Energy saving management method and system based on electromechanical device power consumption prediction

Abstract

This invention discloses an energy-saving management method and system based on power consumption prediction of electromechanical equipment, belonging to the field of data processing technology. The invention calculates the instantaneous total power; calculates the theoretical baseline power based on the partial load factor command value; combines the instantaneous total power to obtain the instantaneous power residual; obtains a multi-dimensional statistical feature vector through segmentation and feature extraction; maps a state label sequence based on a health state feature library to construct a state transition probability matrix; calculates the KL divergence with the health baseline matrix as the implicit loss divergence; when the divergence is greater than a trigger threshold, constructs a virtual dissipation damping term; combines a unidirectional anti-mechanical shock penalty function and an equipment health power consumption characteristic function to construct a total cost function; iteratively solves the optimal partial load factor solution set under the constraint of total output demand and issues control commands. This invention can accurately identify equipment implicit losses and achieve energy-saving optimization control while ensuring mechanical safety.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an energy-saving management method and system based on power consumption prediction of electromechanical equipment. Background Technology

[0002] In actual electromechanical equipment operation scenarios, multiple similar devices are usually connected in parallel to meet the dynamically changing total output demand of the system. The core purpose of energy-saving management for electromechanical equipment groups is to rationally allocate the total output demand to each device according to the power consumption characteristics of each device, that is, to determine the partial load rate of each device, so as to minimize the total input power of the equipment group while ensuring the total effective output of the system. Patent document with application publication number CN120631121A discloses a cloud computing-based data center energy-saving control method and system. By acquiring real-time data and historical power consumption change trajectories of data center equipment, a basic data set including power peak, average, and fluctuation amplitude is established to obtain equipment operating parameter moments. By extracting key feature vectors, a multivariate regression model between equipment status and energy consumption level is constructed to determine the energy conversion efficiency coefficient of various types of equipment under different load conditions. An energy allocation optimization model is constructed, and the objective function is set as minimizing the overall energy consumption cost. The optimal power allocation scheme of each device is determined by calculation, and the energy allocation parameters are updated when abnormal fluctuations in the actual power consumption of the equipment are detected, resulting in an adaptive energy consumption management decision mechanism.

[0003] In terms of feature extraction and state assessment, the aforementioned existing technologies mainly rely on the macroscopic average power, peak power, and large-scale fluctuation amplitude of the equipment for efficiency modeling and state determination. They fail to separate the macroscopic power consumption fluctuations caused by load changes from the microscopic power consumption anomalies caused by internal physical conditions such as increased early mechanical friction and partial discharge. In the load allocation optimization mechanism, the existing technologies mainly take minimizing the current overall energy consumption cost as the single optimization objective and fail to introduce a damping mechanism for the health status of the equipment into the optimization strategy. If a high load is continuously allocated to equipment with hidden losses, it will accelerate its deterioration process and create potential faults. In the dynamic adjustment and execution stage, the existing solutions lack consideration for the mechanical safety boundaries of the equipment when triggering abnormal adjustments or redistributing loads. Rapid changes in commands often lead to instantaneous step jumps in the motor torque of the equipment, which in turn generates huge impact stress on mechanical components and easily causes fatigue damage to the mechanical structure of the equipment. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies are unable to detect the hidden losses of electromechanical equipment caused by early deterioration without eliminating the influence of macro loads. Furthermore, they lack a reasonable suppression mechanism for potentially deteriorated equipment when optimizing load distribution, and lack effective protection against mechanical shocks caused by sudden load increases. As a result, they cannot simultaneously achieve deep energy saving and long-term stable and safe operation of equipment groups.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an energy-saving management method based on electromechanical equipment power consumption prediction, comprising the following steps: Step S1: Collect the operating data, partial load rate command value, rated output and total output demand of the equipment group, and calculate the instantaneous total power of each equipment. Step S2: Calculate the theoretical reference power based on the partial load factor command value, calculate the instantaneous power residual by combining the instantaneous total power, segment and extract features from the instantaneous power residual, and obtain a multidimensional statistical feature vector. Step S3: Based on the pre-stored health status feature library, perform state mapping on the multidimensional statistical feature vector to obtain the state label sequence, and construct the state transition probability matrix based on the state label sequence; Step S4: Calculate the implicit loss divergence based on the state transition probability matrix and the pre-stored health baseline state transition probability matrix; Step S5: If the implicit loss divergence is greater than the preset divergence trigger threshold, then construct a virtual dissipation damping term based on the implicit loss divergence, and construct a total cost function by combining the unidirectional anti-mechanical shock penalty function and the equipment health power consumption characteristic function. Under the condition of satisfying the equality constraint of the total output demand, iteratively solve the total cost function to obtain the optimal partial load rate solution set and issue control commands.

[0006] Preferably, in step S1, the operating data includes voltage and current; The calculation process for obtaining the instantaneous total power is as follows: divide the sum of the three-phase products of voltage and current by the preset conversion factor; Rated output capacity represents the maximum effective output that the equipment can output per unit time under full load operation, while total output demand represents the total effective output that needs to be provided by the group of equipment.

[0007] Preferably, in step S2, the process of constructing a multidimensional statistical feature vector specifically includes: Call the pre-stored device health power consumption characteristic function. The device health power consumption characteristic function is a mapping function obtained by performing multinomial regression fitting on the factory test data of this model of equipment, with the partial load rate as the independent variable and the theoretical input power as the function value. Substitute the partial load rate command value into the device health power consumption characteristic function to obtain the theoretical input power under the partial load rate command value as the theoretical reference power. The difference between the instantaneous total power and the theoretical reference power is taken as the instantaneous power residual of the equipment, and the instantaneous power residuals obtained throughout the entire control cycle are used to form an instantaneous power residual sequence. The instantaneous power residual sequence is evenly divided into multiple segments according to a preset duration. The root mean square value, standard deviation, and kurtosis of all instantaneous power residuals in each segment are calculated to construct a multidimensional statistical feature vector.

[0008] Preferably, step S3, the process of obtaining the state label sequence, specifically includes: The process of retrieving the health status center vector of a device, which contains multiple health status center vectors for the same model of device, involves: The equipment is operated within the allowable load rate range, with the load rate gradually increasing from the lower limit to the upper limit in preset steps, and running stably for a preset time at each load rate. During the full-condition scan, the root mean square value, standard deviation, and kurtosis are calculated in real time to generate multiple multidimensional statistical feature vectors. A clustering algorithm with a preset number of clusters is executed on all generated feature vectors, and the centroid vector of each cluster is output as the health status center vector, and a status label is assigned to each health status center vector. For each segment feature vector calculated within the current control cycle, calculate its Euclidean distance with each health state center vector, select the health state center vector with the smallest distance value as the state label of the segment, and convert multiple multidimensional statistical feature vectors within a control cycle into a state label sequence. The process of constructing the state transition probability matrix specifically includes: Based on the state label sequence, the cumulative frequency of state labels at adjacent positions transitioning from the previous state to the next state is counted, the state transition probability is calculated, and a state transition probability matrix is ​​constructed.

[0009] Preferably, step S4, the process of obtaining the implicit loss divergence, specifically includes: The device's health baseline state transition probability matrix is ​​invoked. The health baseline state transition probability matrix is ​​constructed based on the operating data of the same model of device in its factory condition. Minimize all elements of the health baseline state transition probability matrix and renormalize it to obtain a smoothed baseline matrix. The KL divergence between the state transition probability matrix and the reference matrix of the current control cycle is calculated by summing each element, and is used as the implicit loss divergence.

[0010] Preferably, in step S5, the mathematical expression for the virtual dissipation damping term is: ; in, Let i be the virtual dissipation damping term for the i-th device. This is the damping weighting coefficient. Let be the implicit loss divergence of the i-th device. Let i be the current partial load factor of the i-th device. This is the preset low-load safety zone boundary; This is the divergence trigger threshold.

[0011] Preferably, in step S5, the mathematical expression of the unidirectional anti-mechanical impact penalty function is: ; in, This is a unidirectional mechanical shock protection penalty function. As a penalty weight, Let i be the current solution load rate of the i-th device. This represents the actual load rate of the i-th device immediately before the load redistribution is triggered. This is the preset maximum allowable increase in load factor for a single operation. This represents the total number of devices operating in parallel.

[0012] Preferably, in step S5, the mathematical expression of the total cost function is: ; in, The total cost function is expressed in kW. This represents the total number of devices operating in parallel. For the first The device health power consumption characteristic function of the device. For the first The virtual dissipation damping term of the equipment, This is a unidirectional mechanical shock protection penalty function. This represents the total number of parallel devices. The mathematical expression for the equality constraint is: ; in, This represents the total number of devices operating in parallel. For the first Rated output of the equipment This represents the current total output demand. The iterative process of solving the total cost function specifically includes: The augmented Lagrange multiplier method is used to iteratively solve the total cost function, constructing an augmented Lagrange function. The load rate is iteratively updated using the gradient descent method, and the load rate is truncated at the boundary in each iteration to ensure that it is within the upper and lower limits of the allowable load rate of the equipment. The optimal partial load rate solution set is output after the iteration termination condition is met.

[0013] Preferably, the iteration termination condition is: the absolute value of the change in all load rates in two adjacent iterations is less than the preset convergence tolerance and the absolute value of the violation of equality constraints is less than the preset allowable error, or the number of iterations reaches the preset maximum number of iterations.

[0014] An energy-saving management system based on power consumption prediction of electromechanical equipment includes a data acquisition module, a feature module, a mapping module, a loss module, and a control module. The data acquisition module is used to collect operating data, partial load rate command values, rated output, and total output demand of the equipment group, and to calculate the instantaneous total power of each equipment. The feature module is used to calculate the theoretical reference power based on the partial load factor command value, calculate the instantaneous power residual by combining the instantaneous total power, segment and extract features from the instantaneous power residual, and obtain a multi-dimensional statistical feature vector. The mapping module is used to perform state mapping on multidimensional statistical feature vectors based on a pre-stored health status feature library, obtain a state label sequence, and construct a state transition probability matrix based on the state label sequence. The loss module is used to calculate the implicit loss divergence based on the state transition probability matrix and the pre-stored health baseline state transition probability matrix. The control module is used to construct a virtual dissipation damping term based on the implicit loss divergence if the implicit loss divergence is greater than the preset divergence trigger threshold. It also constructs a total cost function by combining the unidirectional anti-mechanical shock penalty function and the equipment health power consumption characteristic function. Under the condition of satisfying the equality constraint of the total output demand, the total cost function is iteratively solved to obtain the optimal partial load rate solution set and issue control commands.

[0015] The beneficial effects of this invention are as follows: By calculating the implicit loss divergence between the state transition probability matrix and the health baseline matrix within the current control cycle, this invention quantifies the degree of deviation in the transient power consumption fluctuation transition mode of the analyzed equipment, thereby improving the ability to detect very early and weak anomalies. This invention constructs a total cost function containing multiple constraints and penalty mechanisms, transforming the detected implicit loss divergence into a virtual dissipation damping term. When the equipment exhibits early degradation, this damping term can generate a penalty cost comparable to the power consumption of the equipment in the optimization solution, actively reducing the load rate of abnormal equipment. The unidirectional anti-mechanical shock penalty function allows abnormal equipment to unload freely and quickly while limiting the maximum single load of healthy equipment, thus blocking the motor torque step phenomenon caused by command mutation. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the steps of an energy-saving management method based on power consumption prediction of electromechanical equipment, provided in an embodiment of the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1, referring to Figure 1 This paper provides an energy-saving management method based on the prediction of power consumption of electromechanical equipment, including the following steps: Step S1: Collect the operating data, partial load rate command value, rated output and total output demand of the equipment group, and calculate the instantaneous total power of each equipment. Step S2: Calculate the theoretical reference power based on the partial load factor command value, calculate the instantaneous power residual by combining the instantaneous total power, segment and extract features from the instantaneous power residual, and obtain a multidimensional statistical feature vector. Step S3: Based on the pre-stored health status feature library, perform state mapping on the multidimensional statistical feature vector to obtain the state label sequence, and construct the state transition probability matrix based on the state label sequence; Step S4: Calculate the implicit loss divergence based on the state transition probability matrix and the pre-stored health baseline state transition probability matrix; Step S5: If the implicit loss divergence is greater than the preset divergence trigger threshold, then construct a virtual dissipation damping term based on the implicit loss divergence, and construct a total cost function by combining the unidirectional anti-mechanical shock penalty function and the equipment health power consumption characteristic function. Under the condition of satisfying the equality constraint of the total output demand, iteratively solve the total cost function to obtain the optimal partial load rate solution set and issue control commands.

[0019] This invention transforms the instantaneous power residual of equipment into a state transition probability matrix and calculates the implicit loss divergence, thereby capturing and quantifying the early subtle degradation of electromechanical equipment. Compared with traditional allocation strategies that rely solely on macroscopic power, this invention integrates the implicit loss divergence and the anti-mechanical shock mechanism into the total cost function. Under the premise of meeting the total output demand, it can actively suppress the load of potentially faulty equipment, avoid accelerated damage from operating with defects, and effectively prevent mechanical shock to healthy equipment during load transfer. Thus, it achieves the optimal balance between the operating life of the equipment group and the overall energy consumption at the global level.

[0020] In step S1, the operating data includes voltage and current; The calculation process for obtaining the instantaneous total power is as follows: divide the sum of the three-phase products of voltage and current by the preset conversion factor; Rated output capacity represents the maximum effective output that the equipment can output per unit time under full load operation, while total output demand represents the total effective output that needs to be provided by the group of equipment.

[0021] In one specific embodiment of the present invention, during the entire control cycle, the instantaneous values ​​of each phase voltage and current of the i-th device are continuously collected at a preset sampling frequency of 1000Hz. At each sampling time t, according to the calculation principle of three-phase AC power, the sum of the products of the instantaneous values ​​of the three-phase voltage and current at the same time is divided by 1000 to calculate the instantaneous total power at that time. The unit is kW; To capture only the subtle power consumption fluctuations caused by the device's own state and eliminate the impact of changes in macroscopic load levels, this embodiment collects the first... The partial load rate command value calculated by the device in the previous control cycle and sent to the actuator. In this embodiment, the preset control cycle is 60 seconds. All operation steps of this method are executed within one control cycle. If it is the first run and there is no historical record, the current actual operating frequency is read back from the inverter driver of the i-th device via the communication interface. Based on the linear mapping relationship between the frequency and load rate of that device, the read-back frequency is converted into a load rate, which is then used as... The initial value; Let the total number of devices operating in parallel be K. For the th... The equipment has a partial load rate of [number]. This represents the proportion of the device's current output to its rated output. Collection of the first Rated output of the equipment The specific type of rated output depends on the equipment type. Rated output represents the maximum effective output that the equipment can output per unit time under full load operation. If the equipment is a refrigeration compressor, the rated output is the rated cooling capacity, in kW. If the equipment is a water pump, the rated output is the rated flow rate, in m³ / h. If the equipment is a fan, the rated output is the rated air volume, in m³ / h. The rated output is obtained from the parameters on the equipment nameplate. Collect total output demand Total output demand represents the total effective output that needs to be provided by the group of equipment, and the rated output. Total output demand The units are consistent.

[0022] This invention calculates the instantaneous total power by multiplying the instantaneous values ​​of three-phase voltage and current, and clarifies the physical boundary between the rated output and the total output demand, thus avoiding the loss of high-frequency characteristics and time delay caused by intermediate instruments or filtering algorithms.

[0023] Step S2, the process of constructing a multidimensional statistical feature vector, specifically includes: Call the pre-stored device health power consumption characteristic function. The device health power consumption characteristic function is a mapping function obtained by performing multinomial regression fitting on the factory test data of this model of equipment, with the partial load rate as the independent variable and the theoretical input power as the function value. Substitute the partial load rate command value into the device health power consumption characteristic function to obtain the theoretical input power under the partial load rate command value as the theoretical reference power. The difference between the instantaneous total power and the theoretical reference power is taken as the instantaneous power residual of the equipment, and the instantaneous power residuals obtained throughout the entire control cycle are used to form an instantaneous power residual sequence. The instantaneous power residual sequence is evenly divided into multiple segments according to a preset duration. The root mean square value, standard deviation, and kurtosis of all instantaneous power residuals in each segment are calculated to construct a multidimensional statistical feature vector.

[0024] In a specific embodiment of the present invention, after obtaining the partial load rate instruction value, a pre-stored health power consumption characteristic function of the device model is called. The device health power consumption characteristic function is a mapping function, the independent variable is the partial load rate, and the function value is the theoretical input power of the device model when operating at that load rate in its factory state, in kW. The specific form and coefficients of the device health power consumption characteristic function are obtained by performing polynomial regression fitting on the factory test data of the device model. In this example, the fitting is a quadratic polynomial. ,in, For the theoretical input electrical power, , , For the known constants obtained through fitting, In this embodiment, the load factor is... Substituting the partial load factor command value into the fitted quadratic polynomial, the theoretical input power under the partial load factor command value is obtained as the theoretical reference power. ; The difference between the instantaneous total power and the theoretical reference power is taken as the instantaneous power residual of the equipment. The instantaneous power residual reflects the slight deviation in power consumption caused by the equipment's own condition, such as internal deterioration, changes in mechanical friction, or partial discharge. It is no longer affected by the current load rate. The instantaneous power residual obtained throughout the entire control cycle constitutes the instantaneous power residual sequence. The instantaneous power residual sequence is evenly divided into multiple segments according to a preset duration. In this embodiment, the sampling frequency is 1000Hz and the control period duration is 60s. In this embodiment, 1s is selected to divide the instantaneous power residual sequence into 60 segments. Calculate the root mean square, standard deviation, and kurtosis of all instantaneous power residuals within each segment, and construct a multidimensional statistical eigenvector. The multidimensional statistical feature vectors corresponding to each segment are: ,in, The root mean square value, Standard deviation To achieve peak performance, through the above processing, the equipment within one control cycle... The residual sequence was transformed into 60 multidimensional statistical feature vectors arranged in chronological order; Multidimensional statistical feature vectors characterize the energy, fluctuation intensity, and impact distribution of instantaneous power residuals from different perspectives, and can comprehensively preserve the transient information of the equipment.

[0025] This invention utilizes a pre-stored device health power consumption characteristic function to obtain a theoretical reference power, strips away the macroscopic power consumption affected by external control commands, and extracts the instantaneous power residual reflecting the internal physical state of the device. By dividing the residual sequence into short-time segments and extracting the root mean square value, standard deviation, and kurtosis features, this invention can comprehensively characterize the microscopic abnormal behavior of the device from multiple dimensions such as energy magnitude, fluctuation intensity, and peak impact distribution, thereby improving the richness of feature information and the identification of abnormal states.

[0026] Step S3, the process of obtaining the state label sequence, specifically includes: The process of retrieving the health status center vector of a device, which contains multiple health status center vectors for the same model of device, involves: The equipment is operated within the allowable load rate range, with the load rate gradually increasing from the lower limit to the upper limit in preset steps, and running stably for a preset time at each load rate. During the full-condition scan, the root mean square value, standard deviation, and kurtosis are calculated in real time to generate multiple multidimensional statistical feature vectors. A clustering algorithm with a preset number of clusters is executed on all generated feature vectors, and the centroid vector of each cluster is output as the health status center vector, and a status label is assigned to each health status center vector. For each segment feature vector calculated within the current control cycle, calculate its Euclidean distance with each health state center vector, select the health state center vector with the smallest distance value as the state label of the segment, and convert multiple multidimensional statistical feature vectors within a control cycle into a state label sequence. The process of constructing the state transition probability matrix specifically includes: Based on the state label sequence, the cumulative frequency of state labels at adjacent positions transitioning from the previous state to the next state is counted, the state transition probability is calculated, and a state transition probability matrix is ​​constructed.

[0027] In a specific embodiment of the present invention, a health status feature library for this model of device is pre-stored. The health status feature library contains multiple health status center vectors for the same model of device. The process of obtaining the health status center vector specifically includes: Under the factory conditions of the same model of equipment, the allowable adjustment range of its load rate is determined based on the safe operating boundary of the frequency converter of that model of equipment. The allowable adjustment range of the load rate is the scanning range of the equipment under all operating conditions. The equipment is to be operated within its allowable load rate range. During operation, the load rate is to be gradually increased from the lower limit to the upper limit in a preset step of 0.05 through its control system, and the equipment is to be operated stably for a preset time of 5 minutes at each load rate to ensure that sufficient data under the steady-state condition is collected. In addition, the data on the dynamic transition process of the load rate from one value to another should also be included. During the full-condition scanning process, the same sampling steps as in this embodiment are used, namely, sampling the three-phase voltage and current at a frequency of 1000Hz, and continuously calculating the root mean square value, standard deviation and kurtosis in real time with a window duration of 1 second to generate multiple multidimensional statistical feature vectors. Collect all the feature vectors generated above, and execute the K-means clustering algorithm with a preset number of clusters Z. After the algorithm converges, output the centroid vectors of Z clusters, which are the health status center vectors. The health status center vectors constitute the health status feature library of this embodiment. Assign a status label to the health status center vectors in the health status feature library. In this embodiment, the value of Z is determined by the elbow method, and the value of Z in this embodiment is 3. The number calculated within the current control cycle Feature vectors of each segment The Euclidean distance between each health state center vector and the health state center vector is calculated. The state label corresponding to the health state center vector with the smallest distance value is selected as the state label for that segment. Through this mapping, multiple multidimensional statistical feature vectors within a control cycle are converted into a state label sequence of length 60. In this embodiment, the state label is set as follows: ,in, ; Based on the state label sequence, the state transitions of adjacent positions are statistically analyzed. Specifically, among all adjacent state label pairs, the transitions from state to position are statistically analyzed. Transition to state The cumulative frequency is denoted as ,in, ; Construct a state transition probability matrix based on the state label sequence The element in the m-th row and n-th column of the state transition probability matrix The mathematical expression is: ; in, State transition probability matrix The element in the m-th row and n-th column represents the state of the device during the current control cycle. Transition to state The estimated probability is in the range of (0,1), and the sum of all elements in each row of the matrix is ​​always equal to 1. and State transition frequency represents the frequency at which adjacent positions in a state symbol sequence transition from one state to the next. Transfer to , The cumulative number of occurrences is a non-negative integer, and Z is the number of health state center vectors; Row numbers of the state transition probability matrix Represents the state type before the transition, column number The state transition probability matrix represents the probability of a device state transition within the current control cycle, indicating the transient power consumption fluctuation transition mode of the device at this time.

[0028] This invention pre-constructs a health state center vector through full-condition scanning and clustering algorithms, maps continuous high-dimensional feature sequences into discrete state label sequences, and constructs a probability matrix based on the frequency of state transitions. It is no longer limited to the static performance of the equipment at a certain isolated moment, but deeply explores the dynamic evolution law and transition mode of the transient power consumption fluctuation of the equipment, and enhances the feature extraction capability of non-stationary abnormal noise under complex operating conditions.

[0029] Step S4, the process of obtaining the implicit loss divergence, specifically includes: The device's health baseline state transition probability matrix is ​​invoked. The health baseline state transition probability matrix is ​​constructed based on the operating data of the same model of device in its factory condition. Minimize all elements of the health baseline state transition probability matrix and renormalize it to obtain a smoothed baseline matrix. The KL divergence between the state transition probability matrix and the reference matrix of the current control cycle is calculated by summing each element, and is used as the implicit loss divergence.

[0030] In one specific embodiment of the present invention, a health baseline state transition probability matrix of a uniform model device is pre-stored. Its acquisition method and The data is the same, but the data source is the operating data of the same model of equipment in its factory condition. The operating data includes the instantaneous voltage and current values ​​of each phase collected during the offline calibration process of the health equipment at the same sampling frequency of 1000Hz as the online operation, as well as the partial load rate command of the equipment under various operating conditions. The above voltage and current data are used to calculate the instantaneous power and then extract the power residual and feature vector. The partial load rate command value is used to calculate the theoretical reference power. After the same feature extraction, state symbol mapping, transition frequency statistics and Laplace smoothing as the online operation, the following is constructed: ; To further ensure numerical stability in floating-point calculations, and Apply a minimum value to all elements Then, renormalization is performed to obtain the smoothed reference matrix. and the smoothed state transition matrix In this embodiment The value is ; For the The device calculates the smoothed state transition matrix element by element-wise summation. With reference matrix The KL divergence between them serves as the implicit loss divergence. The implicit loss divergence is calculated by summing the elements of the two matrices after flattening them into vectors. When the equipment is in a healthy state, the fluctuation transfer pattern is consistent with the baseline, and the implicit loss divergence is small. If the equipment deteriorates early, the transfer pattern shifts, and the divergence increases significantly. The implicit loss divergence reflects the current degree of implicit loss of the equipment.

[0031] This invention solves the problem of divergence calculation or numerical instability caused by zero-probability transitions in actual operating data by using minimum value smoothing and renormalization techniques. By calculating the KL divergence between the smoothed state transition matrix and the health benchmark matrix, the degree to which the current equipment fluctuation transition mode deviates from the health benchmark can be quantified.

[0032] In step S5, the implicit loss divergence is transformed into the resistance to the increased load on abnormal equipment in the optimization objective, and a virtual dissipation damping term is constructed. The mathematical expression of the virtual dissipation damping term is as follows: ; in, Let i be the virtual dissipation damping term for the i-th device. This is the damping weighting coefficient. Let be the implicit loss divergence of the i-th device. Let i be the current partial load factor of the i-th device. The preset low-load safety zone boundary is set to 0.45 in this embodiment; This is the divergence trigger threshold.

[0033] In one specific embodiment of the present invention, The value of γ is based on the fact that when the implicit loss divergence of a device reaches the trigger threshold of 0.05, the virtual dissipation damping term of the damping term should generate a penalty cost equivalent to the power consumption of the device itself, thereby forming an effective resistance in the optimization solution and forcing the algorithm to reduce the load rate of the device. In this embodiment, for a compressor with a rated power of 200kW, γ is taken as 150kW. This represents the safety zone boundary in the virtual dissipation damping term, indicating the minimum safe operating load that allows equipment with implicit losses to maintain without triggering damping penalties. The value is determined based on the equipment operation and maintenance manual, obtaining the ratio of the equipment's safe operating lower limit to the rated load. In this embodiment, it is 40%, with an operating margin. Therefore, The value is 0.45; The virtual dissipation damping term has no effect when the equipment is in a healthy state. When the equipment has hidden losses, it forces its load rate to be suppressed to a safe low level. When the hidden loss divergence is less than or equal to the divergence trigger threshold, the virtual dissipation damping term is forced to be 0. Therefore, this damping term only applies to equipment that has detected hidden losses. The preset divergence trigger threshold is determined through controlled degradation experiments. The controlled degradation experiments simulate equipment degradation using the power perturbation injection method, and the specific process includes: Select a factory-condition health device of the same model and run it within the allowable load rate range. Following the same sampling and processing method as in this embodiment, collect the instantaneous voltage and current values ​​and partial load rate command values ​​of the device under four typical operating conditions: low load steady state, medium load steady state, high load steady state, and load dynamic change. Collect data continuously for 30 minutes under each operating condition at a sampling frequency of 1000Hz. Based on the method in step S2, calculate the instantaneous power residual sequence under each working condition, and use it as the health baseline residual sequence for the corresponding working condition; On the health baseline residual sequences of the four operating conditions, perturbation signals of different intensities are superimposed by software to generate simulated residual sequences corresponding to the four degradation levels. Four disturbance levels are defined, including: grade For a healthy state, use a healthy baseline residual sequence; grade For very early anomalies, a continuous low-intensity Gaussian white noise perturbation signal is superimposed on the healthy baseline residual sequence. The root mean square value of the white noise is 1.5% of the rated power of the device, which is used to simulate the continuous weak random fluctuations in power consumption caused by very early degradation of the device. grade For moderate anomalies, based on level G1, a rectangular pulse disturbance with a width of 0.5 seconds and an amplitude of 5% of the rated power of the equipment is superimposed at random time intervals of 5 to 15 seconds to simulate intermittent short-term spike impacts. grade For more severe anomalies, based on level G2, the amplitude of the rectangular pulse disturbance is increased to 10% of the equipment's rated power, and the pulse interval is shortened to a random time interval of 2 to 5 seconds to simulate a state where the anomaly is further aggravated and the spike impacts are more frequent and severe. The above-mentioned superimposed disturbance signals are all directly applied to the instantaneous power residual value, in kW, simulating the additional power consumption fluctuations caused by internal equipment degradation. Under each disturbance level, the equipment is run for 30 minutes each under low load, medium load, high load and dynamic load change conditions. The detection rate is the percentage of time during which the output latent loss divergence is greater than the divergence trigger threshold in all operating time periods of the specified disturbance level. The low-load steady-state operating condition is as follows: the partial load rate command of the equipment is fixed at 0.45, which is the lower limit of the load level within the safe operating range of the equipment, simulating the equipment in a low-output operating state; The medium-load steady-state operating condition is as follows: the partial load rate command of the equipment is fixed at 0.60, which is the intermediate load level of the safe operating range of the equipment, simulating the equipment in normal output operation. The high-load steady-state operating condition is as follows: the partial load rate command of the equipment is fixed at 0.85, which is the upper limit of the load level within the safe operating range of the equipment, simulating the equipment in a high-output operating state; The load dynamic change condition is as follows: the partial load rate command of the equipment is continuously changed within the safe operating range with a preset period and amplitude, simulating the operating state of the equipment frequently adjusting its output in actual production. Specifically, the partial load rate command changes continuously between 0.45 and 0.85 with a sinusoidal law with a period of 120 seconds. The false alarm rate under normal operating conditions is at level [missing information]. The percentage of time during which the hidden loss divergence exceeds the divergence trigger threshold; The response delay is from the power disturbance generator to the disturbance level. The time interval from the moment of switching to the target level to the moment when the latent loss divergence first exceeds the divergence trigger threshold and remains thereafter; The candidate thresholds of 0.01, 0.03, 0.05, 0.08, and 0.10 were tested, and the results are shown in Table 1. Table 1

[0034]

[0035] When the latent loss divergence is 0.01, the false alarm rate is as high as 24.5%, resulting in frequent false triggers, making it unsuitable. When the latent loss divergence is 0.03, the false alarm rate of 8.3% is still too high. When the latent loss divergence is 0.05, the detection rate is 98.6%, the false alarm rate is only 1.2%, and the response delay of 28 seconds meets the protection pre-set time requirement, achieving the optimal balance between detection sensitivity and anti-interference capability. When the latent loss divergence is 0.08 or 0.10, the missed detection rate increases significantly, and the safety margin is insufficient. Considering all factors, this embodiment uniquely determines the divergence trigger threshold as 0.05.

[0036] This invention transforms the implicit loss divergence of equipment into physical resistance in the optimization algorithm, and triggers it only when the divergence exceeds a specific threshold. This trigger threshold is calibrated in multiple stages based on controlled degradation experiments and power disturbance injection methods, achieving a balance between detection sensitivity and resistance to environmental noise interference. This ensures that the algorithm only penalizes equipment with actual implicit losses, forcing its load rate to be suppressed to a safe low range, thus avoiding frequent misadjustments caused by normal fluctuations.

[0037] In step S5, the mathematical expression for the unidirectional anti-mechanical impact penalty function is: ; in, This is a unidirectional mechanical shock protection penalty function. As the penalty weight, the value in this embodiment is taken as [value]. kW, Let i be the current solution load rate of the i-th device. This represents the actual load rate of the i-th device at the instant before the current load redistribution is triggered. This value is equivalent to the output command value of the device in the previous control cycle. , The preset maximum allowable increase in load rate for a single instance is 0.08 in this embodiment. This represents the total number of devices operating in parallel.

[0038] The value of this constraint is based on the fact that the anti-impact constraint is a hard constraint that cannot be violated in optimization problems. Once the increase in load factor exceeds the allowable value, the constraint will be enforced. The penalty cost must be much greater than the sum of all other terms in the total cost function, thus mathematically forcing the optimization algorithm to abandon any solution that violates this constraint. In this embodiment, the total rated power of the equipment group is 2000kW, so λ takes the value of kW; The value is determined based on the following: a sudden change in load rate command will cause a corresponding sudden change in motor torque, generating impact stress on mechanical components. Based on the mechanical design specifications provided by the equipment manufacturer and engineering operation experience, the load rate change corresponding to the maximum allowable single torque step change is obtained as the load rate change. This means that the increase in the load rate of healthy equipment must not exceed 0.08 within a control cycle to ensure the safe operation of the equipment.

[0039] The unidirectional mechanical shock protection penalty function only applies to increases in load factor exceeding [a certain threshold]. Severe penalties are imposed on behaviors that cause a drop in load, while no penalties are imposed on unloading behaviors that cause a drop in load rate. This protects healthy devices from load shocks while allowing abnormal devices to unload freely and quickly.

[0040] The unidirectional anti-mechanical shock penalty function of this invention distinguishes between the loading and unloading behavior of the equipment. This penalty function only imposes a huge penalty on commands that cause the instantaneous increase in load rate to exceed the safety threshold, while not restricting the rapid unloading behavior of load reduction. This gives abnormal equipment the freedom to quickly reduce load and avoid danger, and avoids the risk of healthy equipment suffering from a step shock of motor torque due to receiving excessively high sudden change commands when the load is redistributed, thus ensuring the safety of mechanical transmission components.

[0041] In step S5, the mathematical expression for the total cost function is: ; in, The total cost function is expressed in kW. This represents the total number of devices operating in parallel. For the first The device health power consumption characteristic function of the device. For the first The virtual dissipation damping term of the equipment, This is a unidirectional mechanical shock protection penalty function. This represents the total number of parallel devices. The total cost function integrates energy consumption cost, virtual damping penalty based on implicit losses, and unidirectional loading impact limit cost; The total cost function must satisfy the equality constraint, which means that the sum of the effective output of all equipment at their respective load rates is equal to the current total output demand. The mathematical expression for the equality constraint is: ; in, This represents the total number of devices operating in parallel. For the first Rated output of the equipment This represents the current total output demand. The iterative process of solving the total cost function specifically includes: The augmented Lagrange multiplier method is used to iteratively solve the total cost function, constructing an augmented Lagrange function. The load rate is iteratively updated using the gradient descent method, and the load rate is truncated at the boundary in each iteration to ensure that it is within the upper and lower limits of the allowable load rate of the equipment. The optimal partial load rate solution set is output after the iteration termination condition is met.

[0042] In a specific embodiment of the present invention, the objective of this method is to find the partial load rate of each device that minimizes the total cost function while satisfying the total output demand of the device group. Based on the equality constraints, the iterative solution of the total cost function using the augmented Lagrange multiplier method includes the following steps: Constructing the augmented Lagrangian function The mathematical expression for the augmented Lagrange function is: ; in, To augment the Lagrange function, the unit is kW. Let be the total cost function. For Lagrange multipliers, This represents the total number of parallel devices. Let i be the rated output of the i-th equipment. This represents the current total output demand. The secondary penalty coefficient is a preset positive constant used to adjust the degree to which the equality constraints are satisfied. In this embodiment... It is 100. The selection of this parameter results in a penalty of approximately 50kW when the constraint violation is 1kW, which matches the power consumption of the device. The load rate is updated iteratively using the gradient descent method. To ensure that the load rate command generated in each iteration is within the physical safety limits of the equipment, the upper and lower limits of the allowable load rate of the equipment are collected. The mathematical expression for the update and boundary truncation in the t-th iteration is: ; ; in, For the first The updated load rate value of the equipment after the (t+1)th iteration. The lower limit of the allowable load rate for the equipment is 0.4 in this embodiment. The maximum allowable load rate for the equipment is 1.0 in this embodiment. For the first The load rate value of the device at the t-th iteration. The iteration step size for gradient descent is 0.005 in this embodiment. The upper and lower limits of the allowable load rate of the equipment can be obtained by referring to the technical manual or operation and maintenance procedures of this model of equipment. This represents the updated value of the Lagrange multipliers after the (t+1)th iteration. The updated values ​​of the Lagrange multipliers after t iterations. This is the secondary penalty coefficient; This represents the total number of parallel devices. Let i be the rated output of the i-th equipment; This represents the current total output demand. To augment the Lagrange function; To augment the Lagrange function For the Equipment load rate The partial derivatives, the mathematical expression for the partial derivatives are: ; in, It is derived from the derivative of the power consumption function. ;when hour, .

[0043] This invention unifies the physical energy consumption characteristics of the equipment, the virtual damping penalty for degradation, and the shock protection penalty within the same framework. It uses the augmented Lagrange multiplier method to handle the constraint of equal total output, and combines gradient descent and boundary truncation mechanisms for iterative optimization. This ensures that the load rate command generated in each search round is strictly limited to the physical safety range of the equipment. This not only guarantees the efficient solution of complex nonlinear optimization problems, but also ensures the absolute safety and feasibility of the final dynamic allocation strategy at the engineering execution level.

[0044] The iteration termination condition is: the absolute value of the change in all load rates in two adjacent iterations is less than the preset convergence tolerance and the absolute value of the violation of equality constraints is less than the preset allowable error, or the number of iterations reaches the preset maximum number of iterations.

[0045] In one specific embodiment of the present invention, the initial point of iteration is taken as the load rate before triggering, and the multiplier is... Initially set to 0, the iteration termination condition is set as: all in two consecutive iterations The absolute values ​​of the changes are all less than the preset convergence tolerance. And the absolute value of the equality constraint violation If the error is less than the preset allowable error of 0.1, or the number of iterations reaches the preset maximum number of iterations of 500, the process is considered converged and the optimal partial load factor solution set is output.

[0046] The obtained dimensionless optimal partial load factor solution set is transformed into control commands that can be recognized by the actuators of each device. The control commands of each device are sent to the corresponding actuators in parallel to complete the smooth transfer of load.

[0047] An energy-saving management system based on power consumption prediction of electromechanical equipment includes a data acquisition module, a feature module, a mapping module, a loss module, and a control module. The data acquisition module is used to collect operating data, partial load rate command values, rated output, and total output demand of the equipment group, and to calculate the instantaneous total power of each equipment. The feature module is used to calculate the theoretical reference power based on the partial load factor command value, calculate the instantaneous power residual by combining the instantaneous total power, segment and extract features from the instantaneous power residual, and obtain a multi-dimensional statistical feature vector. The mapping module is used to perform state mapping on multidimensional statistical feature vectors based on a pre-stored health status feature library, obtain a state label sequence, and construct a state transition probability matrix based on the state label sequence. The loss module is used to calculate the implicit loss divergence based on the state transition probability matrix and the pre-stored health baseline state transition probability matrix. The control module is used to construct a virtual dissipation damping term based on the implicit loss divergence if the implicit loss divergence is greater than the preset divergence trigger threshold. It also constructs a total cost function by combining the unidirectional anti-mechanical shock penalty function and the equipment health power consumption characteristic function. Under the condition of satisfying the equality constraint of the total output demand, the total cost function is iteratively solved to obtain the optimal partial load rate solution set and issue control commands.

[0048] This invention calculates the theoretical baseline power through the device health power consumption characteristic function, extracts the instantaneous power residual, effectively removes the interference of macro load level changes on power consumption, and captures the subtle power consumption fluctuations caused by internal device degradation. This invention segments the residual sequence by time segment and extracts multi-dimensional statistical features such as root mean square value, standard deviation and kurtosis, and retains multi-dimensional feature information such as transient energy, fluctuation intensity and impact distribution of the device. To further accurately assess the health status of equipment, this invention introduces a state label mapping and a state transition probability matrix. By calculating the implicit loss divergence between the state transition probability matrix and the health baseline matrix within the current control cycle, this invention can quantitatively analyze the degree of deviation of the transient power consumption fluctuation transition mode of the equipment. This divergence assessment method based on dynamic transition mode not only improves the detection capability of very early weak anomalies, but also effectively filters background random interference under normal operating conditions when combined with the set trigger threshold, achieving a balance between anomaly detection rate and system anti-false alarm capability. This invention constructs a total cost function that includes multiple constraints and penalty mechanisms. It transforms the detected implicit loss divergence into a virtual dissipation damping term. When equipment exhibits early degradation, this damping term generates a penalty cost comparable to the power consumption of the equipment during optimization, actively reducing the load rate of abnormal equipment and limiting it to a safe low load level. This invention also sets a unidirectional anti-mechanical shock penalty function, which allows abnormal equipment to unload freely and quickly while limiting the maximum single load of healthy equipment. This prevents the motor torque step phenomenon caused by sudden command changes, effectively protecting mechanical components from impact stress damage. Thus, while accurately meeting the total output requirements, it minimizes the overall energy consumption of the electromechanical equipment group and maximizes its service life.

[0049] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0050] 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, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. Energy saving management method based on electromechanical equipment power consumption prediction, characterized in that, Includes the following steps: Step S1: Collect the operating data, partial load rate command value, rated output and total output demand of the equipment group, and calculate the instantaneous total power of each equipment. Step S2: Calculate the theoretical reference power based on the partial load rate command value, calculate the instantaneous power residual in combination with the instantaneous total power, segment and extract features from the instantaneous power residual, and obtain a multidimensional statistical feature vector. Step S3: Based on the pre-stored health status feature library, perform state mapping on the multidimensional statistical feature vector to obtain a state label sequence, and construct a state transition probability matrix based on the state label sequence; Step S4: Calculate the implicit loss divergence based on the state transition probability matrix and the pre-stored health baseline state transition probability matrix; Step S5: If the implicit loss divergence is greater than the preset divergence trigger threshold, then a virtual dissipation damping term is constructed based on the implicit loss divergence, and a total cost function is constructed by combining the unidirectional anti-mechanical shock penalty function and the equipment health power consumption characteristic function. Under the condition of satisfying the equality constraint of the total output demand, the total cost function is iteratively solved to obtain the optimal partial load rate solution set and control commands are issued.

2. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 1, characterized in that, In step S1, the operating data includes voltage and current; The calculation process for obtaining the instantaneous total power is as follows: the sum of the three-phase products of voltage and current is divided by a preset conversion factor; The rated output represents the maximum effective output that the equipment can output per unit time under full load operation, and the total output demand represents the total effective output that needs to be provided by the equipment group.

3. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 2, characterized in that, The process of constructing the multidimensional statistical feature vector in step S2 specifically includes: Call the pre-stored device health power consumption characteristic function, which is a mapping function obtained by performing multinomial regression fitting on the factory test data of the device, with the partial load rate as the independent variable and the theoretical input power as the function value. Substitute the partial load rate command value into the device health power consumption characteristic function to obtain the theoretical input power under the partial load rate command value as the theoretical reference power. The difference between the instantaneous total power and the theoretical reference power is taken as the instantaneous power residual of the device, and the instantaneous power residuals obtained throughout the entire control cycle are used to form an instantaneous power residual sequence. The instantaneous power residual sequence is evenly divided into multiple segments according to a preset duration. The root mean square value, standard deviation, and kurtosis of all instantaneous power residuals in each segment are calculated to construct a multidimensional statistical feature vector.

4. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 3, characterized in that, The process of obtaining the state label sequence in step S3 specifically includes: The process of retrieving the health status center vectors of devices of the same model includes: accessing the device's health status feature library, which contains multiple health status center vectors for devices of the same model. The equipment is operated within the allowable load rate range, with the load rate gradually increasing from the lower limit to the upper limit in preset steps, and running stably for a preset time at each load rate. During the full-condition scan, the root mean square value, standard deviation, and kurtosis are calculated in real time to generate multiple multidimensional statistical feature vectors. A clustering algorithm with a preset number of clusters is executed on all generated feature vectors, and the centroid vector of each cluster is output as the health status center vector, and a status label is assigned to each health status center vector. For each segment feature vector calculated within the current control cycle, calculate its Euclidean distance with each health state center vector, select the health state center vector with the smallest distance value as the state label of the segment, and convert multiple multidimensional statistical feature vectors within a control cycle into a state label sequence. The process of constructing the state transition probability matrix specifically includes: Based on the state label sequence, the cumulative frequency of state labels at adjacent positions transitioning from the previous state to the next state is counted, the state transition probability is calculated, and a state transition probability matrix is ​​constructed.

5. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 4, characterized in that, In step S4, the process of obtaining the implicit loss divergence specifically includes: The device's health baseline state transition probability matrix is ​​invoked, which is constructed based on the operating data of the same model of device in its factory condition; Minimize all elements of the health baseline state transition probability matrix and renormalize it to obtain a smoothed baseline matrix. The KL divergence between the state transition probability matrix of the current control cycle and the reference matrix is ​​calculated by summing elements one by one, and is used as the implicit loss divergence.

6. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 5, characterized in that, In step S5, the mathematical expression for the virtual dissipation damping term is: ； in, Let i be the virtual dissipation damping term for the i-th device. This is the damping weighting coefficient. Let be the implicit loss divergence of the i-th device. Let i be the current partial load factor of the i-th device. This is the preset low-load safety zone boundary; This is the divergence trigger threshold.

7. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 6, characterized in that, In step S5, the mathematical expression of the unidirectional anti-mechanical impact penalty function is: ； in, This is a unidirectional mechanical shock protection penalty function. As a penalty weight, Let i be the current solution load rate of the i-th device. This represents the actual load rate of the i-th device immediately before the load redistribution is triggered. This is the preset maximum allowable increase in load factor for a single operation. This represents the total number of devices operating in parallel.

8. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 7, characterized in that, In step S5, the mathematical expression for the total cost function is: ； in, The total cost function is expressed in kW. This represents the total number of devices operating in parallel. For the first The device health power consumption characteristic function of the device. For the first The virtual dissipation damping term of the equipment, This is a unidirectional mechanical shock protection penalty function. This represents the total number of parallel devices. The mathematical expression for the equality constraint is: ； in, This represents the total number of devices operating in parallel. For the first Rated output of the equipment This represents the current total output demand. The process of iteratively solving the total cost function specifically includes: The total cost function is solved iteratively using the augmented Lagrange multiplier method to construct the augmented Lagrange function. The load rate is updated iteratively using the gradient descent method, and the load rate is truncated at the boundary in each iteration to ensure that it is within the upper and lower limits of the allowable load rate of the equipment. The optimal partial load rate solution set is output after the iteration termination condition is met.

9. The energy-saving management method based on power consumption prediction of electromechanical equipment as described in claim 8, characterized in that, The iteration termination condition is: the absolute value of the change in all load rates in two adjacent iterations is less than the preset convergence tolerance and the absolute value of the violation of equality constraints is less than the preset allowable error, or the number of iterations reaches the preset maximum number of iterations.

10. An energy-saving management system based on electromechanical equipment power consumption prediction, applied in the energy-saving management method based on electromechanical equipment power consumption prediction as described in any one of claims 1-9, characterized in that, It includes an acquisition module, a feature module, a mapping module, a loss module, and a control module; The acquisition module is used to collect the operating data, partial load rate command value, rated output and total output demand of the equipment group, and calculate the instantaneous total power of each equipment. The feature module is used to calculate the theoretical reference power based on the partial load factor instruction value, calculate the instantaneous power residual in combination with the instantaneous total power, segment and extract features from the instantaneous power residual, and obtain a multidimensional statistical feature vector. The mapping module is used to perform state mapping on the multidimensional statistical feature vector based on a pre-stored health status feature library, obtain a state label sequence, and construct a state transition probability matrix based on the state label sequence. The loss module is used to calculate the implicit loss divergence based on the state transition probability matrix and the pre-stored health baseline state transition probability matrix. The control module is configured to construct a virtual dissipation damping term based on the implicit loss divergence if the implicit loss divergence is greater than a preset divergence trigger threshold, and construct a total cost function by combining the unidirectional anti-mechanical shock penalty function and the equipment health power consumption characteristic function. Under the condition of satisfying the equality constraint of the total output demand, the total cost function is iteratively solved to obtain the optimal partial load rate solution set and issue control commands.

Images