Mechanical and electrical equipment load operation and maintenance management method and system

Abstract

The present application relates to the technical field of industrial automation control and equipment intelligent operation and maintenance, in particular to a mechanical and electrical equipment load operation and maintenance management method and system, comprising: collecting real-time state and total load demand of a group of mechanical and electrical equipment; mapping the state to thermodynamic credit by using a loss mapping model to calculate marginal loss cost; constructing a reverse bidding game model based on the cost, taking the minimization of total marginal loss of the group as the goal under the premise of meeting the demand, and solving the optimal load distribution ratio; generating control instructions accordingly to drive each sub-equipment to execute load adjustment; the present application eliminates the calculation deadlock between fine life management and real-time load response, ensures that the system can respond to the instantaneous mutation of the load within microseconds, and realizes high real-time of control decision.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance management technology, specifically to a method and system for operation and maintenance management based on the load of electromechanical equipment. Background Technology

[0002] With the deep integration of the Industrial Internet and digital management technologies, the operation and management of electromechanical equipment groups has transformed from traditional equipment status monitoring to data-driven asset value lifecycle management. The core objective is to maximize asset utilization and reduce overall operating costs through optimized resource scheduling strategies. However, in existing equipment group operation and maintenance management systems, the decision-making logic for task load allocation often lacks refined consideration based on economic benefits and depreciation costs, generally adopting simple average allocation strategies or static scheduling modes based on a single fixed rule. This resource scheduling method, lacking differentiated value assessment, ignores the significant heterogeneity of individual equipment in terms of remaining useful life value, health depreciation rate, and operation and maintenance costs, and cannot effectively calculate the marginal depreciation costs generated by different equipment undertaking specific load tasks. This leads to assets in the system at high risk of depreciation depreciating rapidly due to overloading, triggering a bottleneck effect in asset management. This not only increases unexpected maintenance expenditures but also prevents the optimal release of the overall lifecycle value of the asset group.

[0003] On the other hand, while some advanced management solutions attempt to incorporate lifetime prediction models for refined decision optimization, these solutions typically rely on high-dimensional nonlinear programming algorithms or complex iterative calculations. This leads to problems such as high computational resource consumption and excessive time consumption when the management system handles high-frequency concurrent load allocation requests. This computational contradiction between the need for refined cost management and the real-time task response speed results in significant lag in scheduling decisions, making it difficult to complete cost-optimal resource allocation within microsecond-level time windows. Consequently, it fails to meet the dual requirements of real-time performance and cost-effectiveness for high-dynamic load response in digital management platforms.

[0004] Furthermore, most existing load management models are static open-loop systems, lacking a dynamic value calibration mechanism based on actual operational data. As equipment operates over time, pre-defined mathematical models fail to perceive the evolution of equipment physical characteristics, leading to a gradual distortion in the accuracy of decision-making models' representation of aging characteristics and wear and tear costs. This, in turn, results in biased long-term operation and maintenance management decisions. Therefore, how to construct an adaptive load management system that balances real-time computation with the economic efficiency of lifespan losses, and how to achieve the optimal solution for cluster resource allocation by introducing dynamic bidding or game theory mechanisms, is a pressing technical challenge in the field of electromechanical equipment operation and maintenance management. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for load operation and maintenance management of electromechanical equipment, which avoids the bottleneck effect caused by traditional average distribution strategies and can achieve adaptive load distribution that minimizes the total marginal loss of the equipment cluster based on the real-time health status of the equipment; specifically, the technical solution of this invention is as follows:

[0006] Based on the electromechanical equipment load operation and maintenance management method, running in a management system including a processor and memory, the method includes:

[0007] Obtain real-time operating status data and current total load demand for each sub-equipment in the electromechanical equipment group;

[0008] Using a pre-set loss mapping model, real-time operating status data is mapped to thermodynamic credit limits, and the marginal loss cost of each sub-device for a unit of new load is calculated based on the thermodynamic credit limits.

[0009] Based on the marginal loss cost value, an inverse bidding game model is constructed. Under the premise of meeting the current total load demand, the optimal load allocation ratio of each sub-device is calculated with the goal of minimizing the total marginal loss of the cluster.

[0010] Based on the optimal load distribution ratio, corresponding control commands are generated and sent to the execution controllers of each sub-device to drive each sub-device to perform load adjustment.

[0011] Among them, the marginal loss cost value represents the lifespan reduction cost incurred by a sub-device under its current health condition when it undertakes an additional unit load.

[0012] Optional, real-time operating status data includes vibration frequency data, current waveform data, and temperature rise rate data;

[0013] The method of mapping real-time operating status data to thermodynamic credit limits using a pre-set loss mapping model includes:

[0014] The vibration frequency data, current waveform data, and temperature rise rate data are normalized respectively to construct a multidimensional state vector.

[0015] The multidimensional state vector is input into the preset health potential energy function to calculate the current health potential energy value of each sub-device;

[0016] The current health potential value is compared with the preset benchmark health potential value to generate the corresponding thermodynamic credit limit.

[0017] The thermodynamic credit limit is positively correlated with the current health potential value, and the thermodynamic credit limit is forcibly reset to zero when the current health potential value is lower than the preset critical threshold.

[0018] Optionally, methods for calculating the marginal loss cost per unit of additional load for each sub-device based on thermodynamic credit limits include:

[0019] Calculate the partial derivative of the health potential energy function with respect to the load variable to obtain the instantaneous loss change rate of each sub-device;

[0020] Obtain the historical cumulative wear of each sub-device, and generate a non-linear weighting coefficient based on the historical cumulative wear.

[0021] Multiply the instantaneous rate of change of loss by the nonlinear weighting coefficient and divide by the thermodynamic credit limit to obtain the marginal loss cost value.

[0022] Specifically, when the thermodynamic credit limit reaches zero, the corresponding marginal loss cost value is set to a preset infinite penalty value to exclude the corresponding sub-device from accepting orders in subsequent games.

[0023] Optionally, methods for constructing a reverse bidding game model to calculate the optimal load allocation ratio for each sub-device include:

[0024] In response to the arrival of the current total load demand, a microsecond-level reverse auction is initiated; each sub-device is used as an intelligent agent node to generate bidding data based on its own marginal loss cost value;

[0025] Sort all bidding data from low to high values ​​to generate a cost sorting sequence; based on the cost sorting sequence, use a greedy selection strategy to sequentially accumulate the load capacity of each sub-device until the accumulated value equals the current total load demand.

[0026] The optimal load allocation ratio is generated by dividing the load capacity of each selected sub-device by the current total load demand.

[0027] The time complexity of the greedy selection strategy is limited to logarithmic linear level to ensure real-time response to load changes.

[0028] Optionally, the method also includes dynamic constraint processing on the reverse bidding game model:

[0029] Get the most recent running duration of each sub-device;

[0030] If the duration of the most recent operation exceeds the preset fatigue threshold, the marginal loss cost value of the corresponding sub-device will be temporarily increased in this reverse auction by introducing a preset fatigue penalty factor.

[0031] If the duration of the most recent operation does not exceed the preset fatigue threshold, the marginal loss cost value of the corresponding sub-device remains unchanged.

[0032] By temporarily increasing the marginal loss cost value, the position of the fatigued sub-equipment in the cost ranking sequence is moved to the back, thereby reducing the probability of it being assigned load and realizing the active maintenance adjustment of the machine group.

[0033] Optionally, the method also includes establishing a closed-loop feedback mechanism for data flow:

[0034] After load adjustment, the actual wear increment of each sub-device is collected in real time;

[0035] The prediction deviation rate is calculated using the actual wear increment and the marginal loss cost value;

[0036] If the prediction error rate is greater than the preset error tolerance threshold, the weight parameters in the loss mapping model are corrected by backpropagation using the actual wear increment.

[0037] If the prediction deviation rate is less than or equal to the preset deviation tolerance threshold, then keep the current parameters of the loss mapping model unchanged.

[0038] By correcting through backpropagation, the accuracy of the loss mapping model in representing the aging characteristics of each sub-device dynamically converges with operating time.

[0039] Optionally, the electromechanical equipment group is a parallel equipment group with redundant configuration, and the control commands include inverter frequency setpoints or PLC start / stop signals;

[0040] The method for generating corresponding control commands includes: obtaining the rated power parameters of each sub-device; multiplying the optimal load allocation ratio by the current total load demand to obtain the target load value of each sub-device;

[0041] Using a preset power-control conversion table, the target load value is converted into the corresponding inverter frequency setting value or PLC start / stop signal;

[0042] Check whether the converted control command is within the safe operating range of each sub-device; if it is within the safe operating range, send the control command; if it is outside the safe operating range, trigger the preset fuse protection logic.

[0043] Based on the electromechanical equipment load operation and maintenance management system, including:

[0044] The multi-dimensional sensing module is used to acquire real-time operating status data of each sub-device and the current total load demand through a sensor array deployed in the electromechanical equipment group;

[0045] The mapping calculation module is used to map real-time operating status data into thermodynamic credit limits using a pre-set loss mapping model, and to calculate the marginal loss cost value of each sub-device for a unit of new load.

[0046] The game scheduling module is used to build a reverse bidding game model based on the marginal loss cost value. Under the premise of meeting the current total load demand, the optimal load allocation ratio of each sub-device is calculated with the goal of minimizing the total marginal loss of the cluster.

[0047] The execution control module is used to generate corresponding control commands based on the optimal load distribution ratio and send the control commands to the execution controllers of each sub-device.

[0048] Optionally, the game scheduling module includes:

[0049] The bidding engine unit is used to respond to load requests and organize the agent nodes corresponding to each sub-device to bid based on the marginal loss cost value.

[0050] The sorting and matching unit is used to sort the bids and generate a list of equipment combinations that meet the total demand based on a greedy algorithm.

[0051] The strategy generation unit is used to convert the device combination list into specific load allocation instructions and send them to the execution control module.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1. This invention effectively extends the overall service life of a group of electromechanical equipment by constructing a marginal loss cost model and a reverse bidding mechanism. The method abandons the average allocation strategy, maps multi-dimensional operating states to thermodynamic credit limits, and calculates the lifespan loss cost of each piece of equipment bearing the additional load. Through differentiated bidding, equipment with good health and low marginal cost bears more load, while sub-healthy equipment automatically reduces its load. This adaptive allocation mechanism of "the capable do more" avoids the accelerated system collapse caused by the weakest link effect and significantly improves the mean time between failures of the group.

[0054] 2. This invention employs a reverse bidding game model based on a greedy strategy, solving the problem that complex algorithms cannot meet real-time requirements. Addressing the issues of long computation time and difficulty in handling load fluctuations in traditional nonlinear programming algorithms, this application transforms the complex scheduling problem into a low-time-complexity sorting and greedy selection process. This computational architecture limits decision-making time to an extremely short range, eliminating computational deadlock between refined lifetime management and real-time load response, ensuring that the system can respond to instantaneous load fluctuations within microseconds, and achieving high real-time performance in control decisions.

[0055] 3. This invention establishes a multi-dimensional state perception and data flow closed-loop feedback mechanism to ensure the accuracy of the decision model throughout its entire lifecycle; by integrating multi-source heterogeneous data such as vibration, current, and temperature rise, it constructs a comprehensive equipment health profile, avoiding the one-sidedness of monitoring a single indicator; at the same time, it uses actual wear increments to backpropagate and correct the model parameters, enabling the loss mapping model to dynamically converge as the equipment aging characteristics evolve; it overcomes the defect of static models gradually becoming distorted over time, ensuring the accuracy of decisions under long-term operation.

[0056] 4. This invention integrates active health regulation and multiple circuit breaker protection logic, which greatly improves the robustness and safety of the system. By introducing a fatigue penalty factor, it forces equipment that has been running for a long time to move to the back of the bidding sequence to obtain a cooling opportunity, thus preventing the accumulation of thermal fatigue. At the same time, it sets a critical threshold for health potential energy and a safe operating range. When the equipment is in an abnormal state, it forces the credit limit to zero or triggers circuit breaker protection. This mechanism rigidly eliminates the risk of equipment operating with defects at the algorithm level, and realizes automated fault isolation and risk avoidance without manual intervention. Attached Figure Description

[0057] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0058] Figure 1 This is a flowchart of the method of the present invention;

[0059] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0061] Example 1:

[0062] Please see Figure 1 A method for the operation and maintenance management of electromechanical equipment load, running in a management system that includes processors and memory, includes the following:

[0063] Obtain real-time operating status data and current total load demand for each sub-equipment in the electromechanical equipment group;

[0064] Using a pre-set loss mapping model, real-time operating status data is mapped to thermodynamic credit limits, and the marginal loss cost of each sub-device for a unit of new load is calculated based on the thermodynamic credit limits.

[0065] Based on the marginal loss cost value, an inverse bidding game model is constructed. Under the premise of meeting the current total load demand, the optimal load allocation ratio of each sub-device is calculated with the goal of minimizing the total marginal loss of the cluster.

[0066] Based on the optimal load distribution ratio, corresponding control commands are generated and sent to the execution controllers of each sub-device to drive each sub-device to perform load adjustment.

[0067] Among them, the marginal loss cost value characterizes the lifespan reduction cost of the sub-equipment bearing an additional unit load in the current health state; it should be noted that the thermodynamic evaluation index in the embodiments of this application, including thermodynamic credit limit and marginal loss cost value, refers to the dimensionless statistical mapping index based on multidimensional operating state data, which is used to characterize the rate of increase of system disorder and wear risk, rather than the physical entropy value based on heat flux density integral in the traditional thermodynamic definition.

[0068] This embodiment details the execution architecture of the electromechanical equipment load operation and maintenance management method based on the spatiotemporal mapping of thermodynamic evaluation indicators and inverse game theory;

[0069] This method runs in a management system that includes high-performance processors, such as industrial-grade edge computing gateways or cloud servers and storage, and acts as the central brain of the electromechanical equipment group to solve the computational deadlock problem between refined life management and real-time load response in traditional operation and maintenance.

[0070] The system performs multi-dimensional state perception steps by capturing real-time operating status data of each sub-device in the electromechanical equipment group at a high frequency sampling rate through sensor arrays deployed on various sub-devices, such as cooling water pump groups in data centers or elevator groups in high-rise buildings; at the same time, it obtains the current total load demand from the upper-level control system, such as BMS or DCS, through the industrial bus.

[0071] Real-time operating status data: The source is the instantaneous physical quantity acquisition of the sensor array. The physical meaning is the original signal reflecting the physical health status of the equipment, and the unit is the corresponding unit of each physical quantity.

[0072] Current total load demand: originates from real-time user requests, and physically represents the total workload that the entire cluster must handle at the current moment, measured in units of... or ;

[0073] The system performs a thermodynamic space mapping step, using a pre-set loss mapping model to map heterogeneous real-time operating status data into a unified evaluation index—thermodynamic credit limit; based on this, the system further calculates the marginal loss cost value of each sub-device for a unit of new load.

[0074] The core of this step is to eliminate the difference in physical dimensions and transform the nonlinear physical aging characteristics into a linear economic carrying capacity indicator.

[0075] Loss mapping model: It originates from a pre-trained mathematical transformation function and has the physical meaning of a transformation operator from physical state to economic carrying capacity;

[0076] Thermodynamic credit limit: It is derived from the results of mapping calculations, and its physical meaning is the current remaining health capital of the equipment, with the unit being credit points;

[0077] Marginal loss cost value: It is derived from differential calculation. Its physical meaning is the life loss cost of a sub-device bearing one additional unit of load in its current healthy state, which is the growth rate of the thermodynamic evaluation index. The unit is cost / unit load.

[0078] The system executes the reverse bidding game step, and builds a reverse bidding game model in memory based on the calculated marginal loss cost value. In this model, each sub-device is instantiated as an intelligent agent, which performs a reverse auction for the current load demand.

[0079] The goal of the system is to find the Pareto optimal solution under the hard constraint of meeting the current total load demand, so as to minimize the total marginal loss of the cluster and thus calculate the optimal load allocation ratio of each sub-device.

[0080] The system executes closed-loop drive control steps, generates corresponding control commands based on the calculated optimal load distribution ratio, and sends them to the execution controllers of each sub-device, such as PLCs or frequency converters, via the industrial bus to drive the physical devices to perform load adjustment.

[0081] This embodiment breaks down the barrier between the physical loss model and the real-time scheduling algorithm by introducing marginal loss cost as an intermediate variable. In highly dynamic scenarios such as electromechanical equipment groups, this solution establishes a strong correlation between equipment health status and load allocation strategy in real time, so that equipment with poor health status and high marginal loss cost automatically reduces its load, while equipment with good health status and low marginal loss cost takes on more load.

[0082] This mechanism achieves adaptive allocation of tasks based on merit at the micro level, and effectively extends the mean time between failures (MTBF) of the entire fleet at the macro level, avoiding the bottleneck effect that accelerates system collapse caused by traditional average allocation strategies.

[0083] Example 2:

[0084] Real-time operating status data includes vibration frequency data, current waveform data, and temperature rise rate data; methods for mapping real-time operating status data to thermodynamic credit limits using a pre-set loss mapping model include:

[0085] The vibration frequency data, current waveform data, and temperature rise rate data are normalized to construct a multidimensional state vector; the multidimensional state vector is input into a preset health potential energy function to calculate the current health potential energy value of each sub-device;

[0086] The current health potential value is compared with the preset benchmark health potential value to generate the corresponding thermodynamic credit limit. The thermodynamic credit limit is positively correlated with the current health potential value, and the thermodynamic credit limit is forcibly reset to zero when the current health potential value is lower than the preset critical threshold.

[0087] This embodiment further refines the specific process of data mapping, aiming to construct an accurate equipment health profile; the real-time operating status data covers vibration frequency data reflecting mechanical wear, current waveform data reflecting electrical stress, and temperature rise rate data reflecting the degree of thermal aging, achieving comprehensive coverage of the physical state of the equipment;

[0088] When using a pre-defined loss mapping model, the system performs interval normalization on the three types of heterogeneous data to eliminate the influence of dimensions and adapt to the domain of the subsequent potential energy function; the specific processing formula is as follows:

[0089] ;

[0090] in, These are the instantaneous measured values ​​of physical quantities such as vibration, current, or temperature rise collected in real time by the sensor. The preset alarm limit value, This is the rated value; it should be noted that this is in response to the calculated value. If the value is greater than 1, the system forces it to be assigned the value 1, thereby constructing a multidimensional state vector with values ​​between 0 and 1. ;

[0091] Input the vector into the preset health potential function In the middle, calculate each sub-device Current health potential value The calculation formula is as follows:

[0092] ;

[0093] The source is real-time calculation, the physical meaning is to characterize the current physical robustness of the device, and the unit is dimensionless;

[0094] The data source is the state components obtained by normalizing vibration frequency data, current waveform data, and temperature rise rate data, ensuring consistency with the data types in the embodiments.

[0095] The source is a constant preset by the Equipment Failure Mode and Effects Analysis (FMEA), and its physical meaning is the weighting coefficient of each state component, and it satisfies... ;

[0096] Based on this, the system will use the current health potential value Compared with the preset baseline health potential value A comparison is performed to generate a thermodynamic credit limit. The calculation process includes a key nonlinear cutoff logic:

[0097] ;

[0098] ;

[0099] The source is calculated, and the physical meaning is the immediate capital that the equipment can use to pay for wear and tear, measured in credit points;

[0100] The source is the equipment's factory technical specifications, and the physical meaning is the potential energy benchmark of the equipment under ideal conditions;

[0101] The source is a preset value, and its physical meaning is the normalized scaling factor;

[0102] The source is the minimum physical limit for safe operation of equipment, such as the ISO standard limit for bearing vibration intensity, and its physical meaning is the fusing threshold.

[0103] This embodiment comprehensively captures the mechanical, electrical, and thermodynamic aging characteristics of equipment by constructing a multi-dimensional state vector, avoiding the one-sidedness of monitoring a single indicator; in particular, by introducing This circuit breaker mechanism forcibly resets the thermodynamic credit limit to zero when the device's health potential energy falls below a critical threshold. This directly disqualifies the device from bidding in subsequent game rounds, rigidly eliminating the risk of operating with defects from an algorithmic perspective and achieving proactive safety protection without human intervention.

[0104] Example 3:

[0105] Methods for calculating the marginal loss cost per unit of additional load for each sub-device based on thermodynamic credit limits include:

[0106] Calculate the partial derivative of the health potential energy function with respect to the load variable to obtain the instantaneous loss change rate of each sub-device;

[0107] Obtain the historical cumulative wear of each sub-equipment, and generate a nonlinear weighting coefficient based on the historical cumulative wear. Multiply the instantaneous loss change rate by the nonlinear weighting coefficient and divide by the thermodynamic credit limit to obtain the marginal loss cost value.

[0108] Specifically, when the thermodynamic credit limit reaches zero, the corresponding marginal loss cost value is set to a preset infinite penalty value to exclude the corresponding sub-device from accepting orders in subsequent games.

[0109] This embodiment details the specific calculation logic of the marginal loss cost value, which serves as the sole benchmark for subsequent bidding. This process calculates transient characteristics based on a physical model and then corrects them by combining historical factors and current credit.

[0110] Specifically, in order to establish the differential relationship between health status and load, the system pre-fits the load-state response function of each sub-device using historical operating data. Considering the state vector Including multiple heterogeneous components such as vibration, current, and temperature rise, the system establishes an independent quadratic polynomial regression model for each normalized state component k:

[0111] ;

[0112] in, For the corresponding state components The independent fitting coefficient vectors, representing the quadratic coefficients, are in units of: , coefficient of the linear term, unit: And the constant term coefficients, which are dimensionless, thus forming a multidimensional state vector. A coefficient matrix with consistent dimensions; This is the load variable; it should be noted that for newly commissioned sub-equipment without historical operating data, the system directly loads the benchmark fitting coefficients pre-stored in memory, generated based on the factory test data of the same model of equipment. As initial values ​​are used, the system periodically updates the above coefficients using the online least squares method based on a preset length sliding time window as equipment operation data accumulates, thus achieving a smooth transition from a general model to an individualized model.

[0113] Based on this, the system uses the chain rule to calculate the health potential function relative to the load variable. For example, the partial derivative of power:

[0114] ;

[0115] in, The health potential function;

[0116] This allows us to obtain the instantaneous loss change rate of each sub-device under the current operating conditions. ;

[0117] This parameter reflects the differential characteristic of how much more wear will occur when a slightly larger load is applied at the physical level;

[0118] At the same time, the historical cumulative wear and tear of each sub-device is read from the database. This value represents the actual wear increment calculated since the equipment was put into operation. The summation is dimensionless, and a nonlinear weighting coefficient is generated using the Sigmoid function. The specific formula is as follows:

[0119] ;

[0120] Marginal loss cost value The calculation formula is as follows:

[0121] ;

[0122] The source is a formula calculation, the physical meaning is the basis for the bidding ranking of sub-devices, and the unit is cost / unit load;

[0123] The source is the differential calculation of load based on real-time status data; its physical meaning is the instantaneous rate of change of loss; and its unit is potential energy loss / unit load.

[0124] The source is historical data mapping, and the physical meaning is a weighting coefficient that increases with the increase of historical cumulative wear.

[0125] The source is a preset constant, and its physical meaning is the amplification gain factor of the aging effect;

[0126] The source is a preset constant, and its physical meaning is the steepness coefficient of the Sigmoid function, which is used to adjust the sensitivity to the wear threshold;

[0127] The source is the equipment life manual, and its physical meaning is the half-life wear threshold;

[0128] The coefficients are derived from regression analysis of historical load and state data using the least squares method, representing the coefficients of the quadratic, linear, and constant terms, respectively.

[0129] The source is the specific components in the multidimensional state vector, such as the normalized vibration frequency or the rate of temperature rise.

[0130] The source is a preset constant, and its physical meaning is to prevent the removal of the zero minimum value;

[0131] In particular, in response to thermodynamic credit limits The system is forced to reset to zero due to excessively low health potential. Specifically, when the thermodynamic credit limit reaches zero, the corresponding marginal loss cost value is set to a preset infinite penalty value; for example, the maximum value that a floating-point number can represent in a computer system, or a value artificially set far exceeding the maximum marginal loss cost that might occur during normal calculations. Order of magnitude constant;

[0132] This embodiment cleverly constructs a price leverage mechanism: equipment with high health and large credit limits has its marginal cost diluted, resulting in lower bids and easier bidding; while equipment with low health or severe historical wear and tear has its marginal cost amplified, resulting in higher bids and difficulty in winning bids.

[0133] In particular, the setting of an infinite penalty value ensures that faulty or extremely aged equipment is automatically excluded from the order queue in a mathematical and logical manner, achieving automated fault isolation and risk avoidance in complex machine cluster networks.

[0134] Example 4:

[0135] Methods for constructing an inverse bidding game model and calculating the optimal load allocation ratio for each sub-device include:

[0136] In response to the arrival of the current total load demand, a microsecond-level reverse auction is initiated; each sub-device is used as an intelligent agent node to generate bidding data based on its own marginal loss cost value;

[0137] Sort all bidding data from low to high values ​​to generate a cost sorting sequence; based on the cost sorting sequence, use a greedy selection strategy to sequentially accumulate the load capacity of each sub-device until the accumulated value is greater than or equal to the current total load demand.

[0138] The optimal load allocation ratio is generated by dividing the load capacity of each selected sub-device by the current total load demand.

[0139] The time complexity of the greedy selection strategy is limited to log-linear level to ensure real-time response to load mutations.

[0140] This embodiment describes the solution process of the reverse bidding game model. This process abandons the traditional complex iterative optimization algorithm and instead adopts a deterministic greedy selection strategy to cope with the high real-time requirements of sudden load changes in electromechanical equipment.

[0141] The specific steps are as follows: Responding to the current total load demand Upon arrival, the system initiates a microsecond-level reverse auction; at this time, each sub-device, acting as an intelligent agent node, calculates the marginal loss cost value based on Example 3. Generate bidding data;

[0142] The system collects bidding data from all online devices and quickly sorts them by value from low to high, for example, using the QuickSort algorithm to generate a cost-sorted sequence. ;

[0143] Based on this sequence, the algorithm iterates through the devices sequentially; for the first [item] in the sequence... Each device, summed up its load capacity , The value is taken as the upper limit of the rated power of the sub-device, up to the cumulative value. If the accumulated value is still less than 0 after traversing all devices. If it does, a full-load operation command is generated and a load reduction alarm is triggered; otherwise, the selected first... The selected device is the winning combination; to ensure the accuracy of load allocation and conform to the calculation logic of the embodiment, for a selected device in the sorting sequence, i.e., the first... Each device, whose capacity to calculate the proportion is dynamically truncated to the current total load demand and the previous... The difference between the cumulative values ​​of each device is the remaining demand, while the previous The load capacity of each device is taken as its rated upper limit;

[0144] Divide the effective load capacity of each sub-device after the above correction by the current total load demand. This generates the optimal load distribution ratio for each sub-device, at which point the sum of the ratios is strictly equal to 1;

[0145] This embodiment reduces a complex nonlinear programming problem to a simple list sorting problem by using marginal cost sorting; this greedy selection strategy strictly limits the computational time complexity to the log-linear level. This allows the system to complete scheduling calculations in milliseconds or even microseconds when facing large-scale clusters, such as hundreds of server fans. This technical feature effectively solves the engineering problem that traditional optimization algorithms take too long to calculate, resulting in control lag and inability to respond to sudden changes in load.

[0146] Example 5:

[0147] The method also includes dynamic constraint processing on the reverse bidding game model:

[0148] Get the most recent running duration of each sub-device;

[0149] In response to the fact that the duration of the most recent operation exceeded the preset fatigue threshold, the marginal loss cost value of the corresponding sub-equipment was temporarily increased in this reverse auction by introducing a preset fatigue penalty factor.

[0150] In response to the fact that the duration of the most recent operation did not exceed the preset fatigue threshold, the marginal loss cost value of the corresponding sub-device remains unchanged; by temporarily increasing the marginal loss cost value, the position of the sub-device in the cost ranking sequence is moved to the back, thereby reducing the probability of it being assigned load, and realizing the active maintenance adjustment of the machine group.

[0151] This embodiment introduces a dynamic constraint processing mechanism into the game model, aiming to prevent thermal fatigue from occurring in the equipment even if it is in good condition due to long-term continuous operation.

[0152] The system obtains the most recent operating duration of each sub-device in real time. ; in response Exceeding the preset fatigue threshold For example, if the equipment has been running continuously for 4 hours, the system determines that it is in a state of sub-fatigue; in this case, the system introduces a preset fatigue penalty factor in this reverse auction. The marginal loss cost of the equipment is temporarily adjusted using the following formula: ; in response Not exceeding Then keep ;in, A preset coefficient greater than 1, for example, a value between 1.5 and 2.0, is used to artificially amplify the bidding cost of fatigue equipment;

[0153] The revised bidding data will force the fatigued equipment to move down in the cost ranking sequence; this means that unless the total load demand is extremely high and all equipment has to be used, the fatigued equipment will be rejected due to its high bid, thus gaining the opportunity to be shut down for cooling.

[0154] This embodiment simulates the pain protection mechanism of an organism; by temporarily increasing the price, the system realizes the active health regulation of the machine group, avoiding the premature performance degradation of a single health device due to long-term continuous full load, and realizing the combination of work and rest and wear balance within the machine group, fundamentally improving the overall service life of the system.

[0155] Example 6:

[0156] The method also includes establishing a data flow closed-loop feedback mechanism: after load adjustment, the actual wear increment of each sub-device is collected in real time;

[0157] The prediction error rate is calculated by using the actual wear increment and the predicted wear increment reconstructed based on the instantaneous wear change rate;

[0158] Specifically, the system extracts the instantaneous loss change rate generated when the instruction was previously calculated. That is, removing the thermodynamic credit limit. and historical weighting coefficient The resulting pure physical differential term, combined with the actual load of this instruction execution. Through formula The predicted wear increment is calculated; this value is a dimensionless value that reflects the decline in health potential.

[0159] Simultaneously, the system collects the integral value of vibration energy of each sub-device during load regulation. Integral value of current heating effect and temperature rise integral value In order to obtain objective physical truth values ​​as a calibration benchmark, the actual wear increment... The calculation does not include the weight parameters to be optimized in the loss mapping model. Instead, it directly calculates the arithmetic mean based on the physical energy integral value and a fixed material fatigue constant to eliminate differences in physical dimensions and maintain consistent magnitudes.

[0160] ;

[0161] in, , and These are the reciprocal constants corresponding to vibration, current, and temperature rise data, respectively, calibrated based on the equipment's ultimate tolerance capability, used to eliminate differences in physical dimensions; the specific calibration method is as follows: take the cumulative vibration energy integral limit value corresponding to when the equipment reaches the end of its design life or experiences a critical failure. ,set up Similarly, set and ;in, This is the cumulative current thermal effect integral limit value corresponding to when the equipment reaches the end of its design life or experiences a critical failure. This is the cumulative temperature rise integral limit value corresponding to when the equipment reaches the end of its design life or experiences a critical failure.

[0162] Specifically, the system calculates the gradient of the prediction bias with respect to the weight parameters. To reduce the complexity of online computation, this embodiment uses a first-order Taylor expansion approximation and ignores the credit limit. The weight coupling effect when used as the denominator simplifies the gradient to:

[0163] ;

[0164] in, The normalized state value for the corresponding dimension characterizes the sensitivity of this state component to the health potential. Based on this gradient, if the prediction deviation rate exceeds a preset deviation tolerance threshold, it indicates that the weight allocation of each state component in the preset loss mapping model can no longer accurately reflect the true characteristics of the equipment. The system utilizes the gradient difference between the actual wear increment and the predicted value to adjust the weight parameters in the loss mapping model. Perform backpropagation correction;

[0165] In response to a prediction deviation rate less than or equal to a preset deviation tolerance threshold, the current parameters of the loss mapping model remain unchanged; through backpropagation correction, the accuracy of the loss mapping model in representing the aging characteristics of each sub-device dynamically converges with operating time.

[0166] This embodiment establishes a data flow closed-loop feedback mechanism. Within a preset time window after load adjustment, the system collects the actual wear increment of each sub-device in real time through sensors. This value can be approximated by the vibration energy integral or the temperature rise integral; simultaneously, the system reads the marginal loss cost value estimated when calculating this instruction and converts it into a predicted wear increment of the same dimension. ;

[0167] Based on this, the system calculates the prediction deviation rate:

[0168] ;

[0169] In response to Greater than the preset deviation tolerance threshold This indicates that the current loss mapping model can no longer accurately reflect the true aging characteristics of the equipment, such as when the equipment suffers unexpected mechanical damage. In this case, the system triggers a backpropagation correction process, utilizing... As labeled data, the weight parameters in the loss mapping model are fine-tuned using gradient descent, as shown in Example 2. This makes the model's predicted output approximate the true value;

[0170] To ensure that the corrected weight parameters always satisfy the normalization constraints of Example 2, i.e. After each gradient update, the system performs projection normalization on the weight vector:

[0171] Calculate the updated temporary weights ;in, The preset backpropagation learning rate has a value range of [value range missing]. ;

[0172] Calculate the final weight ;

[0173] This ensures that the sum of the weight coefficients of each state component remains equal to 1 throughout the iteration process;

[0174] This embodiment solves the problem of static models gradually becoming distorted over time. Through continuous experimental feedback and backpropagation correction, the accuracy of the loss mapping model in representing the aging characteristics of sub-devices will dynamically converge as the running time increases, enabling the system to adapt to changes in the characteristics of the entire life cycle of the equipment and ensuring the accuracy of decision-making under long-term operation.

[0175] Example 7:

[0176] The electromechanical equipment group is a parallel equipment group with redundant configuration, and the control commands include frequency converter frequency setpoints or PLC start / stop signals;

[0177] Methods for generating corresponding control commands include:

[0178] Obtain the rated power parameters of each sub-device; multiply the optimal load allocation ratio by the current total load demand to obtain the target load value of each sub-device;

[0179] Using a preset power-control conversion table, the target load value is converted into the corresponding inverter frequency setting value or PLC start / stop signal;

[0180] Check whether the converted control commands are within the safe operating range of each sub-device; if they are within the safe operating range, send control commands.

[0181] In response to exceeding the safe operating range, the preset circuit breaker protection logic is triggered, automatically clamping the instruction to the safe boundary value or directly issuing a shutdown instruction, and sending an alarm to the operation and maintenance terminal;

[0182] At the same time, the system performs the load gap compensation step:

[0183] Calculate the load gap value that was not triggered due to fuse protection. That is, the difference between the target load value and the actual clamp output value;

[0184] judge Is it greater than the preset dead zone threshold? If so, then... As a new total load demand, a local reverse bidding game is re-initiated in the set of remaining sub-devices that have not triggered the circuit breaker protection. The steps of Example 4 are repeated to allocate the shortfall load to other healthy devices to ensure that the total power output of the cluster always meets the user's request.

[0185] This embodiment describes the specific execution process of converting mathematical operation results into physical actions, and is applicable to parallel equipment groups with redundant configurations, such as multiple variable frequency water pumps connected in parallel.

[0186] When generating control commands, the system obtains the rated power parameters of each sub-device. Based on the aforementioned optimal load distribution ratio and current total load demand Calculate the target load value ;in, For the first Optimal load distribution ratio for each sub-device;

[0187] Using a preset power-control quantity conversion table, which stores the corresponding curves of power and frequency or speed, you can look up... The corresponding inverter frequency setting or PLC start / stop signal;

[0188] Before issuing instructions, the system must perform a safe operating range check; check whether the converted frequency value is within the range of the minimum lubrication frequency to the maximum rated frequency;

[0189] In response to being within a safe operating range, commands are sent via industrial Ethernet;

[0190] In response to exceeding the safe operating range, such as a frequency below 20Hz causing the inability to form an oil film, or a frequency above 55Hz causing overspeed, the preset fuse protection logic is triggered, automatically clamping the instruction to the safe boundary value or directly issuing a shutdown instruction, and sending an alarm to the operation and maintenance terminal.

[0191] This embodiment ensures that the economically optimal solution output by the algorithm must also be a physically safe solution. By adding physical constraint checks and circuit breaker protection, it prevents physical damage to the equipment caused by extreme optimization of the algorithm or sensor drift, such as dry running or overspeeding. While pursuing efficiency, it builds a physical defense for the robustness of the control system.

[0192] Example 8:

[0193] Please see Figure 2 An electromechanical equipment load operation and maintenance management system, used to implement the electromechanical equipment load operation and maintenance management method as described in any one of Examples 1-7, comprising:

[0194] The multi-dimensional sensing module is used to acquire real-time operating status data of each sub-device and the current total load demand through a sensor array deployed in the electromechanical equipment group;

[0195] The mapping calculation module is used to map real-time operating status data into thermodynamic credit limits using a pre-set loss mapping model, and to calculate the marginal loss cost value of each sub-device for a unit of new load.

[0196] The game scheduling module is used to build a reverse bidding game model based on the marginal loss cost value. Under the premise of meeting the current total load demand, the optimal load allocation ratio of each sub-device is calculated with the goal of minimizing the total marginal loss of the cluster.

[0197] The execution control module is used to generate corresponding control commands based on the optimal load distribution ratio and send the control commands to the execution controllers of each sub-device.

[0198] The game scheduling module includes:

[0199] The bidding engine unit is used to respond to load requests and organize the agent nodes corresponding to each sub-device to bid based on the marginal loss cost value.

[0200] The sorting and matching unit is used to sort the bids and generate a list of equipment combinations that meet the total demand based on a greedy algorithm.

[0201] The strategy generation unit is used to convert the device combination list into specific load allocation instructions and send them to the execution control module.

[0202] This embodiment provides an electromechanical equipment load operation and maintenance management system for implementing the above method; the system adopts a design that combines distributed deployment and centralized decision-making in its hardware architecture;

[0203] The multidimensional sensing module consists of MEMS vibration sensors, Hall current sensors, and thermocouple arrays installed in key parts such as the stator and bearing housing. It is responsible for capturing physical signals at a high-frequency sampling rate to ensure the real-time performance and accuracy of the data source.

[0204] The mapping computation module is deployed in the edge computing gateway and has a built-in high-performance floating-point unit (FPU) responsible for performing feature extraction, potential energy function calculation and marginal cost generation, converting physical signals into economic indicators.

[0205] The game scheduling module is the core decision-making unit of the system, comprising three sub-units: the bidding engine unit, which instantiates the Agent objects for each device and generates bids; the sorting and matching unit, which executes the quicksort algorithm and the greedy selection strategy; and the game scheduling module, which executes the quicksort algorithm and the greedy selection strategy. The time complexity is at the microsecond level, generating a list of device combinations that meets the total requirements; the strategy generation unit is used to transform the abstract combination list into specific physical control parameters.

[0206] The execution control module includes the local PLC and frequency converter of each sub-device. They receive instructions from the strategy generation unit and drive the motor to perform actual actions.

[0207] This embodiment decouples complex sensing, computing, decision-making and execution functions through a modular hardware architecture design; in particular, it separates high-frequency data processing from logical decision-making, which not only ensures accurate capture of physical state, but also ensures agile generation of scheduling strategies, perfectly supporting the physical implementation of the method of this invention in industrial sites and realizing intelligent operation and maintenance through software and hardware collaboration.

[0208] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for operation and maintenance management based on a load of an electromechanical device, characterized by, The method runs in a management system comprising a processor and a memory, and comprises: obtaining real-time running state data and current total load demand of each sub-device in the electromechanical device group; mapping the real-time running state data into thermodynamic credit by using a preset wear mapping model, and calculating marginal wear cost values of each sub-device for unit new load based on the thermodynamic credit; based on the marginal wear cost values, constructing a reverse bidding game model, and solving optimal load distribution ratios of each sub-device under the premise of meeting the current total load demand and with the goal of minimizing the total marginal wear of the device group; generating corresponding control instructions according to the optimal load distribution ratios, and sending the control instructions to the execution controllers of each sub-device to drive each sub-device to execute load adjustment; wherein the marginal wear cost value represents the life loss cost of each sub-device in the current health state for bearing an additional unit load; the real-time running state data comprises vibration frequency data, current waveform data and temperature rise rate data; the method of mapping the real-time running state data into thermodynamic credit by using a preset wear mapping model comprises: normalizing the vibration frequency data, the current waveform data and the temperature rise rate data respectively to construct a multi-dimensional state vector; inputting the multi-dimensional state vector into a preset health potential function to calculate the current health potential value of each sub-device, which represents the physical robustness of the device; comparing the current health potential value with a preset baseline health potential value to generate a corresponding thermodynamic credit; wherein the thermodynamic credit is positively correlated with the current health potential value, and the thermodynamic credit is forced to zero when the current health potential value is lower than the preset critical threshold; the method of calculating the marginal wear cost values of each sub-device for unit new load based on the thermodynamic credit comprises: calculating the partial derivative of the health potential function with respect to the load variable to obtain the instantaneous wear change rate of each sub-device; obtaining the historical cumulative wear of each sub-device, and generating a nonlinear weighting coefficient according to the historical cumulative wear; multiplying the instantaneous wear change rate by the nonlinear weighting coefficient, and dividing by the thermodynamic credit to obtain the marginal wear cost value; wherein when the thermodynamic credit is zero, the corresponding marginal wear cost value is set to a preset infinite penalty value to exclude the corresponding sub-device from the subsequent game.

2. The method according to claim 1, wherein, The method of constructing a reverse bidding game model to solve the optimal load distribution ratios of each sub-device comprises: in response to the arrival of the current total load demand, initiating a microsecond-level reverse auction activity; regarding each sub-device as an intelligent agent node, and generating bidding data according to the marginal wear cost value of each sub-device; sorting all the bidding data from low to high to generate a cost sorting sequence; according to the cost sorting sequence, using a greedy selection strategy to accumulate the bearable load of each sub-device in turn until the cumulative value equals the current total load demand; dividing the bearable load of each selected sub-device by the current total load demand to generate the optimal load distribution ratio; The time complexity of the greedy selection strategy is limited to a logarithmic linear level to ensure real-time response to load mutations.

3. The method according to claim 2, wherein, The method further comprises performing dynamic constraint processing on the reverse bidding game model: acquiring the last running duration of each sub-device; if the last running duration exceeds a preset fatigue threshold, temporarily increasing the marginal wear cost value of the corresponding sub-device by introducing a preset fatigue penalty factor in the current reverse auction activity; if the last running duration does not exceed the preset fatigue threshold, keeping the marginal wear cost value of the corresponding sub-device unchanged; temporarily increasing the marginal wear cost value to make the sub-device in the fatigue state move backward in the cost sorting sequence, thereby reducing the probability of being allocated load and achieving active health regulation of the device group.

4. The method according to claim 1, wherein, The method further comprises establishing a data flow closed-loop feedback mechanism: after performing load regulation, real-time collection of actual wear increments of each sub-device; calculating a prediction deviation rate using the actual wear increment and the marginal wear cost value; if the prediction deviation rate is greater than a preset deviation tolerance threshold, performing back propagation correction of the weight parameters in the wear mapping model using the actual wear increment; if the prediction deviation rate is less than or equal to the preset deviation tolerance threshold, keeping the current parameters of the wear mapping model unchanged; through back propagation correction, the wear mapping model dynamically converges the accuracy of representing the aging characteristics of each sub-device with running time.

5. The method according to claim 1, wherein, The electromechanical device group is a parallel device group with redundant configuration, and the control instruction includes a frequency setting value of a frequency converter or a PLC start-stop signal; The method of generating the corresponding control instruction comprises: acquiring the rated power parameters of each sub-device; multiplying the optimal load distribution ratio by the current total load demand to obtain the target load value of each sub-device; using a preset power-control amount conversion table to convert the target load value into the corresponding frequency setting value of the frequency converter or the PLC start-stop signal; checking whether the converted control instruction is within the safe operation interval of each sub-device; if it is within the safe operation interval, sending the control instruction; if it exceeds the safe operation interval, triggering a preset fuse protection logic.

6. A system for managing and maintaining electromechanical equipment loads, for implementing the method for managing and maintaining electromechanical equipment loads according to any one of claims 1 to 5, characterized in that, It comprises: a multi-dimensional perception module for acquiring real-time running state data of each sub-device and current total load demand through a sensor array deployed in the electromechanical device group; a mapping calculation module for mapping the real-time running state data into thermodynamic credit quota using a preset wear mapping model, and calculating the marginal wear cost value of each sub-device for unit incremental load; a game scheduling module for constructing a reverse bidding game model based on the marginal wear cost value, and solving the optimal load distribution ratio of each sub-device to minimize the total marginal wear of the device group under the premise of meeting the current total load demand; an execution control module for generating corresponding control instructions according to the optimal load distribution ratio and sending the control instructions to the execution controller of each sub-device.

7. The electromechanical device load operation and maintenance management system according to claim 6, wherein The game scheduling module comprises: a bidding engine unit for responding to load requests and organizing corresponding agent nodes of each sub-device to make bids based on the marginal wear cost value; A sorting and matching unit is configured to sort the bids and generate a device combination list satisfying the total demand according to a greedy algorithm. A strategy generating unit is configured to convert the device combination list into specific load distribution instructions and send the instructions to the execution control module.

Images