Load balanced distribution control method and system in parallel operation of multiple energy hosts

Abstract

The invention discloses a load balanced distribution control method and system in parallel operation of multiple energy hosts, and relates to the technical field of load distribution, and the method comprises the steps: collecting a real-time operation parameter set of each energy host in a parallel system; calculating a real-time health degree coefficient of each energy host; solving the energy efficiency deviation degree of each energy host under the current load demand, and constructing a load redistribution optimization function; solving the load redistribution optimization function, and dynamically generating and issuing a personalized load setting instruction to each energy host; and monitoring the overall energy efficiency improvement rate of the parallel system and the operation stability index of each energy host, and dynamically backtracking and adjusting the parameter weight of the host health degree attenuation model according to the overall energy efficiency improvement rate and the operation stability index. The technical problem that an existing load distribution method ignores the age limit and the maintenance state of a host, and consequently the overall performance is degraded is solved.

Description

Technical Field

[0001] This application relates to the field of load distribution technology, specifically to a load balancing distribution control method and system for multiple energy generators operating in parallel. Background Technology

[0002] With the deepening of energy structure transformation, the demand for comprehensive utilization of multiple energy forms in the industrial sector is growing, and the system architecture of parallel operation of multiple types and specifications of energy main units is gradually becoming the mainstream mode of energy supply.

[0003] In existing technologies, traditional load distribution methods mostly adopt static strategies based on proportional or average distribution of rated capacity, which fail to fully consider the dynamic changes in the actual energy efficiency characteristics of each energy host due to differences in operating years and maintenance conditions, resulting in higher total energy consumption of the entire parallel system.

[0004] Meanwhile, existing control methods lack a dynamic assessment mechanism for the health status of the host, making it difficult to adaptively adjust the load distribution scheme based on the real-time performance degradation and maintenance recovery of the host. This results in the system operating in a non-optimal energy efficiency range for a long time, and may exacerbate the wear rate of some hosts, affecting the overall operational stability and service life. Summary of the Invention

[0005] This application provides a load balancing distribution control method and system for multiple energy hosts operating in parallel, which solves the technical problem that existing load distribution methods ignore the age and maintenance status of the hosts, leading to overall performance degradation.

[0006] The technical solution to the above-mentioned technical problems in this application is as follows: Firstly, this application provides a load balancing distribution control method for multiple energy mainframes operating in parallel, the method comprising: Collect real-time operating parameter sets of each energy host in the parallel system, wherein the real-time operating parameter sets include at least the current output load, energy consumption rate and historical maintenance flags; Based on the real-time operating parameter set, the real-time health coefficient of each energy host is dynamically calculated through the host health decay model, wherein the host health decay model is coupled with the operating years accumulation factor and the maintenance effect recovery factor. Based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, the energy efficiency deviation of each energy host under the current load demand is calculated, and a load redistribution optimization function with the goal of minimizing the total energy consumption of the parallel system is constructed. The load redistribution optimization function is solved using the particle swarm optimization algorithm, and personalized load setting instructions are dynamically generated and sent to each energy host. After executing the personalized load setting command, the overall energy efficiency improvement rate of the parallel system and the operational stability index of each energy host are monitored, and the parameter weights of the host health decay model are dynamically adjusted back based on the overall energy efficiency improvement rate and the operational stability index.

[0007] Secondly, this application provides a load balancing distribution control system for multiple energy mainframes operating in parallel, including: The information acquisition module is used to collect the real-time operating parameter set of each energy host in the parallel system, wherein the real-time operating parameter set includes at least the current output load, energy consumption rate and historical maintenance markers; The data processing module is used to dynamically calculate the real-time health coefficient of each energy host based on the real-time operating parameter set and through the host health decay model, wherein the host health decay model is coupled with the operating years accumulation factor and the maintenance effect recovery factor. The energy efficiency optimization module is used to solve the energy efficiency deviation of each energy host under the current load demand based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, and to construct a load redistribution optimization function with the goal of minimizing the total energy consumption of the parallel system. The instruction generation module is used to solve the load redistribution optimization function using the particle swarm optimization algorithm, and dynamically generate and send personalized load setting instructions to each energy host. The instruction execution module is used to monitor the overall energy efficiency improvement rate of the parallel system and the operational stability indicators of each energy host after executing the personalized load setting instruction, and dynamically adjust the parameter weights of the host health decay model based on the overall energy efficiency improvement rate and the operational stability indicators.

[0008] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a load balancing distribution control method and system for multiple energy hosts operating in parallel. First, it acquires parameters such as the current output load, energy consumption rate, and historical maintenance records of each energy host in real time, providing a comprehensive data foundation for subsequent health assessment and energy efficiency analysis. Second, it introduces a host health decay model, innovatively coupling the accumulated factor of operating years and the maintenance effect recovery factor to dynamically calculate the real-time health coefficient of each host. Third, based on the real-time health coefficient and a preset standard energy efficiency benchmark curve, the standard curve is scaled using the accumulated factor of operating years and the maintenance effect recovery factor. Combined with current load demand queries, the theoretical energy consumption rate and two estimated energy consumption rates are obtained. Then, weighted summation is performed based on the ratio of the current load to the rated load to obtain an accurate energy efficiency deviation, providing a precise correction basis for load allocation optimization. Subsequently, a particle swarm optimization algorithm was used to solve the aforementioned optimization function. During the iteration process, the inertia weights were dynamically adjusted based on the cumulative factor of each host's operating years and the maintenance effect recovery factor, updating the particle velocity and position vectors. Strict constraints were applied to ensure that the algorithm could better consider the characteristics of hosts in different health states when searching for the optimal solution, improving optimization efficiency and solution accuracy. Finally, the overall energy efficiency improvement rate of the system and the operational stability indicators of each host were continuously monitored, and the parameter weights of the host health decay model were dynamically adjusted accordingly, forming a closed-loop adaptive optimization control mechanism. This mechanism continuously learns and adapts to changes in host performance, continuously optimizing the load allocation strategy.

[0009] Through the above technical solutions, this application maximizes the energy utilization efficiency of the entire parallel system while ensuring the stability of system operation. It effectively solves the problems of overall performance degradation and high energy consumption caused by ignoring the age and maintenance status differences of the main unit in the prior art, and provides strong technical support for the efficient and stable operation of parallel systems with multiple energy main units. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the load balancing distribution control method for multiple energy hosts operating in parallel, as provided in the embodiments of this application. Figure 2 This is a schematic diagram of the load balancing distribution control system for multiple energy hosts operating in parallel, provided in the embodiments of this application.

[0012] The components represented by each number in the attached diagram are explained below: Information acquisition module 11, data processing module 12, energy efficiency optimization module 13, instruction generation module 14, instruction execution module 15. Detailed Implementation

[0013] This application provides a load balancing distribution control method and system for multiple energy hosts operating in parallel, which addresses the technical problem that existing load distribution methods ignore the age and maintenance status of the hosts, leading to overall performance degradation.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0016] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0017] Example 1, as Figure 1 As shown, embodiments of this application provide a load balancing distribution control method for multiple energy mainframes operating in parallel, including: S10: Collect the real-time operating parameter set of each energy host in the parallel system, wherein the real-time operating parameter set includes at least the current output load, energy consumption rate and historical maintenance flags; In this embodiment, firstly, the real-time operating parameter set of each energy host in the parallel system is collected. Specifically, data is collected from each energy host operating in parallel at a preset sampling frequency, such as once per second or once every five seconds.

[0018] Furthermore, the real-time operating parameter set includes at least the current output load, energy consumption rate, and historical maintenance records. The current output load can be directly obtained through power sensors installed at the output of each host unit, and the unit is typically kilowatts or megawatts.

[0019] Energy consumption rate is calculated by collecting the fuel consumption of the main unit per unit time, such as cubic meters per hour for natural gas main units, liters per hour for diesel generators, or electricity consumption, such as kilowatt-hours per hour for electric main units, and combining it with its current output load. The unit is generally kilograms of standard coal per kilowatt-hour or kilowatt-hours per kilowatt-hour.

[0020] Historical maintenance records are structured data records that include the time of the most recent maintenance for each host unit, the maintenance type (e.g., preventative maintenance, fault repair, overhaul, replacement of critical components), maintenance duration, and performance test data before and after maintenance. Real-time operating parameter sets serve as the raw data for subsequent health assessments and energy efficiency analyses.

[0021] S20: Based on the real-time operating parameter set, the real-time health coefficient of each energy host is dynamically calculated through the host health decay model, wherein the host health decay model is coupled with the operating years accumulation factor and the maintenance effect recovery factor. In this embodiment, after obtaining the real-time operating parameter set, the real-time health coefficient of each energy host is dynamically calculated based on this using the host health decay model. This host health decay model innovatively couples the accumulated factor of operating years with the maintenance effect recovery factor to more comprehensively and dynamically reflect the actual health status of the energy host.

[0022] Specifically, the service life accumulation factor is a function that monotonically increases with the host's operational time, with an initial value of 0, gradually approaching 1 as the service life increases. The maintenance effect recovery factor quantifies the impact of maintenance actions on the host's health recovery, and its value ranges from 0 to 1. After the host undergoes maintenance, the value of this factor is adjusted according to the type and depth of the maintenance.

[0023] The steps for constructing the host health decay model include: Based on the equipment model and design parameters of each energy host in the parallel system, retrieve the corresponding standard energy efficiency benchmark curve and rated life data; Based on the historical operation database, sample data sets of each energy host are collected at different operating stages. The sample data sets include at least the cumulative operating years, the types of maintenance records, and the corresponding energy efficiency measurement data. Based on the rated life data and the cumulative operating years, a baseline curve characterizing the performance degradation over operating time is fitted and generated, and the quantitative calculation rules for the cumulative operating years factor are defined. The maintenance records were analyzed to distinguish the contribution weight and time period of different maintenance types to the energy efficiency recovery of the main unit, and the quantitative calculation rules of the maintenance effect recovery factor were defined. Using the cumulative operating years, maintenance records, and measured energy efficiency data as input samples, and the cumulative operating years factor and maintenance effect recovery factor calculated by the quantitative calculation rules as intermediate supervision labels, a model framework for the host health degradation model is constructed. The model framework is trained under supervision using the input samples and intermediate supervision labels until the error converges to an acceptable range, thus completing the model construction.

[0024] In this embodiment, the host health decay model outputs the real-time health coefficient by nonlinearly weighting and fusing the cumulative factor of operating years and the maintenance effect recovery factor.

[0025] First, based on the equipment model and design parameters of each energy host, retrieve the corresponding standard energy efficiency benchmark curve from the technical manual or pre-set equipment parameter database provided by the equipment manufacturer. This curve typically uses load factor (the percentage of current load to rated load) as the horizontal axis and energy efficiency values, such as thermal efficiency and power generation efficiency, as the vertical axis, reflecting the energy efficiency performance of the host at different load points under ideal conditions. Simultaneously, retrieve its rated lifespan data, such as a design operating hour of 80,000 hours or a design operating life of 20 years.

[0026] Secondly, based on the system's historical operation database, sample data sets are collected for each energy unit at different operational stages since its commissioning. This sample data set covers at least the cumulative operating years, i.e., the total operating time from commissioning to the present, accurate to the hour; the types of maintenance records, detailed into categories such as routine inspections, preventative maintenance, fault repairs, overhauls, and replacement of key components; and the corresponding energy efficiency measurement data, i.e., the actual energy efficiency values ​​at the aforementioned cumulative operating years milestones and within specific time periods before and after maintenance, such as the week before and after maintenance, calculated using real-time collected energy consumption rates and output loads.

[0027] Next, based on the rated lifespan data and cumulative operating years, a baseline curve characterizing performance degradation over operating time is fitted and generated. For example, if the rated lifespan is set to L years and the cumulative operating years to t years, the baseline calculation formula for the operating years accumulation factor F can be initially set as F = 1 - e^(-kt / L), where k is the degradation coefficient. This coefficient is determined by fitting the degradation ratio of measured energy efficiency data corresponding to different t values ​​in historical samples relative to the standard energy efficiency baseline curve. This ensures that when t = 0, F = 0, indicating a brand-new state with no degradation; and when t approaches L, F approaches 1, indicating severe performance degradation. Based on this, the quantitative calculation rules for the operating years accumulation factor are formally defined, specifying its value range as [0,1], with a larger value indicating more severe performance degradation due to operating years.

[0028] Then, the maintenance records are analyzed to distinguish the contribution weight and time period of different maintenance types to the energy efficiency recovery of the main unit. For example, the contribution weight of routine inspections is relatively small, possibly set to 0.05, and the time period is short, such as 3 days; the contribution weight of preventive maintenance is moderate, set to 0.2, and the time period is 1 month; the contribution weight of fault repair is between 0.1 and 0.3 depending on the severity of the fault, and the time period is 2 weeks; the contribution weight of overhaul is large, set to 0.6-0.8, and the time period is 6 months to 1 year; the contribution weight of replacing key components is the largest, reaching 0.8-0.95, and the time period is set according to the life of the component, such as 2 years after replacing a new blade. The quantitative calculation rule for the maintenance effect recovery factor F1 is defined as follows: For each maintenance, the initial recovery value R and the time period T are determined according to its type. From the moment the maintenance is completed, F1 starts from R and decreases with time t according to a certain decay law, such as exponential decay F1=Re^(-λt / T), where λ is the decay rate constant, gradually decreasing to 0. When there are multiple maintenance operations, the current maintenance effect recovery factor is the sum of the recovery factors generated by each maintenance operation, but the total recovery factor does not exceed 1.

[0029] Subsequently, using cumulative operating years and maintenance records as timestamps and type identifiers, and measured energy efficiency data as input samples, a model framework for the host health degradation model is constructed using the cumulative operating years factor and maintenance effect recovery factor initially calculated through the aforementioned quantitative calculation rules as intermediate supervision labels. This model framework can be a multilayer perceptron, where input layer neurons correspond to the various dimensions of the input samples, hidden layers are used to learn nonlinear mapping relationships, and the output layer outputs the final real-time health coefficient H. The model's loss function is defined as the mean square error between the output real-time health coefficient H and the target health coefficient derived from the actual degradation degree of measured energy efficiency data relative to the standard energy efficiency baseline curve.

[0030] Finally, the model framework is trained under supervised conditions using input samples and intermediate supervisory labels. During training, the weights and biases of each layer of the model are continuously adjusted, and the loss function is minimized using the backpropagation algorithm until the model's prediction error, i.e., the error between the real-time health coefficient output by the model and the health reference value calculated based on the measured energy efficiency data of the same period, is reached. If the root mean square error (RMSE) on the validation set converges to a preset allowable range, such as setting RMSE < 0.05, then the model is considered to be able to accurately calculate the real-time health coefficient through the coupling effect of the cumulative factor of operating years and the recovery factor of maintenance effect, thus completing the construction of the host health decay model.

[0031] Furthermore, the formula for calculating the real-time health coefficient H can be expressed as H=1-(F×(1-F1)), which comprehensively considers the decay caused by the years of operation and the recovery brought by maintenance. Its value range is (0,1], and the closer the value is to 1, the better the current health status of the host.

[0032] Specifically, the quantitative calculation rules for the cumulative factor of operating years include: The performance degradation inflection point age is set based on the rated life data. When the cumulative operating years are less than the performance degradation inflection point age, the operating years accumulation factor is calculated using a linear mild degradation function. When the cumulative operating years are greater than or equal to the performance degradation inflection point years, the operating years accumulation factor is calculated using an exponentially accelerated decay function.

[0033] In this embodiment, firstly, a performance degradation inflection point age is introduced into the quantitative calculation rules for the cumulative factor of operating years. Since the performance degradation trend of an energy host throughout its entire lifespan is not a simple linear or single exponential pattern, it typically degrades slowly in the early stages of operation, but the degradation rate accelerates significantly after a certain number of years. The performance degradation inflection point age is determined based on the rate of change of the degradation ratio of measured energy efficiency data relative to the standard energy efficiency benchmark curve in historical sample data. When the rate of change exceeds a preset threshold, for example, a sudden increase from an average annual degradation of 1% to an average annual degradation of 3%, the corresponding cumulative operating years are set as the performance degradation inflection point age for that model of host. For example, performance degradation inflection point age = rated lifespan × α, where α is a calibration coefficient, ranging from 0.6 to 0.8.

[0034] For example, a gas turbine main unit has a rated life of 20 years. Through analysis of its historical energy efficiency data, it was found that the performance degradation was relatively slow in the first 8 years of operation, with an average annual degradation of about 0.8%. However, after the 8th year, the degradation rate accelerated to an average of 2.5% per year. Therefore, the 8th year was set as the inflection point year for its performance degradation.

[0035] When the cumulative operation years \(t\) is less than the inflection point years \(t_0\), that is \(t < t_0\), a linear mild attenuation function is used to calculate the operation years cumulative factor \(F\). The function form can be set as \(F = a\times(t / t_0)\), where \(a\) is a proportionality coefficient less than 1, such as 0.3, to ensure that at the inflection point years, the \(F\) value calculated by the linear function is smoothly connected with the value of the subsequent exponential function at this point.

[0036] When the cumulative operation years \(t\) is greater than or equal to the inflection point years \(t_0\), that is \(t\geq t_0\), it switches to an exponential acceleration attenuation function. The function form can be set as \(F = F_0+(1 - F_0)\times(1 - e^{(-b\times(t - t_0) / (L - t_0))})\), where \(F_0\) is the operation years cumulative factor value calculated by the linear function when \(t = t_0\), \(b\) is the acceleration attenuation coefficient, and \(L\) is the rated life years.

[0037] Specifically, if \(t_0 = 8\) years and \(L = 20\) years, when \(t = 8\), \(F_0=a\times(8 / 8)=a = 0.3\). When \(t = 10\) years, substituting into the exponential acceleration attenuation function, \(F = 0.3+(1 - 0.3)\times(1 - e^{(-b\times(10 - 8) / (20 - 8))})=0.3 + 0.7\times(1 - e^{(-2b / 12)})\). By analyzing the measured energy efficiency attenuation data at \(t = 10\) years in historical data, assuming that the \(F\) target value corresponding to the actual attenuation ratio at this time is 0.45, the \(b\) value can be calculated by inverse deduction: \(0.45=0.3 + 0.7\times(1 - e^{(-b / 6)})\to0.15 / 0.7 = 1 - e^{(-b / 6)}\to e^{(-b / 6)}=1 - 0.15 / 0.7\approx0.7857\to -b / 6=\ln(0.7857)\approx - 0.241\), then \(b\approx1.446\).

[0038] Through the design of the piecewise function, the performance attenuation characteristics of the energy host in different operation stages are simulated, making the calculation of the operation years cumulative factor more in line with the actual situation and laying a foundation for the accurate evaluation of the subsequent health coefficient.

[0039] Furthermore, the quantitative calculation rules of the maintenance effect recovery factor include: Match the preset contribution weight coefficient according to the maintenance type of the历次维护记录; Calculate the remaining efficacy contribution value of a single maintenance according to the interval between the completion time point of each maintenance and the current time, combined with the preset time limit period of the corresponding maintenance type; Perform a weighted sum of all the remaining efficacy contribution values corresponding to the历次维护记录 to obtain the maintenance effect recovery factor.

[0040] It should be noted that there is an unclear expression "历次维护记录" in the original text. It may need to be further clarified according to the specific context for a more accurate translation.In this embodiment, firstly, for each maintenance record in the previous maintenance logs, according to the maintenance type, such as preventive maintenance, fault repair, overhaul, replacement of key components, etc., the corresponding coefficient value is matched from a preset maintenance contribution weight coefficient table. This coefficient table is constructed based on industry standard maintenance guidelines and historical energy efficiency recovery data statistics. For example, the contribution weight coefficient for replacing key components is set to 0.9, overhaul to 0.7, preventive maintenance to 0.3, daily inspection to 0.05, and fault repair fluctuates between 0.2 and 0.5 according to the severity of the fault, with higher values ​​for severe fault repair and lower values ​​for minor fault repair.

[0041] Secondly, for each maintenance, calculate its remaining effectiveness contribution value. Remaining effectiveness contribution value = contribution weight coefficient × time decay coefficient. If the time interval is greater than or equal to the effective period, the time decay coefficient = 0, indicating that the maintenance effect is completely ineffective; if the time interval is less than the effective period, the time decay coefficient = 1 - (time interval / effective period). The closer the time, the closer the coefficient is to 1, and the more the effect is retained.

[0042] For example, a certain energy main unit underwent a major overhaul three months ago. The preset contribution weighting coefficient for the overhaul was 0.7, corresponding to a time period of 180 days. The interval between the current time and the maintenance completion time is 90 days. Since 90 days < 180 days, the time decay coefficient = 1 - (90 / 180) = 0.5. Therefore, the remaining efficiency contribution value of this major overhaul = 0.7 × 0.5 = 0.35.

[0043] If the host underwent preventative maintenance 6 months ago, its contribution weighting coefficient is 0.3, and the effective period is 90 days. Since the current interval is 180 days ≥ 90 days, the time decay coefficient is 0. Therefore, the remaining effectiveness contribution value of this preventative maintenance is 0.3 × 0 = 0.

[0044] If the host performed a routine inspection one month ago, the contribution weight coefficient is 0.05, the time period is 15 days, the current interval is 30 days ≥ 15 days, the time decay coefficient is 0, and the remaining performance contribution value is 0.05 × 0 = 0.

[0045] At this point, the remaining performance contribution values ​​of each maintenance record are weighted and summed to obtain the maintenance effect recovery factor.

[0046] S30: Based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, solve the energy efficiency deviation of each energy host under the current load demand, and construct a load redistribution optimization function with the goal of minimizing the total energy consumption of the parallel system. In this embodiment of the application, firstly, the actual energy efficiency curve under the current health state is calculated based on the real-time health coefficient H of each energy host and its corresponding standard energy efficiency benchmark curve.

[0047] Specifically, the energy efficiency value at each load rate point on the standard energy efficiency baseline curve is multiplied by the real-time health coefficient H to obtain the actual energy efficiency value of the host under the current health status. That is, the actual energy efficiency curve = standard energy efficiency baseline curve × H. The energy efficiency deviation is defined as the percentage difference between the energy efficiency value of the standard energy efficiency baseline curve and the energy efficiency value of the actual energy efficiency curve at the same load rate, relative to the energy efficiency value of the standard energy efficiency baseline curve. It is used to quantify the degree of energy efficiency loss caused by the decline in health status.

[0048] Furthermore, the energy efficiency deviation of each energy host under the current load demand is solved. That is, for the load rate range that each host may undertake under the current total system load demand, the average deviation of the actual energy efficiency curve from the standard energy efficiency benchmark curve in this range is calculated.

[0049] Secondly, a load redistribution optimization function is constructed with the goal of minimizing the total energy consumption of the parallel system. The independent variable is the distributed load of each energy host, and the constraints include the maximum load limit, minimum load limit, load adjustment rate limit, and balance of total system load demand for each host.

[0050] Specifically, based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, the energy efficiency deviation of each energy unit under the current load demand is calculated, including: The standard energy efficiency benchmark curve is scaled using the accumulated factor of operating years to obtain the first equivalent energy efficiency curve; The standard energy efficiency baseline curve is scaled using the maintenance effect recovery factor to obtain a second equivalent energy efficiency curve; Based on the current load demand, the theoretical energy consumption rate, the first estimated energy consumption rate, and the second estimated energy consumption rate are obtained by querying the standard energy efficiency benchmark curve, the first equivalent energy efficiency curve, and the second equivalent energy efficiency curve, respectively. Calculate the percentage of the first difference between the first estimated energy consumption rate and the theoretical energy consumption rate relative to the theoretical energy consumption rate, as the first energy efficiency deviation component; Calculate the percentage of the second difference between the second estimated energy consumption rate and the theoretical energy consumption rate relative to the theoretical energy consumption rate, as the second energy efficiency deviation component; The first weight and the second weight are determined based on the ratio of the current load demand to the rated load corresponding to the standard energy efficiency benchmark curve. The energy efficiency deviation component and the second energy efficiency deviation component are weighted and summed according to the first weight and the second weight to obtain the energy efficiency deviation of each energy host.

[0051] In this embodiment, firstly, the energy consumption rate value of the standard energy efficiency reference curve is adjusted according to the ratio of "1 / accumulated operating years factor" to scale the standard energy efficiency reference curve and obtain the first equivalent energy efficiency curve. For example, if the accumulated operating years factor F=0.4, it means that the energy efficiency has decreased by 40% due to the operating years. Then, the theoretical energy consumption rate at a certain load rate point on the standard energy efficiency reference curve is 100kW. After scaling by the accumulated operating years factor, the first estimated energy consumption rate = 100 / (1-F) = 100 / 0.6≈166.67kW. That is, the energy consumption rate of the first equivalent energy efficiency curve at this point is 166.67kW, which directly reflects the negative impact of the operating years on energy consumption.

[0052] Secondly, the energy consumption rate value of the standard energy efficiency baseline curve is adjusted according to the ratio of "1 / maintenance effect recovery factor" to obtain the second equivalent energy efficiency curve. If the maintenance effect recovery factor F1=0.35, it means that maintenance brings a 35% energy efficiency recovery. Then the second estimated energy consumption rate = theoretical energy consumption rate / (1+F1) = 100 / (1+0.35)≈74.07kW. That is, the maintenance effect restores the energy consumption rate from the reduced 166.67kW to 74.07kW. The second equivalent energy efficiency curve reflects the improvement effect of maintenance on energy consumption.

[0053] Secondly, regarding the current total load demand of the system, assuming that the load that a certain energy host needs to bear accounts for 60% of its rated load, then look up the theoretical energy consumption rate E0, the first estimated energy consumption rate E1, and the second estimated energy consumption rate E2 corresponding to the 60% load rate on the standard energy efficiency benchmark curve, the first equivalent energy efficiency curve, and the second equivalent energy efficiency curve, respectively.

[0054] Then, the first energy efficiency deviation component D1 = (|E1-E0| / E0) × 100% is calculated, which reflects the energy consumption deviation ratio when only the operating years are considered; the second energy efficiency deviation component D2 = (|E0-E2| / E0) × 100% reflects the energy consumption improvement ratio when only the maintenance effect is considered, and a positive value here indicates a reduction in energy consumption.

[0055] Subsequently, the weights are determined based on the ratio R between the current load demand and the rated load corresponding to the standard energy efficiency baseline curve, i.e., based on the load factor. When the current load / rated load is ≤50%, the first weight is 1-(current load / rated load) and the second weight is the current load / rated load. When the current load / rated load is >50%, the first weight is the current load / rated load and the second weight is 1-(current load / rated load).

[0056] Specifically, when R = current load / rated load ≤ 50%, it indicates that the host is in a low-load operating state. At this time, the basic degradation caused by the years of operation is more significant. Therefore, the first weight is 1-R and the second weight is R. When R > 50%, the host is in a high-load operating state. The performance recovery brought by maintenance is more prominent under high load. Therefore, the first weight is R and the second weight is 1-R.

[0057] For example, if the current load factor R = 60% > 50%, then the first weight = 0.6, the second weight = 0.4, and the energy efficiency deviation D = 0.6 × D1 + 0.4 × D2. By dynamically adjusting the weights, the energy efficiency deviation calculation can better reflect the primary and secondary relationships of the actual energy efficiency influencing factors under different load conditions.

[0058] Furthermore, based on the current load demand, the theoretical energy consumption rate, the first estimated energy consumption rate, and the second estimated energy consumption rate are obtained by querying the standard energy efficiency benchmark curve, the first equivalent energy efficiency curve, and the second equivalent energy efficiency curve, respectively, including: Based on the current load demand, a matching search is performed on the discrete load-energy rate data points of the standard energy efficiency baseline curve; If a data point that perfectly matches the current load demand is found, the corresponding energy consumption rate is directly read as the theoretical energy consumption rate. If no data point is found that perfectly matches the current load demand, the two nearest data points are selected, and the theoretical energy consumption rate under the current load demand is obtained by linear interpolation. The load factor range is determined based on the ratio of the current load demand to the rated load of each energy host. If the load rate is lower than the preset low load threshold, the first estimated energy consumption rate is obtained on the first equivalent energy efficiency curve using the same method as querying the theoretical energy consumption rate, and the energy consumption rate of the corresponding data point on the standard energy efficiency benchmark curve is directly used as the second estimated energy consumption rate on the second equivalent energy efficiency curve. If the load rate is in the preset medium-high load range, the second estimated energy consumption rate is obtained on the second equivalent energy efficiency curve using the same method as querying the theoretical energy consumption rate, and the energy consumption rate of the corresponding data point on the standard energy efficiency benchmark curve is used as the first estimated energy consumption rate on the first equivalent energy efficiency curve.

[0059] In this embodiment, firstly, a matching search is performed on the discrete data points of the standard energy efficiency baseline curve. If the current load demand corresponds to a load rate of 25%, meaning no perfectly matching data point is found, then two neighboring data points, 20% and 30%, are selected. Assuming their corresponding theoretical energy consumption rates are E20 and E30, respectively, the theoretical energy consumption rate E at a 25% load rate is calculated using the linear interpolation formula E=E20+(E30-E20)×(25%-20%) / (30%-20%). If the current load rate is 30%, then the energy consumption rate of the data point corresponding to the 30% load rate is directly read as the theoretical energy consumption rate E0.

[0060] Secondly, determine the current load rate range. If the preset low load threshold is 30% of the rated load, when the calculated current load rate is 25%, which is lower than 30%, the first estimated energy consumption rate E1 is obtained by linear interpolation on the first equivalent energy efficiency curve. On the second equivalent energy efficiency curve, since the improvement effect of maintenance on energy efficiency under low load conditions is limited, the energy consumption rate of the corresponding load rate data point on the standard energy efficiency benchmark curve is directly used as the second estimated energy consumption rate E2, that is, the maintenance effect is considered negligible under low load conditions.

[0061] If the current load factor is 60%, falling within the medium-to-high load range of 30% to 100%, the second estimated energy consumption rate E2 is obtained through linear interpolation on the second equivalent energy efficiency curve. On the first equivalent energy efficiency curve, considering that the basic degradation over the years under medium-to-high load conditions has been fully reflected through previous function calculations, the energy consumption rate of the corresponding data point on the standard energy efficiency benchmark curve is directly used as the first estimated energy consumption rate E1, avoiding the impact of double-counting the years of operation. This interval-based differentiated processing further improves the accuracy and rationality of energy consumption rate queries under different load conditions, making the subsequent calculation basis for energy efficiency deviation more reliable.

[0062] S40: The load redistribution optimization function is solved using the particle swarm optimization algorithm, and personalized load setting instructions are dynamically generated and sent to each energy host. In this embodiment, the particle swarm parameters are first initialized. The position vector of each particle represents a set of allocated load values ​​for each energy host. The allocated load of each host must be between its minimum load limit and maximum load limit, and the sum of the components of all particle position vectors must be equal to the current total load demand of the system.

[0063] Secondly, a fitness function is constructed. With the goal of minimizing the total energy consumption of the parallel system, the distributed load of each energy host is substituted into its actual energy efficiency curve. Based on the curve corrected by the health coefficient, the energy consumption of a single host is calculated, which is the distributed load × energy consumption rate. The sum is used to obtain the total energy consumption of the system. The total energy consumption value is used as the particle fitness value. The smaller the fitness value, the better the particle position.

[0064] Secondly, personalized load setting instructions are dynamically generated and distributed to each energy host, enabling high-efficiency hosts to take on a larger proportion of the load, while low-efficiency hosts take on a smaller proportion of the load.

[0065] The load redistribution optimization function, which aims to minimize the total energy consumption of the parallel system, includes: The load to be allocated for each energy host in the parallel system is used as the optimization variable; The equation constraint is that the total load demand in the parallel system is equal to the sum of the load values ​​to be allocated by each energy host. The allowable load range of each energy host is used as an inequality constraint, wherein the lower limit of the allowable load range is determined based on the cumulative factor of the operating years, and the upper limit is determined based on the maintenance effect recovery factor; The energy efficiency deviation of each energy host is used as a correction factor for the corresponding host to calculate the estimated energy consumption under the load value to be allocated; The load redistribution optimization function is constructed with the goal of minimizing the sum of the estimated energy consumption of all energy hosts.

[0066] In this embodiment of the application, firstly, the load to be allocated of each energy host in the parallel system is used as the optimization variable. Let the load to be allocated of the i-th energy host be Pi, i=1,2,...,n, and n is the total number of energy hosts. Then the optimization variable vector is [P1,P2,...,Pn].

[0067] Secondly, the equality constraint balances the total system load demand, while the inequality constraint defines the allowable load range for each host, i.e., the lower limit of allowable load ≤ Pi ≤ the upper limit of allowable load. The lower limit of allowable load is dynamically adjusted based on the accumulated factor of operating years. The longer the operating years and the smaller the accumulated factor, the more severe the performance degradation of the host, and the lower the minimum stable operating load is correspondingly increased. The lower limit of allowable load is set as: rated load × (1 - accumulated factor of operating years × 0.3). The allowable load limit is adjusted based on the maintenance effect recovery factor. The larger the maintenance effect recovery factor, the better the performance recovery of the main unit after maintenance, and the stronger the maximum output capacity. The allowable load limit = rated load × (1 + maintenance effect recovery factor × 0.2). Among them, 0.3 and 0.2 are the basic coefficients, which can be calibrated according to the type of energy main unit and industry standards.

[0068] Furthermore, the energy efficiency deviation D of each energy host is used as a correction coefficient to adjust its estimated energy consumption under the assigned load Pi. Specifically, if the theoretical energy consumption rate of a certain energy host under the load rate corresponding to the assigned load Pi is E0 according to the standard energy efficiency benchmark curve, then the actual estimated energy consumption rate after considering the energy efficiency deviation is E = E0 × (1 + D / 100), where D is the energy efficiency deviation.

[0069] For example, if D=20%, then the actual estimated energy consumption rate E=E0×1.2, that is, the energy consumption increases by 20% due to the decline in health. Therefore, the estimated energy consumption of a single energy host is Pi×E=Pi×E0×(1+D / 100), and the total estimated energy consumption of the system is the sum of the estimated energy consumption of all hosts.

[0070] Finally, with the goal of minimizing the sum of the estimated energy consumption of all energy hosts, the load redistribution optimization function is constructed. The load redistribution optimization function is expressed as f(P1,P2,...,Pn)=Σ[Pi×E0i×(1+Di / 100)], where i=1,2,...,n, and must satisfy: ΣPi=total system load demand, allowable lower limit i≤Pi≤allowable upper limit i, allowable lower limit i=rated load i×(1-accumulated factor of operating years i×0.3), allowable upper limit i=rated load i×(1+maintenance effect recovery factor i×0.2). This optimization function directly incorporates the energy efficiency deviation of each host into the energy consumption calculation, enabling the optimization process to accurately reflect the impact of health differences on actual energy consumption, thereby achieving energy efficiency-priority load allocation.

[0071] Specifically, step S40 in the method includes: Initialize the particle swarm, where the position vector of each particle represents a set of load allocation schemes, each dimension of the position vector corresponds to a power host, and the initial value of each dimension is randomly generated within the allowable load range of the corresponding power host; The load redistribution optimization function is used as the fitness function of the particle swarm optimization algorithm to calculate the fitness value of each particle. In each iteration, the velocity vector and position vector of each particle are updated based on the individual particle's historical best position and the particle swarm's global best position. The inertia weight used in the update process is dynamically adjusted according to the cumulative factor of the operating years of the energy host and the maintenance effect recovery factor of the corresponding dimension. Perform constraint processing on the updated particle position vector; Determine whether the preset maximum number of iterations has been reached, or whether the improvement of the global optimal fitness value of the particle swarm within a preset number of consecutive iterations is lower than the preset convergence threshold; If any condition is met, the iteration is terminated, and the global optimal position vector of the particle swarm is taken as the optimal load allocation scheme. The optimal load allocation scheme is decoded into personalized load setting instructions for each energy host and sent to the control system of the corresponding energy host.

[0072] In this embodiment, the particle swarm is first initialized to a population size of 50, a maximum number of iterations of 100, an initial inertia weight of w=0.729, and learning factors c1=c2=1.494. The position vector dimension of each particle is consistent with the number of power hosts n. For example, when n=3 power hosts, the particle position vector is [P1,P2,P3], where P1, P2, and P3 are the allocated loads of the 3 power hosts, respectively. When generating the initial positions, for each power host Pi, a value is randomly selected between its lower and upper allowable load limits, and normalization is performed to ensure that ΣPi=the total system load requirement.

[0073] Next, calculate the particle fitness value. Substitute each Pi in the particle position vector into the actual energy efficiency curve of the corresponding host, and combine it with the energy consumption rate corrected by the health coefficient to calculate the energy consumption of a single unit Pi×Ei. Summing these values ​​gives the total system energy consumption as the fitness value.

[0074] For example, if a particle is located at [400, 300, 300]kW, and the Ei values ​​of the three main units are 0.8kW / kW, 0.75kW / kW, and 0.82kW / kW respectively, then the total energy consumption is 400×0.8+300×0.75+300×0.82=791kW, and the fitness value of the particle is 791.

[0075] Next, the inertia weight is dynamically adjusted. For example, the update formula for the inertia weight w is w = w1 × (1 - α × Fi - β × F1i), where w1 is the basic inertia weight of 0.729, α and β are weight coefficients, α = 0.2 and β = 0.15, Fi is the cumulative factor of the operating years of the i-th host, and F1i is its maintenance effect recovery factor. For a host corresponding to a certain dimension in the position vector, if its cumulative factor of operating years Fi = 0.4, indicating a 40% decay, and its maintenance effect recovery factor F1i = 0.35, indicating a 35% recovery, then the inertia weight correction term for that dimension = 0.2 × 0.4 + 0.15 × 0.35 = 0.1325, and after correction, w = 0.729 × (1 - 0.1325) = 0.729 × 0.8675 ≈ 0.633.

[0076] Furthermore, hosts with longer operating years and poorer maintenance and recovery performance have smaller inertial weights in their corresponding dimensions, which reduces the search step size of particles in that dimension and avoids unstable load distribution caused by low host health. Conversely, hosts with high health have larger inertial weights in their corresponding dimensions, enhancing search flexibility.

[0077] Subsequently, particle velocity and position are updated. After the update, the position vector needs to be constrained. If the load Pi allocated to a host exceeds the allowable range, it is truncated to the nearest boundary value, and the loads of other hosts are adjusted proportionally to meet the total load constraint.

[0078] Specifically, the components of the particle position vector are normalized and scaled so that the sum of the scaled components equals the total load requirement. For each component of the normalized and scaled particle position vector, boundary correction is performed according to the allowable load range of the corresponding power source, adjusting components exceeding the range to their corresponding boundary values. The boundary-corrected particle position vector is then normalized and scaled again so that the sum of the scaled components equals the total load requirement, and boundary correction is performed again. This normalization, scaling, and boundary correction process is repeated until the absolute error between the sum of the particle position vector components and the total load requirement is less than a preset tolerance threshold, and all components are within the allowable load range of the corresponding power source.

[0079] Finally, determine the iteration termination condition. The iteration terminates when the number of iterations reaches 100, or when the change in the global optimal fitness value Δf is less than 0.5kW in 20 consecutive iterations, which is when the convergence threshold is reached.

[0080] Finally, the optimal location vector is decoded into personalized load setting instructions to ensure that each host smoothly receives and executes load adjustments, thereby achieving the dynamic optimization goal of minimizing the total energy consumption of the parallel system.

[0081] The feature is that, in each iteration, the velocity vector and position vector of each particle are updated based on the individual particle's historical best position and the global best position of the particle swarm, including: An adaptive inertial weight is set for updating the velocity vector and position vector of each particle, wherein the value of the adaptive inertial weight is positively correlated with the cumulative factor of the operating years of the energy host in the corresponding dimension and negatively correlated with the maintenance effect recovery factor. Based on the individual particle's historical best position, the particle swarm's global best position, and the adaptive inertia weight, the particle's velocity vector is updated according to the preset basic formula of the particle swarm optimization algorithm. Add the updated velocity vector to the current particle position vector to obtain the updated position vector.

[0082] In this embodiment, firstly, an adaptive inertial weight is set for updating the velocity vector and position vector of each particle. The adaptive inertial weight = basic inertial weight + operating years influence coefficient × operating years cumulative factor - maintenance status influence coefficient × maintenance effect recovery factor.

[0083] Among them, the basic inertia weight value is an adjustable parameter of the particle swarm algorithm, for example, 0.5-0.7; the operating years influence coefficient is a preset constant greater than zero, used to adjust the positive influence of operating years on the weight, for example, 0.2; the maintenance status influence coefficient is a preset constant greater than zero, used to adjust the negative influence of maintenance status on the weight, for example, 0.15.

[0084] For example, when the cumulative factor Fi of the operating years of a certain energy host is large, such as Fi=0.6, it indicates that the equipment is aging more seriously and the maintenance effect recovery factor F1i is small, such as F1i=0.2, it indicates that the maintenance effect is limited. Then its adaptive inertia weight = 0.6+0.2×0.6-0.15×0.2=0.6+0.12-0.03=0.69. At this time, the inertia weight is high, which means that in the particle iteration process, the load allocation dimension corresponding to this host retains more historical search experience.

[0085] Secondly, the particle's velocity vector is updated, taking into account the particle's current motion inertia. This is controlled by an adaptive inertia weight, which is a dynamically adjusted inertia weight based on the cumulative factor of the corresponding energy host's operating years and the maintenance effect recovery factor. This weight is multiplied by the particle's current velocity to reflect the particle's characteristic of maintaining its original motion trend.

[0086] Simultaneously, individual cognitive and social cognitive components are introduced. The individual cognitive component is obtained by multiplying a learning factor by the difference between the particle's current position and its historical best position, guiding the particle towards the optimal solution it has previously found. The social cognitive component is obtained by multiplying another learning factor by the difference between the particle's current position and the global best position of the particle swarm, prompting the particle to move towards the global best solution found by the entire swarm. Superimposing these three components yields the updated particle velocity vector.

[0087] Finally, the updated velocity vector is directly added to the current particle position vector to obtain the particle's new position in the solution space, thus realizing the dynamic exploration and optimization of the load allocation scheme.

[0088] S50: After executing the personalized load setting command, monitor the overall energy efficiency improvement rate of the parallel system and the operating stability index of each energy host, and dynamically backtrack and adjust the parameter weights of the host health decay model based on the overall energy efficiency improvement rate and the operating stability index.

[0089] In this embodiment, the energy efficiency improvement rate is first calculated as (total system energy consumption before optimization - total system energy consumption after optimization) / total system energy consumption before optimization × 100%, and is calculated by collecting energy consumption data before and after optimization in real time. Operational stability indicators include the load fluctuation coefficient, start-stop frequency, and over-temperature alarm frequency of each energy host.

[0090] Among them, the load fluctuation coefficient must be controlled within ±5%, the number of start-stop cycles should not exceed once per hour, the over-temperature alarm frequency should not exceed 3 times per 24 hours, and if any indicator exceeds the threshold for 3 consecutive sampling cycles, the stability of the host operation is judged to have decreased.

[0091] Secondly, a parameter weight backtracking adjustment mechanism is established. The weighted combination of the overall energy efficiency improvement rate and the operational stability index is used as the trigger condition for model correction. When the energy efficiency improvement rate is lower than the preset target value, such as 5%, or when the comprehensive score of the stability index is lower than 80 points, the parameter weight adjustment process is initiated.

[0092] Specifically, a multi-objective optimization model is constructed with the weights of the cumulative factor of operating years, the weight of the maintenance effect recovery factor, and the weight of energy efficiency deviation in the health decay model as the parameters to be adjusted. The optimization objectives are to maximize the energy efficiency improvement rate and the comprehensive score of the stability index, with the sum of the weights of the three factors being 1 as a constraint.

[0093] The model is solved using the gradient descent method. If the energy efficiency improvement rate is insufficient, the weight of the maintenance effect recovery factor is increased to strengthen the positive impact of maintenance effect on the upper limit of load, so that the host with better performance recovery after maintenance can bear higher loads. If the stability decreases, the weight of the service life accumulation factor is increased to strictly limit the lower limit of the load of old hosts and avoid excessive load fluctuations due to excessive pursuit of energy efficiency.

[0094] In summary, compared with existing technologies, this application constructs a closed-loop control mechanism of "health perception - dynamic constraint - energy efficiency correction - adaptive optimization" by deeply integrating the health characteristic parameters of the energy host into the entire process of load allocation optimization.

[0095] In summary, the embodiments of this application have at least the following technical effects: This application provides a load balancing distribution control method for multiple energy hosts operating in parallel. First, it acquires parameters such as the current output load, energy consumption rate, and historical maintenance records of each energy host in real time, providing a comprehensive data foundation for subsequent health assessment and energy efficiency analysis. Second, it introduces a host health decay model, innovatively coupling the accumulated factor of operating years and the maintenance effect recovery factor to dynamically calculate the real-time health coefficient of each host. Third, based on the real-time health coefficient and a preset standard energy efficiency benchmark curve, the standard curve is scaled using the accumulated factor of operating years and the maintenance effect recovery factor. Combined with current load demand queries, the theoretical energy consumption rate and two estimated energy consumption rates are obtained. Then, weighted summation is performed based on the ratio of the current load to the rated load to obtain an accurate energy efficiency deviation, providing a precise correction basis for load distribution optimization. Subsequently, a particle swarm optimization algorithm is used to solve the aforementioned optimization function. During the iteration process, the inertia weight is dynamically adjusted based on the cumulative factor of the operating years of each host and the recovery factor of the maintenance effect, updating the velocity and position vectors of the particles, and strict constraint processing is applied. This allows the algorithm to better consider the characteristics of hosts in different health states when searching for the optimal solution, improving optimization efficiency and solution accuracy. Finally, the overall energy efficiency improvement rate of the system and the operational stability indicators of each host are continuously monitored, and the parameter weights of the host health decay model are dynamically adjusted accordingly, forming a closed-loop adaptive optimization control mechanism. This mechanism continuously learns and adapts to changes in host performance and continuously optimizes the load allocation strategy. Through the above technical solution, this application maximizes the energy utilization efficiency of the entire parallel system while ensuring system operational stability. It effectively solves the problems of overall performance degradation and high energy consumption caused by ignoring the age and maintenance status differences of hosts in existing technologies, providing strong technical support for the efficient and stable operation of parallel systems with multiple energy hosts.

[0096] Example 2, as Figure 2 As shown, based on the same inventive concept as the load balancing distribution control method for multiple energy hosts operating in parallel provided in Embodiment 1, this application also provides a load balancing distribution control system for multiple energy hosts operating in parallel, including: The information acquisition module 11 is used to acquire the real-time operating parameter set of each energy host in the parallel system, wherein the real-time operating parameter set includes at least the current output load, energy consumption rate and historical maintenance markers; Data processing module 12 is used to dynamically calculate the real-time health coefficient of each energy host based on the real-time operating parameter set and through the host health decay model, wherein the host health decay model is coupled with the operating years accumulation factor and the maintenance effect recovery factor. The energy efficiency optimization module 13 is used to solve the energy efficiency deviation of each energy host under the current load demand based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, and to construct a load redistribution optimization function with the goal of minimizing the total energy consumption of the parallel system. Instruction generation module 14 is used to solve the load redistribution optimization function using particle swarm optimization algorithm, and dynamically generate and send personalized load setting instructions to each energy host. The instruction execution module 15 is used to monitor the overall energy efficiency improvement rate of the parallel system and the operational stability index of each energy host after executing the personalized load setting instruction, and dynamically backtrack and adjust the parameter weights of the host health decay model based on the overall energy efficiency improvement rate and the operational stability index.

[0097] Furthermore, in one embodiment of the application, the step of constructing the host health decay model includes: Based on the equipment model and design parameters of each energy host in the parallel system, retrieve the corresponding standard energy efficiency benchmark curve and rated life data; Based on the historical operation database, sample data sets of each energy host are collected at different operating stages. The sample data sets include at least the cumulative operating years, the types of maintenance records, and the corresponding energy efficiency measurement data. Based on the rated life data and the cumulative operating years, a baseline curve characterizing the performance degradation over operating time is fitted and generated, and the quantitative calculation rules for the cumulative operating years factor are defined. The maintenance records were analyzed to distinguish the contribution weight and time period of different maintenance types to the energy efficiency recovery of the main unit, and the quantitative calculation rules of the maintenance effect recovery factor were defined. Using the cumulative operating years, maintenance records, and measured energy efficiency data as input samples, and the cumulative operating years factor and maintenance effect recovery factor calculated by the quantitative calculation rules as intermediate supervision labels, a model framework for the host health degradation model is constructed. The model framework is trained under supervision using the input samples and intermediate supervision labels until the error converges to an acceptable range, thus completing the model construction.

[0098] Furthermore, in one embodiment of the application, the quantitative calculation rule for the cumulative factor of operating years includes: The performance degradation inflection point age is set based on the rated life data. When the cumulative operating years are less than the performance degradation inflection point age, the operating years accumulation factor is calculated using a linear mild degradation function. When the cumulative operating years are greater than or equal to the performance degradation inflection point years, the operating years accumulation factor is calculated using an exponentially accelerated decay function.

[0099] Furthermore, in one embodiment of the application, the quantitative calculation rule for the maintenance effect recovery factor includes: Based on the maintenance type of each maintenance record, a preset contribution weight coefficient is matched; Based on the interval between the completion time of each maintenance and the current time, and combined with the preset time period of the corresponding maintenance type, the remaining efficiency contribution value of a single maintenance is calculated. The maintenance effect recovery factor is obtained by weighted summing of all remaining performance contribution values ​​corresponding to each maintenance record.

[0100] Furthermore, based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, the energy efficiency deviation of each energy unit under the current load demand is calculated, including: The standard energy efficiency benchmark curve is scaled using the accumulated factor of operating years to obtain the first equivalent energy efficiency curve; The standard energy efficiency baseline curve is scaled using the maintenance effect recovery factor to obtain a second equivalent energy efficiency curve; Based on the current load demand, the theoretical energy consumption rate, the first estimated energy consumption rate, and the second estimated energy consumption rate are obtained by querying the standard energy efficiency benchmark curve, the first equivalent energy efficiency curve, and the second equivalent energy efficiency curve, respectively. Calculate the percentage of the first difference between the first estimated energy consumption rate and the theoretical energy consumption rate relative to the theoretical energy consumption rate, as the first energy efficiency deviation component; Calculate the percentage of the second difference between the second estimated energy consumption rate and the theoretical energy consumption rate relative to the theoretical energy consumption rate, as the second energy efficiency deviation component; The first weight and the second weight are determined based on the ratio of the current load demand to the rated load corresponding to the standard energy efficiency benchmark curve. The energy efficiency deviation component and the second energy efficiency deviation component are weighted and summed according to the first weight and the second weight to obtain the energy efficiency deviation of each energy host.

[0101] Furthermore, based on the current load demand, the theoretical energy consumption rate, the first estimated energy consumption rate, and the second estimated energy consumption rate are obtained by querying the standard energy efficiency benchmark curve, the first equivalent energy efficiency curve, and the second equivalent energy efficiency curve, respectively, including: Based on the current load demand, a matching search is performed on the discrete load-energy rate data points of the standard energy efficiency baseline curve; If a data point that perfectly matches the current load demand is found, the corresponding energy consumption rate is directly read as the theoretical energy consumption rate. If no data point is found that perfectly matches the current load demand, the two nearest data points are selected, and the theoretical energy consumption rate under the current load demand is obtained by linear interpolation. The load factor range is determined based on the ratio of the current load demand to the rated load of each energy host. If the load rate is lower than the preset low load threshold, the first estimated energy consumption rate is obtained on the first equivalent energy efficiency curve using the same method as querying the theoretical energy consumption rate, and the energy consumption rate of the corresponding data point on the standard energy efficiency benchmark curve is directly used as the second estimated energy consumption rate on the second equivalent energy efficiency curve. If the load rate is in the preset medium-high load range, the second estimated energy consumption rate is obtained on the second equivalent energy efficiency curve using the same method as querying the theoretical energy consumption rate, and the energy consumption rate of the corresponding data point on the standard energy efficiency benchmark curve is used as the first estimated energy consumption rate on the first equivalent energy efficiency curve.

[0102] Furthermore, in one embodiment of the application, a load redistribution optimization function is constructed with the objective of minimizing the total energy consumption of the parallel system, including: The load to be allocated for each energy host in the parallel system is used as the optimization variable; The equation constraint is that the total load demand in the parallel system is equal to the sum of the load values ​​to be allocated by each energy host. The allowable load range of each energy host is used as an inequality constraint, wherein the lower limit of the allowable load range is determined based on the cumulative factor of the operating years, and the upper limit is determined based on the maintenance effect recovery factor; The energy efficiency deviation of each energy host is used as a correction factor for the corresponding host to calculate the estimated energy consumption under the load value to be allocated; The load redistribution optimization function is constructed with the goal of minimizing the sum of the estimated energy consumption of all energy hosts.

[0103] In one embodiment, the instruction generation module 14 is specifically used for: The load redistribution optimization function is solved using a particle swarm optimization algorithm, and personalized load setting instructions are dynamically generated and distributed to each energy host, including: Initialize the particle swarm, where the position vector of each particle represents a set of load allocation schemes, each dimension of the position vector corresponds to a power host, and the initial value of each dimension is randomly generated within the allowable load range of the corresponding power host; The load redistribution optimization function is used as the fitness function of the particle swarm optimization algorithm to calculate the fitness value of each particle. In each iteration, the velocity vector and position vector of each particle are updated based on the individual particle's historical best position and the particle swarm's global best position. The inertia weight used in the update process is dynamically adjusted according to the cumulative factor of the operating years of the energy host and the maintenance effect recovery factor of the corresponding dimension. Perform constraint processing on the updated particle position vector; Determine whether the preset maximum number of iterations has been reached, or whether the improvement of the global optimal fitness value of the particle swarm within a preset number of consecutive iterations is lower than the preset convergence threshold; If any condition is met, the iteration is terminated, and the global optimal position vector of the particle swarm is taken as the optimal load allocation scheme. The optimal load allocation scheme is decoded into personalized load setting instructions for each energy host and sent to the control system of the corresponding energy host.

[0104] Further, in one embodiment, the method is characterized in that, in each iteration, the velocity vector and position vector of each particle are updated based on the individual particle's historical best position and the swarm's global best position, including: An adaptive inertial weight is set for updating the velocity vector and position vector of each particle, wherein the value of the adaptive inertial weight is positively correlated with the cumulative factor of the operating years of the energy host in the corresponding dimension and negatively correlated with the maintenance effect recovery factor. Based on the individual particle's historical best position, the particle swarm's global best position, and the adaptive inertia weight, the particle's velocity vector is updated according to the preset basic formula of the particle swarm optimization algorithm. Add the updated velocity vector to the current particle position vector to obtain the updated position vector.

[0105] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0106] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0107] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A load balancing distribution control method for multiple energy generators operating in parallel, characterized in that, The method includes: Collect real-time operating parameter sets of each energy host in the parallel system, wherein the real-time operating parameter sets include at least the current output load, energy consumption rate and historical maintenance flags; Based on the real-time operating parameter set, the real-time health coefficient of each energy host is dynamically calculated through the host health decay model, wherein the host health decay model is coupled with the operating years accumulation factor and the maintenance effect recovery factor. Based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, the energy efficiency deviation of each energy host under the current load demand is calculated, and a load redistribution optimization function with the goal of minimizing the total energy consumption of the parallel system is constructed. The load redistribution optimization function is solved using the particle swarm optimization algorithm, and personalized load setting instructions are dynamically generated and sent to each energy host. After executing the personalized load setting command, the overall energy efficiency improvement rate of the parallel system and the operational stability index of each energy host are monitored, and the parameter weights of the host health decay model are dynamically adjusted back based on the overall energy efficiency improvement rate and the operational stability index.

2. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 1, characterized in that, The steps for constructing the host health decay model include: Based on the equipment model and design parameters of each energy host in the parallel system, retrieve the corresponding standard energy efficiency benchmark curve and rated life data; Based on the historical operation database, sample data sets of each energy host are collected at different operating stages. The sample data sets include at least the cumulative operating years, the types of maintenance records, and the corresponding energy efficiency measurement data. Based on the rated life data and the cumulative operating years, a baseline curve characterizing the performance degradation over operating time is fitted and generated, and the quantitative calculation rules for the cumulative operating years factor are defined. The maintenance records were analyzed to distinguish the contribution weight and time period of different maintenance types to the energy efficiency recovery of the main unit, and the quantitative calculation rules of the maintenance effect recovery factor were defined. Using the cumulative operating years, maintenance records, and measured energy efficiency data as input samples, and the cumulative operating years factor and maintenance effect recovery factor calculated by the quantitative calculation rules as intermediate supervision labels, a model framework for the host health degradation model is constructed. The model framework is trained under supervision using the input samples and intermediate supervision labels until the error converges to an acceptable range, thus completing the model construction.

3. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 2, characterized in that, The quantitative calculation rules for the cumulative factor based on the number of years of operation include: The performance degradation inflection point age is set based on the rated life data. When the cumulative operating years are less than the performance degradation inflection point age, the operating years accumulation factor is calculated using a linear mild degradation function. When the cumulative operating years are greater than or equal to the performance degradation inflection point years, the operating years accumulation factor is calculated using an exponentially accelerated decay function.

4. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 2, characterized in that, The quantitative calculation rules for the maintenance effect recovery factor include: Based on the maintenance type of each maintenance record, a preset contribution weight coefficient is matched; Based on the interval between the completion time of each maintenance and the current time, and combined with the preset time period of the corresponding maintenance type, the remaining efficiency contribution value of a single maintenance is calculated. The maintenance effect recovery factor is obtained by weighted summing of all remaining performance contribution values ​​corresponding to each maintenance record.

5. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 1, characterized in that, Based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, the energy efficiency deviation of each energy unit under the current load demand is calculated, including: The standard energy efficiency benchmark curve is scaled using the accumulated factor of operating years to obtain the first equivalent energy efficiency curve; The standard energy efficiency baseline curve is scaled using the maintenance effect recovery factor to obtain a second equivalent energy efficiency curve; Based on the current load demand, the theoretical energy consumption rate, the first estimated energy consumption rate, and the second estimated energy consumption rate are obtained by querying the standard energy efficiency benchmark curve, the first equivalent energy efficiency curve, and the second equivalent energy efficiency curve, respectively. Calculate the percentage of the first difference between the first estimated energy consumption rate and the theoretical energy consumption rate relative to the theoretical energy consumption rate, as the first energy efficiency deviation component; Calculate the percentage of the second difference between the second estimated energy consumption rate and the theoretical energy consumption rate relative to the theoretical energy consumption rate, as the second energy efficiency deviation component; The first weight and the second weight are determined based on the ratio of the current load demand to the rated load corresponding to the standard energy efficiency benchmark curve. The energy efficiency deviation component and the second energy efficiency deviation component are weighted and summed according to the first weight and the second weight to obtain the energy efficiency deviation of each energy host.

6. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 5, characterized in that, Based on the current load demand, the theoretical energy consumption rate, the first estimated energy consumption rate, and the second estimated energy consumption rate are obtained by querying the standard energy efficiency benchmark curve, the first equivalent energy efficiency curve, and the second equivalent energy efficiency curve, respectively, including: Based on the current load demand, a matching search is performed on the discrete load-energy rate data points of the standard energy efficiency baseline curve; If a data point that perfectly matches the current load demand is found, the corresponding energy consumption rate is directly read as the theoretical energy consumption rate. If no data point is found that perfectly matches the current load demand, the two nearest data points are selected, and the theoretical energy consumption rate under the current load demand is obtained by linear interpolation. The load factor range is determined based on the ratio of the current load demand to the rated load of each energy host. If the load rate is lower than the preset low load threshold, the first estimated energy consumption rate is obtained on the first equivalent energy efficiency curve using the same method as querying the theoretical energy consumption rate, and the energy consumption rate of the corresponding data point on the standard energy efficiency benchmark curve is directly used as the second estimated energy consumption rate on the second equivalent energy efficiency curve. If the load rate is in the preset medium-high load range, the second estimated energy consumption rate is obtained on the second equivalent energy efficiency curve using the same method as querying the theoretical energy consumption rate, and the energy consumption rate of the corresponding data point on the standard energy efficiency benchmark curve is used as the first estimated energy consumption rate on the first equivalent energy efficiency curve.

7. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 1, characterized in that, Construct a load redistribution optimization function with the objective of minimizing the total energy consumption of the parallel system, including: The load to be allocated for each energy host in the parallel system is used as the optimization variable; The equation constraint is that the total load demand in the parallel system is equal to the sum of the load values ​​to be allocated by each energy host. The allowable load range of each energy host is used as an inequality constraint, wherein the lower limit of the allowable load range is determined based on the cumulative factor of the operating years, and the upper limit is determined based on the maintenance effect recovery factor; The energy efficiency deviation of each energy host is used as a correction factor for the corresponding host to calculate the estimated energy consumption under the load value to be allocated; The load redistribution optimization function is constructed with the goal of minimizing the sum of the estimated energy consumption of all energy hosts.

8. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 1, characterized in that, The load redistribution optimization function is solved using a particle swarm optimization algorithm, and personalized load setting instructions are dynamically generated and distributed to each energy host, including: Initialize the particle swarm, where the position vector of each particle represents a set of load allocation schemes, each dimension of the position vector corresponds to a power host, and the initial value of each dimension is randomly generated within the allowable load range of the corresponding power host; The load redistribution optimization function is used as the fitness function of the particle swarm optimization algorithm to calculate the fitness value of each particle. In each iteration, the velocity vector and position vector of each particle are updated based on the individual particle's historical best position and the particle swarm's global best position. The inertia weight used in the update process is dynamically adjusted according to the cumulative factor of the operating years of the energy host and the maintenance effect recovery factor of the corresponding dimension. Perform constraint processing on the updated particle position vector; Determine whether the preset maximum number of iterations has been reached, or whether the improvement of the global optimal fitness value of the particle swarm within a preset number of consecutive iterations is lower than the preset convergence threshold; If any condition is met, the iteration is terminated, and the global optimal position vector of the particle swarm is taken as the optimal load allocation scheme. The optimal load allocation scheme is decoded into personalized load setting instructions for each energy host and sent to the control system of the corresponding energy host.

9. The load balancing distribution control method for parallel operation of multiple energy generators according to claim 8, characterized in that, In each iteration, the velocity and position vectors of each particle are updated based on the individual particle's historical best position and the swarm's global best position, including: An adaptive inertial weight is set for updating the velocity vector and position vector of each particle, wherein the value of the adaptive inertial weight is positively correlated with the cumulative factor of the operating years of the energy host in the corresponding dimension and negatively correlated with the maintenance effect recovery factor. Based on the individual particle's historical best position, the particle swarm's global best position, and the adaptive inertia weight, the particle's velocity vector is updated according to the preset basic formula of the particle swarm optimization algorithm. Add the updated velocity vector to the current particle position vector to obtain the updated position vector.

10. A load balancing distribution control system for multiple energy generators operating in parallel, characterized in that, The load balancing distribution control method for parallel operation of multiple energy generators as described in any one of claims 1-9 includes: The information acquisition module is used to collect the real-time operating parameter set of each energy host in the parallel system, wherein the real-time operating parameter set includes at least the current output load, energy consumption rate and historical maintenance markers; The data processing module is used to dynamically calculate the real-time health coefficient of each energy host based on the real-time operating parameter set and through the host health decay model, wherein the host health decay model is coupled with the operating years accumulation factor and the maintenance effect recovery factor. The energy efficiency optimization module is used to solve the energy efficiency deviation of each energy host under the current load demand based on the real-time health coefficient and the preset standard energy efficiency benchmark curve, and to construct a load redistribution optimization function with the goal of minimizing the total energy consumption of the parallel system. The instruction generation module is used to solve the load redistribution optimization function using the particle swarm optimization algorithm, and dynamically generate and send personalized load setting instructions to each energy host. The instruction execution module is used to monitor the overall energy efficiency improvement rate of the parallel system and the operational stability indicators of each energy host after executing the personalized load setting instruction, and dynamically adjust the parameter weights of the host health decay model based on the overall energy efficiency improvement rate and the operational stability indicators.

Images