corresponding to the sequence.

[0147] The system randomly generates an initial population and uses the fitness objective function Z constructed in the previous embodiment as the evaluation criterion. To transform the minimization problem into a maximization problem, the fitness evaluation formula for an individual is set as follows: :

[0148] ;

[0149] in, To prevent extremely small constants with a denominator of zero, crossover and mutation are then performed based on the roulette wheel algorithm, and a global search is conducted. After each generation of evolution, a simple genetic algorithm is prone to getting trapped in local optima. Therefore, the system extracts the fitness of the population. The top-ranking individuals perform local neighborhood actions on their gene chains based on HVAC business logic:

[0150] The first neighborhood action is a pre-crisis action. The system scans all the tasks assigned to elite individuals, extracts the subtask with the smallest relaxation time constraint index, forcibly cuts the smallest subtask from its current position in the individual's assignment sequence, and inserts it into the first position of its corresponding technician task sequence, that is, the very beginning of the gene sequence chain.

[0151] The second neighborhood action is a co-location binding action, where the system searches the gene chain for predictive maintenance work orders. Introducing geospatial matching functions Extract the building number information of the predicted work order and iterate through all initial work order tasks. When the same geographical location is met, that is:

[0152] ;

[0153] The system forcibly performs gene recombination, and The assigned gene loci are overwritten as the same technician, and the two are forcibly bound together in the sorted gene segment so that their index positions are adjacent, ensuring that the same technician performs continuous work;

[0154] The termination condition is triggered when the algorithm reaches the preset maximum number of iterations for convergence, or when the optimal fitness of the population remains unchanged for multiple consecutive generations.

[0155] The final globally optimal individual is decoded, translated into specific personnel dispatch instructions, and the optimal dispatch sequence is output.

[0156] Conventional genetic algorithms are prone to generating a large number of invalid dead solutions when faced with scheduling models with extremely strong penalties, resulting in extremely slow convergence. This embodiment introduces first and second neighborhood actions in each generation. The first action is dedicated to eliminating timeout defaults, and the second action is dedicated to eliminating long-distance travel.

[0157] The first and second actions are deeply integrated into the underlying framework of the AI ​​algorithm engine through strict operator logic, so as to achieve the optimal scheduling results.

[0158] The mobile execution module includes a task guidance unit, a closed-loop feedback unit, and a reverse correction unit;

[0159] The task guidance unit is used to receive and send the optimal dispatch sequence and corresponding relaxation time constraint indicators to the maintenance technician's terminal to guide task execution;

[0160] Obtain the optimal dispatch sequence and the corresponding relaxation time constraint index, convert the optimal dispatch sequence into a task instruction package, and send it to the mobile terminal of the corresponding maintenance technician through the communication network. Based on the relaxation time constraint index, generate dynamic thermal inertia time on the mobile terminal.

[0161] Dynamic thermal inertia timing The update formula is:

[0162] ;

[0163] in, This refers to the initial absolute time at which the scheduling task is issued. The relaxation time constraint index, This refers to the current absolute moment.

[0164] The core task of this embodiment is to convert the numerical combinations calculated in the cloud into operational instructions and time prompts for employees. The specific processing steps are as follows:

[0165] Obtain the optimal global order sequence, reduce its dimensionality and split it, extract the task sequence belonging to a specific technician k, and package the task's geographical coordinates, fault tags, required spare parts list and other information into a task instruction package, and push it to the technician's smartphone or industrial PDA maintenance technician terminal through the 4G / 5G communication network.

[0166] Traditional dispatch systems typically impose a rigid, static time requirement, such as arrival within two hours of accepting an order. This system, however, generates a dynamic thermal inertia timeframe for each task on the mobile app interface, based on building physics properties calculated by preceding modules. This timeframe changes over time. Its update formula is

[0167] ;

[0168] at this time The absolute latest start time for this task has been defined. This point is the physical critical point at which the indoor temperature exceeds the user's tolerance limit and is about to trigger a complaint.

[0169] The complex equivalent RC calculation results of building thermodynamics are encapsulated into a simple countdown UI component, which intuitively allows you to perceive the delay time of the building's air conditioning failure.

[0170] This embodiment uses the terminal's local time for local difference calculation and updates. The intention is to ensure the countdown runs accurately even if the technician is in a network signal dead zone such as an underground server room, avoiding missed deadlines due to network outages. The optimal solution generated by the cloud algorithm relies on strict execution by on-site personnel. By demonstrating dynamic hot inertia timing, the technician's behavior is gently guided, prompting them to arrive at the site according to the algorithm's planned schedule.

[0171] By precisely timing the countdown, technicians can always arrive and repair equipment just before the indoor environment gets too hot and the user becomes anxious. This allows customers to smoothly navigate equipment downtime, greatly improving their SLA satisfaction with HVAC service providers and reducing technicians' anxiety and work fatigue.

[0172] The closed-loop feedback unit is used to collect actual maintenance feedback data after the work order is completed;

[0173] After the mobile terminal receives the work order completion instruction, it triggers the data collection instruction to obtain the actual maintenance feedback data. The actual maintenance feedback data includes the actual maintenance time and equipment recovery energy efficiency. The actual maintenance time is the absolute time difference from when the maintenance technician clicks to start the work on the mobile terminal to when the work ends.

[0174] The processing logic for restoring equipment energy efficiency is as follows:

[0175] After maintenance is completed and the HVAC system is restarted, the outlet and return water temperatures, flow rates, and actual power consumption of the HVAC system are collected within a duration window, and then analyzed using thermodynamic formulas. The actual performance coefficient after equipment repair was obtained through reverse calculation:

[0176] ;

[0177] Where c is the specific heat capacity of the fluid. For fluid density, To submit traffic, For the temperature difference between the inlet and outlet water, This represents the actual electrical power consumed.

[0178] In this embodiment, equipment energy efficiency is used as the standard for assessing maintenance quality. The purpose is to abandon the simple maintenance management in traditional business. Only through precise calculation of energy efficiency data can we objectively reflect whether the internal structure of the equipment has been thoroughly repaired.

[0179] The purpose of setting a duration window is that, in the initial stage of HVAC equipment startup, the energy efficiency is extremely unstable because the heat capacity and pressure in the system have not yet been balanced. Accurate data can only be collected after the system has passed the transient period and entered steady-state operation.

[0180] By directly capturing flow rate, temperature difference, and power consumption through underlying sensors to calculate energy efficiency, the possibility of technicians artificially inflating repair results on the App is eliminated, providing service providers with refined quality control over maintenance.

[0181] The reverse correction unit is used to reverse correct equipment deterioration models and building thermal inertia physical models using machine learning algorithms based on actual maintenance feedback data;

[0182] The actual maintenance feedback data is transmitted back to the cloud platform, where a first reverse correction and a second reverse correction are performed. The first reverse correction is used to correct the timing equipment degradation model.

[0183] Calculate the mean square error loss function between the system's predicted recovery energy efficiency before maintenance and the equipment's recovery energy efficiency; use the gradient descent algorithm to perform reverse iterative correction on the basic aging constant and deterioration nonlinear exponent parameters in the time-series equipment deterioration model to obtain the updated basic aging constant and deterioration nonlinear exponent parameters.

[0184] The second inverse correction is used to correct the thermal inertia physical model of the building:

[0185] The actual indoor temperature drop curve during the maintenance period is extracted and compared with the temperature curve predicted by the building's thermal inertia physical model to obtain the time-series temperature prediction residual. The recursive least squares algorithm is used to minimize the sum of squares of the cumulative temperature prediction residuals as the objective to perform reverse online identification and numerical correction on the equivalent heat capacity and equivalent thermal resistance in the building's thermal inertia physical model, resulting in updated equivalent heat capacity and equivalent thermal resistance.

[0186] In this embodiment, the core task is to receive the actual maintenance results collected in the preceding steps, eliminate the prediction deviation between the theoretical model and the real physical world through algorithm self-learning, perform the first reverse correction to correct the time-series equipment deterioration model, and calculate the loss function. The system first extracts the constructed deterioration model, calculates the predicted recovery energy efficiency before maintenance based on the maintenance actions, compares it with the actual equipment recovery energy efficiency just collected from the field, and constructs the mean squared error (MSE) loss function:

[0187] ;

[0188] in, To restore the energy efficiency of the equipment, To predict energy efficiency recovery before maintenance,

[0189] If the energy efficiency of the equipment recovery is found to be lower than the predicted energy efficiency before maintenance, it means that the system previously underestimated the natural aging degree of the equipment. The gradient descent algorithm is then activated to calculate the partial derivatives of the loss function with respect to the basic aging constant and the deterioration nonlinear exponent parameter, respectively. The learning rate is applied along the negative gradient direction to make step adjustments, and the updated parameter values ​​are output.

[0190] The second reverse correction is to correct the building's thermal inertia physical model and obtain the time-series prediction residual: During the maintenance period, the indoor temperature must have shifted. The system extracts the actual indoor temperature drop / rise curve during this period and aligns it with the theoretical curve predicted based on the RC differential equation and subtracts it to obtain the time-series temperature prediction residual at each moment.

[0191] The principle of recursive least squares correction is that the built environment is a slowly changing system, and the system uses the RLS algorithm with a forgetting factor to process the residuals. The optimization objective is to minimize the cumulative sum of squared residuals, and the equivalent heat capacity and equivalent thermal resistance matrix vectors in the model are identified and updated online in the background.

[0192] Gradient descent is chosen for the device because the aging parameters of the device are highly nonlinear, making it suitable to use gradient-based deep learning optimization to gradually approach the global optimum.

[0193] The recursive least squares (RLS) method is chosen for the built environment because thermodynamic RC networks belong to classical linear time-invariant systems. RLS is the most mature, fastest-converging, and most computationally efficient algorithm in the field of system identification.

[0194] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0195] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0196] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0197] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A cloud-based intelligent scheduling system for HVAC equipment work orders, characterized in that, It includes an event triggering module, a state hot inertia module, a target optimization module, and a move execution module: The event triggering module is used to collect the operating status data of HVAC equipment and the unstructured repair report text input by the user, extract features from the unstructured repair report text to generate standard fault labels, and perform baseline comparison and anomaly detection on the operating status data to generate an initial work order task. The state thermal inertia module is used to construct a time-series equipment deterioration model, calculate the cumulative deterioration function of HVAC equipment in real time, generate a quantitative value of equipment health, calculate the predicted time consumption for repairing abnormal states of HVAC equipment based on the building thermal inertia physical model, and convert the predicted time consumption into a relaxation time constraint index. The target optimization module is used to judge the initial work order tasks, equipment health quantification values ​​and relaxation time constraint indicators, and automatically insert predictive maintenance work orders. Based on the work order task optimization algorithm, iteratively solves the updated task set and relaxation time constraint indicators to generate the optimal dispatch sequence. The mobile execution module is used to receive and send the optimal work order sequence and relaxation time constraint index to the maintenance technician terminal. After the work order is completed, it collects actual maintenance feedback data and uses the actual maintenance feedback data to reverse correct the equipment deterioration model and the building thermal inertia physical model.

2. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 1, characterized in that, The event triggering module includes: Obtain unstructured repair request text input by the user, and use a natural language processing algorithm model to map the unstructured repair request text into a work order feature vector; Calculate the cosine similarity between the work order feature vector and the feature vector of the i-th standard fault template in the fault knowledge base. Traverse the fault knowledge base and perform cosine similarity judgment on the work order feature vectors: The label corresponding to the standard fault template feature vector with the largest cosine similarity and greater than the preset semantic matching threshold is selected as the standard fault label. Collect operating status data of HVAC equipment, including temperature, pressure, acoustic signature and vibration data, and record the data corresponding to the operating status data at the current sampling time as a real-time status vector; Retrieve the fault-free baseline state vector, calculate the residual vector between the real-time state vector and the baseline state vector, construct the residual evaluation function, and quantify the severity of the equipment's deviation from the normal baseline. Set a dynamic early warning threshold for the residual evaluation function and construct an event triggering logic mechanism. When an event is triggered, activate the event and generate an initial work order task. The event determination includes a first determination, a second determination, and a third determination.

3. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 2, characterized in that, The state thermal inertia module includes a model building unit and a time prediction unit; The model building unit is used to receive operating status data, build a time-series equipment deterioration model, calculate the cumulative deterioration function of HVAC equipment in real time, and generate a quantitative value of equipment health. Receive the running status data from the event triggering module, extract the sound and vibration data from the running status data, calculate the dynamic deterioration rate of the HVAC equipment at the current moment, and construct a time-series equipment deterioration model; Based on historical operating loads, the nonlinear exponential parameters of equipment degradation are obtained, and the deterioration rate of HVAC equipment since the last maintenance is calculated in real time. up to the current moment The cumulative deterioration function is obtained, and the cumulative deterioration function is normalized to obtain the quantified value of equipment health.

4. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 3, characterized in that, The time consumption prediction unit includes: Extract temperature data from the operating status data, including the actual indoor temperature. and outdoor ambient temperature A physical model of building thermal inertia based on a first-order equivalent thermal capacity and thermal resistance network is established. The thermodynamic differential equation of the physical model of building thermal inertia is as follows: ； in, The equivalent heat capacity of the building. The equivalent thermal resistance of the building shell. The internal heat load generated by indoor personnel and equipment, The actual heating or cooling capacity provided by the HVAC equipment under the current abnormal condition; The predicted time is obtained by calculating the time it takes for the indoor temperature to rise from the current actual temperature to the critical temperature within the user's comfort range based on the thermodynamic differential equation. Obtain the scheduling safety buffer constant, and subtract the predicted time from the scheduling safety buffer constant to obtain the relaxation time constraint index.

5. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 4, characterized in that, The target optimization module includes a predictive maintenance decision-making unit, a constraint modeling unit, and a variable neighborhood optimization unit. The predictive maintenance decision unit is used to judge the quantitative value of equipment health and insert predictive maintenance work orders; Receive device health quantification values, set security alert thresholds, and establish a threshold-triggered comparison mechanism: When the measured health value of the equipment is less than the safety warning threshold, the current HVAC equipment is at risk of sudden downtime. A predictive maintenance work order is generated, and the predictive maintenance work order and the initial work order task are merged to update the task queue and generate an updated task set.

6. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 5, characterized in that, The constraint modeling unit is used to construct a scheduling optimization mathematical model based on the updated task set and relaxation time constraint index; Obtain the set of maintenance workers to be scheduled. For the updated task set, define decision variables, including path allocation variables and time node variables: When employee k executes two or more elements of the updated task set consecutively, the path assignment variable is 1; otherwise, it is 0. Iterate through the updated task set, obtain the actual start time and actual completion time of each subtask, and set a latest start time constraint boundary. Current scheduling time Relaxation time constraint index; A fitness objective function is constructed based on scheduling resources and timeout penalties. The optimized values ​​of the objective function include a first optimized value and a second optimized value.

7. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 6, characterized in that, The neighborhood optimization unit is used to iteratively solve the work order task optimization algorithm to generate the optimal dispatch sequence. A genetic algorithm employing a fusion variable neighborhood search mechanism is used as the task optimization algorithm. The updated task set is encoded into an initial population chromosome, which includes task sorting and worker allocation. The fitness objective function of each individual in the population is calculated, and global iterative evolution is performed. After each generation of evolution, elite individuals in the population are extracted for variable neighborhood local search. The variable neighborhood local search includes first neighborhood actions and second neighborhood actions. The first neighborhood action includes: Identify the subtask with the smallest relaxation time constraint index among the current elite individuals, and place the technician to which the current subtask belongs at the first position in the task execution sequence; The second neighborhood action includes: Identify predictive maintenance work orders in the updated task set, traverse the initial work order tasks with the same building number, and force the initial work order tasks to be assigned to the same maintenance technician for continuous operation. When the number of iterations reaches the preset maximum convergence algebra, the iteration is terminated, the globally optimal individual with the minimum fitness objective function is decoded and output, the globally optimal individual is transformed into a task execution order and personnel matching scheme, and the optimal dispatch sequence is output.

8. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 7, characterized in that, The mobile execution module includes a task guidance unit, a closed-loop feedback unit, and a reverse correction unit; The task guidance unit is used to receive and send the optimal dispatch sequence and the corresponding relaxation time constraint index to the maintenance technician's terminal to guide task execution. Obtain the optimal dispatch sequence and the corresponding relaxation time constraint index, convert the optimal dispatch sequence into a task instruction package, and send it to the mobile terminal of the corresponding maintenance technician through the communication network. Based on the relaxation time constraint index, generate a dynamic thermal inertia timer on the mobile terminal. The dynamic thermal inertia timing The update formula is: ； in, This refers to the initial absolute time at which the scheduling task is issued. The relaxation time constraint index, This refers to the current absolute moment.

9. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 8, characterized in that, The closed-loop feedback unit is used to collect actual maintenance feedback data after the work order is completed; After the mobile terminal receives the work order completion instruction, it triggers the data collection instruction to obtain the actual maintenance feedback data. The actual maintenance feedback data includes the actual maintenance time and equipment recovery energy efficiency. The actual maintenance time is the absolute time difference from when the maintenance technician clicks to start the work on the mobile terminal to when the work ends. The processing logic for restoring the energy efficiency of the device is as follows: After the maintenance work is completed and the HVAC equipment is restarted, the outlet and return water temperatures, flow rates, and actual power consumption of the HVAC equipment are collected within the duration window to calculate the actual performance coefficient of the equipment after repair.

10. The intelligent scheduling system for HVAC equipment work orders based on a cloud platform as described in claim 9, characterized in that, The reverse correction unit is used to reverse correct the equipment deterioration model and the building thermal inertia physical model using machine learning algorithms based on actual maintenance feedback data; The actual maintenance feedback data is transmitted back to the cloud platform, where a first reverse correction and a second reverse correction are performed respectively. The first reverse correction is used to correct the timing equipment degradation model. Calculate the mean square error loss function between the system's predicted recovery energy efficiency before maintenance and the equipment's recovery energy efficiency; use the gradient descent algorithm to perform reverse iterative correction on the basic aging constant and deterioration nonlinear exponent parameters in the time-series equipment deterioration model to obtain the updated basic aging constant and deterioration nonlinear exponent parameters. The second reverse correction is used to correct the building's thermal inertia physical model: The actual indoor temperature drop curve during the maintenance period is extracted and compared with the temperature curve predicted by the building's thermal inertia physical model to obtain the time-series temperature prediction residual. The recursive least squares algorithm is used to minimize the sum of squares of the cumulative temperature prediction residuals as the objective to perform reverse online identification and numerical correction on the equivalent heat capacity and equivalent thermal resistance in the building's thermal inertia physical model, resulting in updated equivalent heat capacity and equivalent thermal resistance.

Images