Primary energy efficient charging host's charging control system and method

Abstract

The application discloses a charging control system and method of a primary energy efficiency charging host, and relates to the technical field of energy efficiency optimization of charging hosts. The method comprises the following steps: acquiring power parameters of current charging demand, and based on the power parameters and intrinsic state parameters of a target charging host, enumerating and acquiring all feasible on-off combinations of power supply on-off switches; based on an efficiency prediction model, predicting apparent system efficiency of the target charging host under each on-off combination; according to multiple apparent system efficiencies, selecting an on-off combination corresponding to the highest apparent system efficiency as a target on-off combination, and controlling on-off states of the power supply on-off switches in response to the target on-off combination. The application solves the problems of low energy efficiency under light load working conditions, high standby power consumption, rough switching strategy and lack of energy efficiency prediction mechanism in the prior art, and improves the energy efficiency level of the charging host in the full load range.

Description

Technical Field

[0001] This invention relates to the field of energy efficiency optimization technology for charging hosts, specifically to a charging control system and method for a Level 1 energy efficiency charging host. Background Technology

[0002] Current mainstream charging hosts generally adopt a multi-power module parallel architecture, with multiple DC-DC power modules sharing the charging load. However, the conversion efficiency of power modules varies non-linearly with the load rate, with a significant drop in efficiency under light load and peak efficiency under medium to high load. Existing module management methods for charging hosts have the following shortcomings: fixed module activation schemes keep all modules powered on at all times, resulting in excessively low load rates for each module under light load, causing the overall efficiency to fall far short of Level 1 energy efficiency requirements, and high standby power consumption; simple threshold switching schemes use static empirical thresholds, failing to consider the impact of operating condition changes on the efficiency curve, lacking an energy efficiency prediction mechanism, and posing a risk of efficiency reduction after switching; redundancy management schemes based on current sharing prioritize reliability and do not involve energy efficiency-based module start-stop control.

[0003] Therefore, there is an urgent need for a charging control method that can dynamically adjust the number of online modules and accurately predict the system efficiency under different configurations. Summary of the Invention

[0004] This invention addresses the technical problems of low energy efficiency, high standby power consumption, crude switching strategies, and lack of energy efficiency prediction mechanisms in existing multi-module parallel charging hosts. It provides a charging control system and method for a Level 1 energy efficiency charging host.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] In a first aspect, the present invention provides a charging control system for a Class 1 energy efficiency charging host, comprising:

[0007] The combined enumeration module is used to obtain the power parameters of the current charging demand, and based on the power parameters and the intrinsic state parameters of the target charging host, enumerate all feasible on / off combinations of power supply switches.

[0008] An efficiency prediction module is used to predict the apparent system efficiency of the target charging host under each of the on / off combinations based on a pre-built efficiency prediction model, wherein the apparent system efficiency is the efficiency under the relative load balancing strategy.

[0009] The on / off control module is used to select the on / off combination corresponding to the highest apparent system efficiency as the target on / off combination, and to control the on / off state of multiple power supply on / off switches in the target charging host in response to the target on / off combination.

[0010] Secondly, the present invention provides a charging control method for a Class 1 energy efficiency charging host, comprising:

[0011] Obtain the power parameters of the current charging demand, and based on the power parameters and the intrinsic state parameters of the target charging host, enumerate and obtain all feasible on / off combinations of power supply switches.

[0012] Based on a pre-built efficiency prediction model, the apparent system efficiency of the target charging host under each of the on / off combinations is predicted, wherein the apparent system efficiency is the efficiency under the relative load balancing strategy.

[0013] Based on the apparent system efficiency, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination, and the on / off state control of the multiple power supply on / off switches in the target charging host is performed in response to the target on / off combination.

[0014] The beneficial effects of this invention are:

[0015] Compared to existing technologies, this invention first enumerates all feasible power supply switching combinations to construct a complete candidate configuration space, avoiding the loss of better energy efficiency solutions due to configuration omissions. Secondly, based on an efficiency prediction model that integrates physical models and data corrections, it predicts the system efficiency under each switching combination, solving the problem that traditional threshold switching cannot predict post-switching efficiency performance and preventing ineffective switching. Thirdly, using the apparent system efficiency under a relative load balancing strategy as the optimization target, it selects the optimal switching combination for control, ensuring that each online module avoids the low-efficiency range under light load and always operates in the high-efficiency load range, improving overall efficiency by 3% to 8% under light load conditions. Furthermore, all relays can be disconnected when there is no charging demand, achieving zero standby power consumption. This invention solves the technical problems of low energy efficiency under light load, high standby power consumption, and coarse switching strategies in existing technologies, meeting the Level 1 energy efficiency requirements of the charging host across the entire load range. Attached Figure Description

[0016] Figure 1 A schematic diagram of the charging control system of the first-level energy efficiency charging host provided by the present invention;

[0017] Figure 2 This is a flowchart illustrating the charging control method for a Level 1 energy efficiency charging host provided by the present invention.

[0018] In the attached diagram, the components represented by each number are as follows:

[0019] Combined enumeration module 11, efficiency prediction module 12, on / off control module 13. Detailed Implementation

[0020] Example 1, as Figure 1As shown, this embodiment of the invention provides a charging control system for a Level 1 energy efficiency charging host, including:

[0021] The combination enumeration module 11 is used to obtain the power parameters of the current charging demand, and based on the power parameters and the intrinsic state parameters of the target charging host, enumerate and obtain all feasible on / off combinations of power supply switches.

[0022] First, in electric vehicle charging scenarios, the charging host needs to dynamically adjust its operating status according to the vehicle's charging demand. The charging host is usually composed of multiple power modules connected in parallel. Each power module has a relay connected in series in its power supply path as a power on / off switch. By controlling the on / off state of the relay, it can be determined whether the module participates in power output online.

[0023] The power parameter of the current charging demand refers to the charging power value requested by the vehicle's battery management system from the charging host. For example, when the vehicle's current battery state of charge is low, it requests a charging power of 60kW. The intrinsic state parameters of the target charging host include the number of power modules, the rated power value of each power module, and the current on / off state of each power supply switch. The intrinsic state parameters determine the hardware configuration and capability boundaries of the charging host.

[0024] Specifically, the core of this step lies in enumerating all feasible power supply switching combinations. Different switching combinations result in different load rates for each module and varying overall system efficiency. Only by comprehensively enumerating and evaluating each combination can the optimal combination that maximizes system efficiency be found. For example, for a charging host containing N power modules, theoretically there are 2^N switching combinations, but not all combinations are feasible. Feasible switching combinations must satisfy power capacity constraints, meaning the total rated power of the online modules is not less than the current charging demand. For instance, when the charging demand is 60kW and each module has a rated power of 30kW, at least two modules need to be online; therefore, combinations with only one online module are not feasible. By enumerating all feasible combinations, a complete candidate solution space can be provided for subsequent energy efficiency prediction and optimal configuration selection.

[0025] Specifically, the power parameters of the current charging demand are obtained, and based on the power parameters and the intrinsic state parameters of the target charging host, all feasible on / off combinations of the power supply switch are enumerated, including:

[0026] Obtain the current charging demand and extract the corresponding power parameters;

[0027] Obtain the intrinsic state parameters of the target charging host, wherein the intrinsic state parameters include at least the number of power modules, the rated power value of each power module, and the current on / off state of each power supply switch.

[0028] Based on intrinsic state parameters, determine the consistency of multiple power supply on / off switches;

[0029] If they are consistent, then the number of connected switches is used as the only variable to perform simplified combination enumeration on multiple power supply switches to obtain multiple on / off combinations.

[0030] If they are inconsistent, multiple power supply switches are enumerated in a heterogeneous combination manner based on the combination enumeration method to obtain multiple on / off combinations.

[0031] First, the current charging demand is obtained, and the corresponding power parameters are extracted. The current charging demand refers to the charging request sent by the vehicle's battery management system to the charging host, which includes a target charging power value, such as 60kW. The power parameters are the power values ​​in this request, used to determine how many power modules need to be activated to meet the charging demand.

[0032] Secondly, the intrinsic state parameters of the target charging host are obtained. These parameters include at least the number of power modules, the rated power value of each power module, and the current on / off state of each power supply switch. The number of power modules refers to the total number of DC-DC power modules installed inside the charging host, for example, four. The rated power value of each power module refers to the maximum power that each module can output under rated operating conditions, for example, 30kW per module. The current on / off state of each power supply switch records whether the relay corresponding to each power module is in the ON or OFF state at the current moment. An ON state indicates that the power module is online and participating in power output, while an OFF state indicates that the power module is offline and not participating in power output. This current on / off state information is used to identify the modules that differ between the current on / off combination and the target on / off combination during subsequent switching decisions, determining which power modules need to be turned on and which need to be turned off.

[0033] Furthermore, based on intrinsic state parameters, the consistency of multiple power supply on / off switches is determined. Consistency refers to whether all power modules have the same rated power and the same efficiency characteristics. When all modules are of the same model and have not undergone differentiated aging, they can be considered consistent; when the module models are different or the aging degrees differ significantly, the modules are inconsistent. The purpose of determining consistency is to select different enumeration strategies to reduce computational complexity.

[0034] Specifically, if the quantities are identical, the number of connected switches is used as the sole variable to perform simplified combination enumeration of multiple power supply on / off switches, obtaining multiple on / off combinations. When the rated power and efficiency characteristics of each power module are completely identical, the on / off combination of the power supply on / off switches depends only on the number of switches in the on state, and is independent of which specific switch is on. For example, if the charging host contains four identical power modules, activating any two of them will result in the same system energy efficiency performance, without needing to distinguish whether the combination of module 1 and module 2 or module 1 and module 3 is activated. Therefore, it is only necessary to enumerate all integer values ​​between the minimum value that satisfies the power capacity constraint and the total number of power modules, such as the number of online modules being 2, 3, or 4, without enumerating the specific permutations and combinations of modules, thus significantly reducing the number of enumerations.

[0035] Furthermore, if inconsistencies exist, a heterogeneous combination enumeration method is used to enumerate multiple power supply on / off switches to obtain multiple on / off combinations. When the modules are inconsistent, different combinations of modules will produce different energy efficiency performances, so it is necessary to enumerate all possible module subsets. For example, if two modules are enabled out of four, there are six different combinations, each of which needs to be evaluated separately. Through combination enumeration, it is ensured that no possible solutions that are better than the current configuration are overlooked, providing a complete candidate space for subsequent energy efficiency optimization.

[0036] Finally, multiple on / off combinations are obtained, which represent different numbers of online modules or different module combination schemes that meet the current charging demand power capacity constraints. These combinations are used to provide input parameters for the subsequent efficiency prediction model, evaluate the system energy efficiency performance under each configuration, and select the optimal configuration with the highest efficiency.

[0037] The efficiency prediction module 12 is used to predict the apparent system efficiency of the target charging host under each on / off combination based on a pre-built efficiency prediction model, wherein the apparent system efficiency is the efficiency under the relative load balancing strategy.

[0038] Secondly, after enumerating all feasible on / off combinations, it is necessary to evaluate the system energy efficiency performance under each combination in order to select the optimal configuration. The core of this step lies in the pre-built efficiency prediction model, which can predict the corresponding system efficiency based on the load rate of each online module in the on / off combination.

[0039] The efficiency prediction model employs a hybrid architecture combining a physical model and a data correction model. The physical model is built upon the loss decomposition principle of power modules, including components such as conduction loss, switching loss, core loss, and fixed loss, providing a basic efficiency prediction. The data correction model uses machine learning methods to learn the residual between the physical model's prediction and the actual efficiency, compensating for non-ideal factors not considered by the physical model, such as device aging, parasitic parameter changes, and temperature coupling effects. The combination of these two approaches enables high-precision predictions even with limited training data.

[0040] Specifically, the construction of the efficiency prediction model includes:

[0041] By combining the intrinsic state parameters of the target charging host with the prior knowledge base, a physical efficiency model corresponding to each power supply on / off switch is constructed and calibrated.

[0042] Obtain the historical charging management logs of the target charging host;

[0043] Based on historical charging management logs, a physical efficiency model is introduced to calculate the sample physical predicted efficiency, and the sample efficiency residual between the sample physical predicted efficiency and the sample actual efficiency in the historical charging management logs is calculated.

[0044] Based on historical charging management logs, sample load features and sample loss features of power supply on / off switches are extracted respectively.

[0045] Using sample load features and sample loss features as inputs, and sample efficiency residuals as supervision, an efficiency correction model based on regression learning is constructed and trained.

[0046] The physical efficiency model and the efficiency correction model are correlated and connected in series to form an efficiency prediction model;

[0047] The physical efficiency model includes at least the conduction loss component, the switching loss component, the core loss component, and the fixed loss component.

[0048] First, by combining the intrinsic state parameters of the target charging host with a priori knowledge base, a physical efficiency model corresponding to each power supply switch is constructed and calibrated. The intrinsic state parameters include the number of power modules, the rated power value of each power module, and the current on / off state of each power supply switch. The priori knowledge base stores the physical parameters and loss characteristics of the core components in the power modules, providing known device-level parameter inputs for calculating each loss component in the physical efficiency model. Optionally, this priori knowledge base is constructed based on the datasheet data of the components used in the power modules and measured loss data of similar equipment under standard operating conditions, such as the on-resistance value of the power switch, switching time parameters, the Steinmetz coefficient of the core material, and the forward voltage drop of the diode. By integrating the nominal values ​​in the device datasheets with the measured correction values ​​under standard laboratory operating conditions, a set of prior parameters covering different temperatures and current levels is formed, enabling the physical efficiency model to provide reasonable basic efficiency predictions even when historical operating data is lacking.

[0049] Specifically, the physical efficiency model is constructed based on the loss decomposition principle of power modules, and includes at least the following components: conduction loss, switching loss, core loss, and fixed loss. The conduction loss component refers to the Joule heat loss generated when current flows through the power switch and diode, and is proportional to the square of the output current. The switching loss component refers to the loss generated by the power switch during turn-on and turn-off, and is related to the input voltage, output current, and switching frequency. The core loss component refers to the loss generated by the transformer or inductor core in an alternating magnetic field, and is related to the switching frequency and flux density swing. The fixed loss components include gate drive power consumption, control circuit power consumption, and cooling fan power consumption, and are independent of the load size.

[0050] By using intrinsic state parameters and device parameters from the prior knowledge base, the parameters of each loss component in the physical efficiency model can be calibrated, enabling it to output a basic efficiency prediction value based on the current hardware configuration of the charging host.

[0051] Secondly, obtain the historical charging management logs of the target charging host. These logs record detailed data from the past 30 or 90 days of operation, including the start and end times of each charging task, charging power, output current and voltage of each power module, module temperature, ambient temperature, and actual measured system efficiency. These historical charging management logs serve as a sample source for training the efficiency correction model. By extracting operational records under different operating conditions, a sample dataset covering various load rates, temperature conditions, and aging stages can be constructed to learn the mapping relationship between the physical efficiency model's prediction bias and operating parameters.

[0052] Subsequently, based on historical charging management logs, a physical efficiency model is introduced to calculate the sample physical predicted efficiency, and the sample efficiency residual between the sample physical predicted efficiency and the sample actual efficiency in the historical charging management logs is calculated. Specifically, for each historical operation record in the historical charging management logs, its operating parameters, such as output power, input voltage, output voltage, and module temperature, are input into the physical efficiency model to obtain the physical predicted efficiency under that operating condition. Simultaneously, the actual measured efficiency under that operating condition is read from the historical operation record as the sample actual efficiency. The sample efficiency residual equals the sample actual efficiency minus the sample physical predicted efficiency, reflecting the prediction bias of the physical efficiency model under that operating condition. This sample efficiency residual originates from the simplification assumptions of the physical model, unmodeled parasitic effects, individual device differences, and aging factors.

[0053] Furthermore, based on historical charging management logs, sample load characteristics and sample loss characteristics of the power supply on / off switches were extracted. The sample load characteristics include parameters closely related to the current operating conditions, such as module load rate, input voltage, output voltage, module temperature, and ambient temperature. These parameters directly affect the power module's losses. The sample loss characteristics include cumulative operating time and relay switching count. Cumulative operating time reflects the aging degree of the power module, and the relay switching count reflects the intensity of relay use. These parameters affect the degradation of device performance and changes in loss characteristics.

[0054] Furthermore, using sample load features and sample loss features as inputs and sample efficiency residuals as supervision, an efficiency correction model based on regression learning is constructed and trained.

[0055] For example, taking Gaussian process regression as an example, the specific architecture of the efficiency correction model is as follows: The input layer receives a seven-dimensional input feature formed by concatenating a five-dimensional load feature vector and a two-dimensional loss feature vector. The five-dimensional load feature includes module load rate, input voltage, output voltage, module temperature, and ambient temperature. The two-dimensional loss feature includes cumulative running time and relay switching count. The Gaussian process regression model assumes that the efficiency residuals of each sample point follow a joint Gaussian distribution, and uses a kernel function to measure the similarity between different input feature vectors. The kernel function adopts a combination of radial basis function and white noise kernel. The radial basis function kernel is used to capture the smooth nonlinear relationship between the input features and the efficiency residuals, and its length scale parameter is initially set to 1.0, and the signal variance is initially set to 0.1. The white noise kernel is used to fit the measurement noise, and its noise variance is initially set to 0.01.

[0056] During model training, the seven-dimensional input features from historical samples are used as training input, and the corresponding sample efficiency residuals are used as training objectives. The hyperparameters in the kernel function, including length scale, signal variance, and noise variance, are optimized by maximizing the logarithmic marginal likelihood function. After training, for new input feature vectors, the Gaussian process regression model outputs the mean of the prediction residuals as an efficiency correction value, and simultaneously outputs the prediction variance as a measure of the uncertainty of this correction value. Incremental training is triggered every 100 newly accumulated valid samples, re-optimizing the kernel function hyperparameters to enable the model to continuously adapt to device aging and environmental changes.

[0057] Through the construction and training of the Gaussian process regression model, the efficiency correction model can accurately learn the nonlinear mapping relationship between the prediction bias of the physical efficiency model and the operating parameters, providing accurate residual compensation for the physical prediction efficiency.

[0058] Finally, the physical efficiency model and the efficiency correction model are linked and connected in series to output the efficiency prediction model. Specifically, the prediction process of the efficiency prediction model is as follows: First, based on the load distribution results under the current on / off combination, the operating parameters of each module are extracted and input into the physical efficiency model to obtain the physical predicted efficiency; second, the load characteristics and loss characteristics under the current operating conditions are extracted and input into the efficiency correction model to obtain the prediction residual; finally, the physical predicted efficiency and the prediction residual are added together to obtain the final apparent system efficiency prediction value.

[0059] Through the above hybrid modeling architecture, the physical model provides the reliability and interpretability of basic efficiency predictions, while the data correction model compensates for non-ideal factors not considered by the physical model. The combination of the two can achieve high-precision predictions with limited training data.

[0060] Apparent system efficiency refers to the overall system efficiency of the charging host under a relative load balancing strategy. The relative load balancing strategy assumes that all online power modules distribute the total output power according to the principle of equal load rates. Specifically, for power modules of the same model with the same rated power, this strategy is equivalent to each module sharing the total power equally; for power modules of different models with different rated power, this strategy is equivalent to each module sharing the total power according to its rated power ratio, thus keeping the load rate of each module consistent. The reason for using a relative load balancing strategy for efficiency prediction is that in subsequent actual control, the system will execute load distribution according to this strategy, therefore the predicted result is consistent with the system efficiency under actual operating conditions.

[0061] The number of online modules corresponding to each on / off combination, the rated power value of each power module, and the current charging demand power parameter are input into the efficiency prediction model. The model outputs the apparent system efficiency under that on / off combination. After predicting all candidate on / off combinations one by one, a set of apparent system efficiency values ​​corresponding to each on / off combination is obtained, providing a quantitative basis for the selection of the optimal combination.

[0062] Specifically, based on a pre-built efficiency prediction model, the apparent system efficiency of the target charging host under each on / off combination is predicted, where the apparent system efficiency is the efficiency under the relative load balancing strategy, including:

[0063] Based on the relative load balancing strategy, load allocation calculations are performed for each on / off combination, taking into account the intrinsic state parameters, to obtain the load allocation results.

[0064] Combine the load allocation results, match and input them into the corresponding physical efficiency model in the efficiency prediction model to obtain the physical prediction efficiency;

[0065] Based on the load distribution results and intrinsic state parameters, the load characteristics and loss characteristics of each power supply switch are extracted and input into the corresponding efficiency correction model in the efficiency prediction model to obtain the efficiency residual.

[0066] The efficiency residuals are superimposed on the physical predicted efficiency to obtain the apparent switching efficiency of each power supply switch.

[0067] Based on the mapping relationship between apparent switching efficiency and on / off combinations, the apparent switching efficiency is systematically combined and integrated to obtain multiple apparent system efficiencies.

[0068] First, based on the relative load rate balancing strategy, load allocation calculations are performed for each on / off combination, taking into account intrinsic state parameters, to obtain the load allocation results. The relative load rate balancing strategy requires that the load rates of each online power module be equal. Specifically, let the total charging demand power be Q, and the number of online modules be n. For modules of the same model, the rated power of each module is the same, denoted as P, and the power allocated to each module is Q divided by n. For modules of different models, the rated power of each module is different, and the rated power of the i-th module is denoted as Pi. i Let ΣP be the sum of the rated power of all online modules. Then the power allocated to the i-th module is equal to Q multiplied by P. i Divide by the sum of the rated power of all online modules, i.e., P i Divide by ΣP. The load allocation result records the output power value that each online module should bear.

[0069] Secondly, based on the load allocation results, the data is matched and input into the corresponding physical efficiency model in the efficiency prediction model to obtain the physical predicted efficiency. For each online module, its allocated power value, as well as current operating parameters such as input voltage, output voltage, and module temperature, are input into the physical efficiency model corresponding to that module. The physical efficiency model calculates the basic efficiency prediction value of the module under the current operating conditions based on components such as conduction loss, switching loss, core loss, and fixed loss, and outputs the physical predicted efficiency.

[0070] Furthermore, based on the load allocation results and intrinsic state parameters, the load characteristics and loss characteristics of each power supply switch are extracted and input into the corresponding efficiency correction model in the efficiency prediction model to obtain the efficiency residual. Specifically, the load characteristics include module load rate, input voltage, output voltage, module temperature, and ambient temperature, where the module load rate equals the power allocated to the module divided by the module's rated power. The loss characteristics include the module's cumulative operating time and the number of relay switching operations. These feature vectors are input into the efficiency correction model, and the model outputs the predicted residual between the physical predicted efficiency and the actual efficiency of the module under the current operating conditions.

[0071] Furthermore, the efficiency residuals are superimposed on the physical predicted efficiency to obtain the apparent switching efficiency of each power supply switch. The apparent switching efficiency equals the physical predicted efficiency plus the prediction residuals output by the efficiency correction model. This apparent switching efficiency reflects a more accurate efficiency prediction of the power module under current operating conditions after residual correction, compensating for prediction biases caused by factors such as simplification assumptions, unmodeled parasitic effects, and device aging in the physical model.

[0072] Finally, based on the mapping relationship between apparent switching efficiency and on / off combinations, the apparent switching efficiencies are integrated into a system to obtain multiple apparent system efficiencies. System integration refers to merging the apparent switching efficiencies of all online modules under the same on / off combination into the overall system efficiency using a power-weighted method.

[0073] Specifically, the total system output power equals the sum of the output power of each online module. The total system input power equals the sum of the output power of each online module divided by its corresponding apparent switching efficiency; that is, the input power of each module equals its output power divided by its apparent switching efficiency. The apparent system efficiency equals the total system output power divided by the total system input power. This calculation method is applicable to both cases where modules of the same model share power equally and cases where modules of different models distribute power according to their rated power ratio, and it has universality. For the special case of modules of the same model sharing power equally, since the output power of each module is equal and the apparent switching efficiency is the same, the apparent system efficiency equals the apparent switching efficiency of a single module. Through the above system combination integration, each on / off combination corresponds to an apparent system efficiency, which is used for the selection of the optimal combination in the subsequent process.

[0074] The on / off control module 13 is used to select the on / off combination corresponding to the highest apparent system efficiency as the target on / off combination, and to control the on / off status of multiple power supply on / off switches in the target charging host in response to the target on / off combination.

[0075] Furthermore, after obtaining the apparent system efficiency for each on / off combination, the optimal one needs to be selected as the target on / off combination. The selection rule is as follows: compare the apparent system efficiency values ​​of all feasible on / off combinations, and determine the on / off combination with the highest efficiency as the target on / off combination. This target on / off combination represents the power module configuration scheme that enables the charging host system to achieve optimal efficiency under the current charging power demand.

[0076] However, in actual operation, switching between on / off combinations cannot solely consider optimal energy efficiency; system stability and equipment lifespan must also be taken into account. Switching every time the efficiency difference is minimal will lead to frequent relay operation, shortening its mechanical lifespan. Furthermore, the switching process may generate transient disturbances in output voltage and current, affecting charging quality. Therefore, after selecting the target on / off combination, stability conditions must be assessed.

[0077] The stability condition judgment includes two aspects. First, energy efficiency improvement judgment: calculate the difference between the highest apparent system efficiency and the current system efficiency corresponding to the current on / off combination, and determine whether this difference exceeds a preset efficiency improvement threshold. If it does not exceed the threshold, it indicates that the energy efficiency benefit brought by the switch is limited and it is not worthwhile to perform the switch. Second, minimum dwell time judgment: determine whether the time interval from the last on / off state control execution time to the current time exceeds a preset minimum dwell time. If it does not exceed the threshold, it indicates that the system is still within the stable period after the last switch, and frequent switches should be avoided. Only when both of the above conditions are met will the on / off combination corresponding to the highest efficiency be selected as the target on / off combination and the switch be performed.

[0078] Furthermore, before determining stability conditions, a relay lifespan penalty mechanism needs to be introduced to avoid increasing maintenance costs due to excessive use of relays nearing the end of their lifespan. Specifically, the relay lifespan penalty mechanism is a correction method that applies an efficiency discount to on / off combinations involving high-risk relays during energy efficiency assessment. Since the remaining lifespan of each power supply switch gradually shortens with the cumulative number of switching operations, and relay failure at the end of its lifespan will prevent the corresponding power module from switching normally, affecting the availability of the charging host, a penalty correction needs to be applied to the apparent system efficiency when selecting target on / off combinations, prioritizing combinations with similar efficiency but using low-risk relays. Through this lifespan penalty mechanism, the system achieves a balance between optimal energy efficiency and relay lifespan.

[0079] Specifically, based on the efficiency of multiple apparent systems, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination, which also includes:

[0080] Obtain real-time loss characteristics of multiple power supply on / off switches;

[0081] Based on the real-time loss characteristics and the preset nonlinear mapping function, the life penalty coefficient of each power supply switching switch is calculated.

[0082] Based on the lifetime penalty coefficient, lifetime penalties are applied to the efficiency of multiple apparent systems to obtain the apparent corrected system efficiency.

[0083] Based on multiple apparent corrections of system efficiency, the on / off combination corresponding to the highest efficiency value is selected and updated as the target on / off combination.

[0084] First, obtain the real-time loss characteristics of multiple power supply switching switches. Real-time loss characteristics are quantitative indicators reflecting the current health status and remaining lifespan of each relay, and include at least the cumulative number of switching operations, average load current, ambient temperature, and cumulative energizing time for each power supply switching switch. The cumulative number of switching operations directly affects the mechanical wear of the relay contacts, the average load current affects the electrical wear of the contacts, the ambient temperature affects the oxidation rate of the contacts, and the cumulative energizing time reflects the overall aging degree of the relay.

[0085] Secondly, based on the real-time loss characteristics and a preset nonlinear mapping function, the life penalty coefficient for each power supply switching switch is calculated. This preset nonlinear mapping function maps the ratio of the cumulative number of relay switching operations to the rated mechanical life to the penalty coefficient.

[0086] The calculation of the life penalty coefficient for each power supply switching switch, based on real-time loss characteristics and a preset nonlinear mapping function, also includes:

[0087] Based on the confidence failure probability corresponding to different feature terms in the real-time loss characteristics, the first lifetime penalty factor and the second lifetime penalty factor are determined.

[0088] The first coefficient term and the second coefficient term corresponding to the first lifetime penalty factor and the second lifetime penalty factor are calculated using a nonlinear mapping function.

[0089] Historical fault data for each power supply switch is obtained, and the lifetime contribution of different feature items in the real-time loss characteristics is determined by combining statistical analysis methods.

[0090] Using the normalized lifespan contribution as the weight, the first coefficient term and the second coefficient term are weighted and summed to obtain the lifespan penalty coefficient.

[0091] First, based on the confidence failure probabilities corresponding to different feature items in the real-time loss characteristics, the first lifetime penalty factor and the second lifetime penalty factor are determined. Specifically, the real-time loss characteristics include multiple feature items, such as cumulative switching counts, cumulative power-on time, average load current, and ambient temperature. Each feature item can be used to obtain its corresponding confidence failure probability by querying historical fault statistics or accelerated life test data based on its current value. The confidence failure probability represents the estimated probability of the relay failing within a certain period of time under the current value of that feature item, and its value ranges from 0 to 1.

[0092] The above characteristics are divided into two categories. The first category reflects the mechanical lifespan of the relay, including the cumulative number of switching operations and the cumulative energizing time. These characteristics are directly related to the mechanical wear of the relay contacts. The second category reflects the electrical and environmental stresses experienced by the relay, including the average load current and ambient temperature. These characteristics are related to the electrical wear and oxidation rate of the relay contacts.

[0093] Specifically, the first life penalty factor is the result of combining all the confidence failure probabilities in the first type of feature terms, representing the degree of failure risk of the relay in terms of mechanical life. For example, the confidence failure probability corresponding to the cumulative number of switching times is 0.3, and the confidence failure probability corresponding to the cumulative energization time is 0.1. The maximum value of the two, 0.3, is taken as the first life penalty factor, representing the dominant risk in terms of mechanical wear.

[0094] The second lifetime penalty factor is the result of combining all the confidence failure probabilities in the second type of characteristic terms, representing the degree of failure risk of the relay in terms of electrical overstress. For example, the confidence failure probability corresponding to the average load current is 0.2, and the confidence failure probability corresponding to the ambient temperature is 0.1. The maximum value of the two, 0.2, is taken as the second lifetime penalty factor, representing the dominant risk in terms of electrical stress.

[0095] By compressing multiple feature terms into two penalty factors, information from both mechanical and electrical life dimensions is preserved, and subsequent calculations are simplified, avoiding the problem of difficulty in determining weights when directly weighting and summing multiple feature terms.

[0096] Secondly, the first and second coefficient terms corresponding to the first and second lifetime penalty factors are calculated using a nonlinear mapping function. The nonlinear mapping function converts the lifetime penalty factors into penalty coefficient terms, ensuring that high failure risk corresponds to low coefficient terms and low failure risk corresponds to high coefficient terms. Optionally, this nonlinear mapping function can be a sigmoid function or an exponential decay function.

[0097] For example, an exponential decay form is used, where the coefficient term equals the result of a power operation with the natural constant as the base and a negative lifetime penalty factor as the exponent. When the lifetime penalty factor is 0, the coefficient term is 1; when the lifetime penalty factor is 0.3, the coefficient term is approximately 0.74; when the lifetime penalty factor is 0.6, the coefficient term is approximately 0.55; and when the lifetime penalty factor is 1, the coefficient term is approximately 0.37. The mathematical form of this mapping relationship is determined by fitting relay accelerated life test data, ensuring that the value of the coefficient term at different failure risk levels conforms to the actual nonlinear relationship between the remaining life of the relay and the failure probability. Through this nonlinear mapping, a higher lifetime penalty factor indicates a higher failure risk, and a lower coefficient term indicates a stronger penalty.

[0098] Furthermore, historical fault data for each power supply switch is acquired, and statistical analysis methods are used to determine the lifespan contribution of different characteristic items in the real-time loss characteristics. Historical fault data records the fault occurrence time of relays of the same model during past use, as well as the numerical records of each loss characteristic item before the fault. For example, within a continuous one-year operation monitoring cycle, the cumulative number of switching operations, average load current, ambient temperature, and cumulative energization time for each relay fault are recorded. Statistical analysis methods can employ principal component analysis, random forest feature importance assessment, or correlation coefficient methods. By analyzing the correlation strength between each characteristic item and the relay fault time, the contribution weight of each characteristic item to the relay lifespan is calculated.

[0099] Specifically, statistical analysis methods refer to methods that quantify the impact of various loss characteristics on relay lifespan from historical fault data using mathematical calculations. Taking the correlation coefficient method as an example: First, fault records of relays of the same model during historical operation are collected. Each record includes characteristic values ​​such as the cumulative number of switching operations before the relay fault, average load current, ambient temperature, and cumulative energization time, as well as the actual service life of the relay. Then, the Pearson correlation coefficient between each characteristic and its service life is calculated. The larger the absolute value of the correlation coefficient, the higher the linear correlation between the characteristic and its service life, and the more significant its impact on lifespan. Finally, the absolute value of the correlation coefficient for each characteristic is divided by the sum of the absolute values ​​of the correlation coefficients for all characteristics to obtain the normalized lifespan contribution. Through statistical analysis, the influence weight of each loss characteristic in the relay's lifespan can be objectively and quantitatively determined. The larger the lifespan contribution value, the more significant the impact of that characteristic on the relay's lifespan, and the higher its weight should be assigned in the subsequent calculation of the lifespan penalty coefficient.

[0100] Furthermore, using the normalized lifetime contribution as the weight, a weighted sum is performed on the first and second coefficient terms to obtain the lifetime penalty coefficient. Specifically, the lifetime contribution of each feature term is divided by the sum of the lifetime contributions of all feature terms to obtain the normalized weight. The first coefficient term corresponds to the comprehensive mapping result of the first type of feature terms, and the second coefficient term corresponds to the comprehensive mapping result of the second type of feature terms. The lifetime penalty coefficient is equal to the sum of the first coefficient term multiplied by the normalized weights of the first type of feature terms plus the sum of the second coefficient term multiplied by the normalized weights of the second type of feature terms. Through the above weighted summation, the lifetime penalty coefficient can comprehensively reflect the combined impact of each loss feature term on the remaining lifetime of the relay, making the lifetime penalty more accurate and comprehensive.

[0101] Furthermore, based on the lifetime penalty coefficient, lifetime penalties are applied to multiple apparent system efficiencies to obtain the apparent corrected system efficiency. Specifically, the lifetime penalty is implemented as follows: for each on / off combination, the set of relays that need to be activated relative to the current on / off combination is identified, i.e., relays that need to be turned on and relays that need to be turned off. The minimum lifetime penalty coefficient of all relays in this activation set is taken as the combination penalty coefficient for that on / off combination. The apparent corrected system efficiency is equal to the apparent system efficiency multiplied by the combination penalty coefficient. If an on / off combination requires the use of a relay that is nearing the end of its lifespan, its combination penalty coefficient is small, and the apparent corrected system efficiency is significantly reduced.

[0102] Furthermore, based on multiple apparent system efficiencies, the on / off combination corresponding to the highest efficiency value is selected and updated as the target on / off combination. Through the aforementioned lifetime penalty correction, the on / off combination with the highest original apparent system efficiency but requiring the use of high-risk relays may have a lower efficiency after correction than other combinations; while the combination with the second highest original apparent system efficiency but using low-risk relays may outperform. This mechanism enables the system to proactively avoid relays nearing the end of their lifespan while pursuing optimal energy efficiency, thereby extending the overall system lifespan and reducing maintenance costs.

[0103] Furthermore, based on the efficiency of multiple apparent systems, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination, including:

[0104] Stability conditions are determined sequentially, including:

[0105] Determine whether the difference between the apparent system efficiency corresponding to the highest efficiency and the current system efficiency under the current on / off combination exceeds a preset efficiency improvement threshold;

[0106] If it exceeds, then determine whether the time interval from the last on / off state control to the current time exceeds the preset minimum dwell time;

[0107] If all stability conditions are met, the on / off combination corresponding to the highest value is selected as the target on / off combination.

[0108] First, determine whether the difference between the apparent system efficiency corresponding to the highest efficiency and the current system efficiency under the current on / off combination exceeds a preset efficiency improvement threshold. The current on / off combination refers to the power module configuration currently running on the charging host, and its corresponding current system efficiency can be calculated in real-time by an efficiency prediction model or obtained from historical operating data. The efficiency improvement threshold is a pre-set minimum energy efficiency gain threshold, for example, 0.5%. The purpose of setting this threshold is to avoid frequent relay operation due to small efficiency differences, which would shorten mechanical lifespan, and to prevent transient disturbances during switching from causing unnecessary impacts on charging quality.

[0109] If the efficiency difference does not exceed the efficiency improvement threshold, it indicates that the energy efficiency benefits brought by the switch are limited and it is not worthwhile to perform the switch. In this case, the current on / off combination should be maintained.

[0110] Furthermore, if the efficiency difference exceeds a preset efficiency improvement threshold, it is further determined whether the time interval from the last on / off state control execution time to the current time exceeds a preset minimum dwell time. The minimum dwell time refers to the shortest stable operating time the system must maintain under the current configuration after each switch; for example, it can be set to 30 seconds or 60 seconds. The purpose of setting the minimum dwell time is to ensure that the system has sufficient time to reach thermal and electrical steady state after each switch, avoiding excessive wear on relays and power modules caused by repeated switching in a short period, while ensuring the stability of charging output. If the time interval does not exceed the minimum dwell time, it indicates that the system is still within the stable period after the last switch, and further switching should be avoided.

[0111] If both of the above stability conditions are met—that is, the efficiency difference exceeds the efficiency improvement threshold and the time interval exceeds the minimum residence time—then the on / off combination corresponding to the highest one is selected as the target on / off combination. Specifically, through the above dual stability constraint mechanism, the system can balance the relationship between relay lifespan, system stability, and energy efficiency gains while pursuing optimal energy efficiency.

[0112] Furthermore, before formally implementing on / off state control, the power allocation in the target on / off combination needs to be finely optimized. In the aforementioned steps, the calculation of apparent system efficiency is based on the assumption of a relative load rate balancing strategy, that is, the total output power is allocated to each online power module according to the principle of equal load rate or proportional rated power. This relative load rate balancing strategy is simple to calculate and applicable to most operating scenarios, but it cannot make the system efficiency reach the theoretical maximum value under all operating conditions. Due to differences in aging degree, uneven temperature distribution, or individual device deviations among the power modules in actual operation, their efficiency curves show slight differences. In this case, appropriately adjusting the power allocation scheme so that the more efficient modules bear more load and the less efficient modules bear less load may result in a system efficiency better than a balanced allocation.

[0113] Therefore, when the consistency of multiple power supply switching switches in the target on / off combination is inconsistent, that is, when the rated power of each power module is different or the aging degree is significantly different, this step additionally performs optimization solution.

[0114] Specifically, in response to the target on / off combination, the on / off state control of multiple power supply on / off switches in the target charging host is performed, prior to which the following is also included:

[0115] If the consistency of multiple power supply switching switches in the target on / off combination is inconsistent, then the optimization problem is defined by combining the efficiency prediction model, taking the highest apparent system efficiency as the optimization objective, and the power allocation value of each power module as the independent variable.

[0116] Based on the intrinsic state parameters, a set of boundary constraints corresponding to the optimization problem is defined. The set of boundary constraints includes at least the power capacity constraint, the minimum number of modules constraint, the maximum number of modules constraint, and the rated power constraint of each power module.

[0117] Based on the boundary constraint set, an iterative numerical optimization method is used to solve the optimization problem and obtain the optimal power allocation value for each power module.

[0118] The system retains the highest apparent system efficiency among the optimal power allocation value and the relative load allocation result under the load balancing strategy, and updates the target on / off combination accordingly.

[0119] First, if the consistency of multiple power supply switches in the target on / off combination is inconsistent, then, combining the efficiency prediction model, the optimization problem is defined with the highest apparent system efficiency as the objective and the power allocation value of each power module as the independent variable. Specifically, when the power modules in the target on / off combination are not the same model, or when the aging levels of the modules differ, the efficiency characteristics of each module are different. In this case, the scheme of allocating power according to the rated power ratio under the relative load balancing strategy may not necessarily maximize system efficiency, because modules with better efficiency characteristics should bear more power, and modules with poorer efficiency characteristics should bear less power, in order to optimize the overall system efficiency. Therefore, it is necessary to use the power allocation value of each power module as a decision variable, and define the optimization problem with the goal of maximizing the overall system efficiency.

[0120] Optionally, the optimization problem is defined as follows: Suppose there are n online power modules in the target on / off combination, and the power allocation value of the i-th module is x. i The unit is kilowatts. The overall system efficiency η is equal to the sum of the output power of all modules divided by the sum of the input power of all modules, i.e., η = Σx i / Σ(x i / η i (x i)), where η i (x i Let ) represent the i-th module with an output power of x. i The efficiency is predicted by an efficiency prediction model based on parameters such as the module's power value, input voltage, output voltage, module temperature, cumulative operating time, and relay switching count. The optimization objective is to maximize η. The decision variable is the power allocation value x for each module. i The unit is kilowatt, and it is a continuous variable.

[0121] By solving this optimization problem, we can obtain the power allocation combination of each module that maximizes the overall efficiency of the system. This allows modules with better efficiency characteristics to bear a larger proportion of the power, while modules with poorer efficiency characteristics bear a smaller proportion of the power, thus achieving the optimal configuration of system efficiency while meeting the total power requirements.

[0122] Secondly, based on the intrinsic state parameters, a set of boundary constraints corresponding to the optimization problem is defined. This set includes at least power capacity constraints, minimum module number constraints, maximum module number constraints, and the rated power constraint for each power module. The power capacity constraint requires that the sum of the output power of all online modules equals the total charging demand power Q. The minimum and maximum module number constraints limit the range of values ​​for the number of online modules; the minimum module number is determined by both reliability and power capacity, while the maximum module number is determined by the total number of modules in the charging host. The rated power constraint for each online module requires that its output power does not exceed its rated power value and is not lower than zero.

[0123] Furthermore, based on the boundary constraint set, an iterative numerical optimization method is employed to solve the optimization problem and obtain the optimal power allocation values ​​for each power module. Optionally, the iterative numerical optimization method can employ algorithms such as gradient descent, interior-point method, or sequential quadratic programming. Using the power allocation values ​​of each module as initial values, and under the premise of satisfying the boundary constraints, the system iteratively searches for the power allocation combination that maximizes the overall system efficiency. An efficiency prediction model is used in each iteration to evaluate the system efficiency under the current power allocation scheme and guide the adjustment of the optimization direction.

[0124] For example, taking the sequential quadratic programming algorithm as an example, the process of solving the optimization problem is as follows:

[0125] First, set the initial solution. The load allocation result under the relative load balancing strategy is used as the initial power allocation value, that is, the initial power allocation value of the i-th module is equal to the total charging demand power Q multiplied by the rated power P of the i-th module. iDivide by the sum of the rated power of all online modules. This initial solution satisfies all constraints in the boundary constraint set. Next, in each iteration, calculate the objective function value and constraint function value at the current solution. The objective function value is calculated using an efficiency prediction model: for each online module, its power allocation value, along with current input voltage, output voltage, module temperature, cumulative running time, relay switching count, and other parameters, are input into the efficiency prediction model to obtain the module's predicted efficiency value. Then, the overall system efficiency is calculated, which equals the sum of the output power of all modules divided by the sum of the input power of all modules, where the input power of each module is equal to its output power divided by its predicted efficiency value.

[0126] Furthermore, a quadratic programming subproblem is constructed. At the current solution, a second-order Taylor expansion of the objective function yields a quadratic approximation function; a first-order Taylor expansion of the constraint functions yields a linear approximation function. This forms the basis of a quadratic programming subproblem, where the objective function is a quadratic form and the constraints are linear. Solving this subproblem yields the search direction. By solving this subproblem, the direction of variable increments that minimizes the approximate objective function is obtained.

[0127] Further, a line search is performed to determine the step size. Starting from the current solution, the search seeks a feasible step size that improves the objective function value while satisfying the constraints by gradually decreasing the step size. Common line search methods include backtracking line search, which starts with a step size of 1 and multiplies by a decay factor less than 1, such as 0.5, each time until a step size satisfying the constraints is found. The updated solution is the current solution plus the step size multiplied by the search direction. Convergence is determined by calculating the relative change in the objective function value between two adjacent iterations, i.e., the absolute value of the difference between the current and previous objective function values ​​divided by the previous objective function value. When this relative change is less than a preset convergence threshold, such as 1e-4, or when the number of iterations reaches a preset maximum number of iterations, such as 100, the iteration terminates, and the current solution is output as the optimal power allocation value; otherwise, the next iteration continues.

[0128] Finally, the system retains the highest apparent system efficiency between the optimal power allocation value and the load allocation result under the relative load balancing strategy, and updates the target on / off combination accordingly. Specifically, the apparent system efficiency under the optimal power allocation scheme obtained from the optimization solution is compared with the apparent system efficiency under the relative load balancing strategy, and the one with higher efficiency is selected as the final power allocation scheme. If the optimization solution is better, the output power command of each module is updated according to the optimized power allocation value, and the target on / off combination remains unchanged; if the load balancing strategy result is better, the original load allocation result is maintained. Through the above optimization steps, in the case of different module models or inconsistent module aging, the system can further explore the space for energy efficiency improvement, making the power allocation of each module more reasonable and achieving optimal overall efficiency.

[0129] Specifically, the above steps aim to further explore the energy efficiency optimization potential at the power allocation level, given a fixed on / off combination, to maximize system efficiency with existing hardware configurations. For modules of the same model and similar aging levels, the efficiency curves of each module are essentially the same, and the balanced allocation is already close to optimal; therefore, the benefits of this optimization step are limited and can be skipped. This step, as an optional fine-tuning optimization link between steps S20 and S30, does not affect the main control flow, but can bring additional energy efficiency improvements in scenarios with different module models or significant differences in module aging.

[0130] Finally, after determining the target on / off combination, the on / off status control of multiple power supply on / off switches in the target charging host is performed in response to the target on / off combination.

[0131] Specifically, in response to the target on / off combination, the on / off state control of multiple power supply on / off switches in the target charging host includes:

[0132] A set of modules that identify the differences between the current on / off combination and the target on / off combination;

[0133] For newly added power modules that need to be connected in the difference module cluster, first turn on the corresponding power supply switch, and use the online power modules as the power transfer source to control the output of the newly added power modules to be increased to match the load distribution result in the target on / off combination at a preset first power transfer rate.

[0134] For power modules to be removed from the differential module set, the online power modules are used as the power receiving source. The output of the power modules to be removed is controlled to be reduced to the same as the load distribution result in the target on / off combination by a preset second power transfer rate.

[0135] After the power transfer is completed and the total output parameters of the target charging host are verified to meet the preset steady-state discrimination conditions, the power supply switch corresponding to the power module to be de-powered is disconnected.

[0136] Record the timestamp of power transfer completion and update the cumulative switching count counter for each power supply switch.

[0137] First, identify the difference set of modules between the current on / off combination and the target on / off combination. The current on / off combination refers to the power module configuration currently running on the charging host, while the target on / off combination is the optimal configuration determined after the aforementioned energy efficiency assessment, lifespan penalty, and stability judgment. The difference set of modules includes two categories: one is newly added power modules that need to be connected, i.e., currently offline but required to be online in the target on / off combination; the other is power modules that need to be removed, i.e., currently online but required to be offline in the target on / off combination.

[0138] Secondly, for newly added power modules requiring connection within the differential module cluster, a soft-start process of connection followed by loading is executed. First, the power supply switch corresponding to the new power module is closed, energizing the module and putting it into standby mode, at which point its output power is zero. Then, using the currently online power modules as the power transfer source, the output power of the new module is gradually increased according to a preset first power transfer rate. This first power transfer rate can be set, for example, to 10% of the rated power per second. During the power increase process, the output power of the already online modules decreases synchronously and equally, ensuring that the sum of the output power of all online modules always equals the total charging demand. When the output power of the new module rises to the power value allocated to it in the target on / off combination, the power loading process for the new module is complete.

[0139] Simultaneously, for power modules that need to be removed from the differential module set, a soft exit process of unloading first and then disconnecting is executed. Using the power module that will remain online after the switchover as the power receiving source, the output power of the module to be removed is gradually reduced according to a preset second power transfer rate. This second power transfer rate can be set, for example, to 10% of the rated power per second. During the power reduction process, the output power of the module to be removed gradually decreases, while the output power of the receiving source module increases synchronously and equally, ensuring that the sum of the output power of all online modules always equals the total charging power demand. When the output power of the module to be removed decreases to the power value allocated to that module in the target on / off combination, the power value of the module to be removed in the target on / off combination is zero. At this point, the output power of the module to be removed is zero, and the power unloading process is complete.

[0140] The first power transfer rate and the second power transfer rate represent the rising rate during the power loading phase and the falling rate during the power unloading phase, respectively, indicating the magnitude of power change per unit time. For example, they can be set to 10% of the rated power per second. This power transfer rate is set comprehensively based on the output response capability of the charging host and the load's tolerance to power change rate, balancing the relationship between switching speed and output stability. An excessively high rate can cause transient surges in output voltage and current, potentially triggering load protection mechanisms or causing power outages; an excessively low rate will prolong the switching time, increasing the relay's load-carrying time during power transfer and accelerating contact wear. By setting a reasonable power transfer rate, the impact of the switching process on relay lifespan can be minimized while ensuring output quality.

[0141] After power transfer is complete and the total output parameters of the target charger are verified to meet the preset steady-state discrimination conditions, the power supply switch corresponding to the power module to be de-energized is disconnected. The steady-state discrimination conditions are used to confirm that the system is in a stable operating state after power transfer; for example, output voltage fluctuation can be set to less than ±1%, output current fluctuation to less than ±2%, and a sustained stable time exceeding 0.5 seconds. After the steady-state discrimination conditions are met, the relay of the module to be de-energized is then disconnected to ensure that the module has no current output when disconnected, thus avoiding damage to the relay contacts caused by arcing during load disconnection.

[0142] Finally, the timestamp of power transfer completion is recorded, and the cumulative switching count counter for each power supply switch is updated. The timestamp is used for subsequent minimum dwell time determination, and the cumulative switching count counter is used for life penalty calculation and relay health status monitoring. Each time a relay is switched on or off, its corresponding cumulative switching count is incremented by one. Through the above soft switching strategy, the system achieves smooth power transfer during switching, avoiding drastic fluctuations in output voltage and current, while also protecting the relay contacts and extending the relay's lifespan.

[0143] In summary, the embodiments of this application have at least the following technical effects:

[0144] This invention first enumerates all feasible power supply switching combinations to construct a complete candidate configuration space, avoiding the loss of better energy-efficient solutions due to configuration omissions. Secondly, based on an efficiency prediction model that integrates a physical model and data correction, it predicts the system efficiency under each switching combination, solving the problem that traditional threshold switching cannot predict post-switching efficiency performance and effectively preventing ineffective switching. Thirdly, using the apparent system efficiency under a relative load balancing strategy as the optimization target, it selects the optimal switching combination for control, ensuring that each online module avoids the low-efficiency range under light load and always operates in the high-efficiency load range, improving overall system efficiency by 3% to 8% under light load conditions. Furthermore, all relays can be disconnected when there is no charging demand, achieving zero standby power consumption.

[0145] This invention solves the technical problems of low energy efficiency under light load, high standby power consumption, and coarse switching strategy in the prior art, and meets the first-level energy efficiency requirements of the charging host in the full load range.

[0146] Example 2, as Figure 2 As shown, based on the same inventive concept as the charging control system of the Level 1 energy efficiency charging host provided in Embodiment 1, this embodiment of the invention also provides a charging control method for the Level 1 energy efficiency charging host, including:

[0147] Obtain the power parameters of the current charging demand, and based on the power parameters and the intrinsic state parameters of the target charging host, enumerate all feasible on / off combinations of power supply switches.

[0148] Based on a pre-built efficiency prediction model, the apparent system efficiency of the target charging host under each on / off combination is predicted, where the apparent system efficiency is the efficiency under the relative load balancing strategy.

[0149] Based on the efficiency of multiple apparent systems, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination, and the on / off state control of multiple power supply on / off switches in the target charging host is performed in response to the target on / off combination.

[0150] Specifically, the power parameters of the current charging demand are obtained, and based on the power parameters and the intrinsic state parameters of the target charging host, all feasible on / off combinations of the power supply switch are enumerated, including:

[0151] Obtain the current charging demand and extract the corresponding power parameters;

[0152] Obtain the intrinsic state parameters of the target charging host, wherein the intrinsic state parameters include at least the number of power modules, the rated power value of each power module, and the current on / off state of each power supply switch.

[0153] Based on intrinsic state parameters, determine the consistency of multiple power supply on / off switches;

[0154] If they are consistent, then the number of connected switches is used as the only variable to perform simplified combination enumeration on multiple power supply switches to obtain multiple on / off combinations.

[0155] If they are inconsistent, multiple power supply switches are enumerated in a heterogeneous combination manner based on the combination enumeration method to obtain multiple on / off combinations.

[0156] Furthermore, the construction of the efficiency prediction model includes:

[0157] By combining the intrinsic state parameters of the target charging host with the prior knowledge base, a physical efficiency model corresponding to each power supply on / off switch is constructed and calibrated.

[0158] Obtain the historical charging management logs of the target charging host;

[0159] Based on historical charging management logs, a physical efficiency model is introduced to calculate the sample physical predicted efficiency, and the sample efficiency residual between the sample physical predicted efficiency and the sample actual efficiency in the historical charging management logs is calculated.

[0160] Based on historical charging management logs, sample load features and sample loss features of power supply on / off switches are extracted respectively.

[0161] Using sample load features and sample loss features as inputs, and sample efficiency residuals as supervision, an efficiency correction model based on regression learning is constructed and trained.

[0162] The physical efficiency model and the efficiency correction model are correlated and connected in series to form an efficiency prediction model;

[0163] The physical efficiency model includes at least the conduction loss component, the switching loss component, the core loss component, and the fixed loss component.

[0164] Specifically, based on a pre-built efficiency prediction model, the apparent system efficiency of the target charging host under each on / off combination is predicted, where the apparent system efficiency is the efficiency under the relative load balancing strategy, including:

[0165] Based on the relative load balancing strategy, load allocation calculations are performed for each on / off combination, taking into account the intrinsic state parameters, to obtain the load allocation results.

[0166] Combine the load allocation results, match and input them into the corresponding physical efficiency model in the efficiency prediction model to obtain the physical prediction efficiency;

[0167] Based on the load distribution results and intrinsic state parameters, the load characteristics and loss characteristics of each power supply switch are extracted and input into the corresponding efficiency correction model in the efficiency prediction model to obtain the efficiency residual.

[0168] The efficiency residuals are superimposed on the physical predicted efficiency to obtain the apparent switching efficiency of each power supply switch.

[0169] Based on the mapping relationship between apparent switching efficiency and on / off combinations, the apparent switching efficiency is systematically combined and integrated to obtain multiple apparent system efficiencies.

[0170] Furthermore, based on the efficiency of multiple apparent systems, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination, including:

[0171] Specifically, stability conditions are determined sequentially, including:

[0172] Determine whether the difference between the apparent system efficiency corresponding to the highest efficiency and the current system efficiency under the current on / off combination exceeds a preset efficiency improvement threshold;

[0173] If it exceeds, then determine whether the time interval from the last on / off state control to the current time exceeds the preset minimum dwell time;

[0174] If all stability conditions are met, the on / off combination corresponding to the highest value is selected as the target on / off combination.

[0175] In response to the target on / off combination, the on / off state control of multiple power supply on / off switches in the target charging host is performed, prior to which the following is also included:

[0176] If the consistency of multiple power supply switching switches in the target on / off combination is inconsistent, then the optimization problem is defined by combining the efficiency prediction model, taking the highest apparent system efficiency as the optimization objective, and the power allocation value of each power module as the independent variable.

[0177] Based on the intrinsic state parameters, a set of boundary constraints corresponding to the optimization problem is defined. The set of boundary constraints includes at least the power capacity constraint, the minimum number of modules constraint, the maximum number of modules constraint, and the rated power constraint of each power module.

[0178] Based on the boundary constraint set, an iterative numerical optimization method is used to solve the optimization problem and obtain the optimal power allocation value for each power module.

[0179] The system retains the highest apparent system efficiency among the optimal power allocation value and the load allocation result under the relative load rate balancing strategy, and updates the target on / off combination accordingly.

[0180] Specifically, in response to the target on / off combination, the on / off state control of multiple power supply on / off switches in the target charging host includes:

[0181] A set of modules that identify the differences between the current on / off combination and the target on / off combination;

[0182] For newly added power modules that need to be connected in the difference module cluster, first turn on the corresponding power supply switch, and use the online power modules as the power transfer source to control the output of the newly added power modules to be increased to match the load distribution result in the target on / off combination at a preset first power transfer rate.

[0183] For power modules to be removed from the differential module set, the online power modules are used as the power receiving source. The output of the power modules to be removed is controlled to be reduced to the same as the load distribution result in the target on / off combination by a preset second power transfer rate.

[0184] After the power transfer is completed and the total output parameters of the target charging host are verified to meet the preset steady-state discrimination conditions, the power supply switch corresponding to the power module to be de-powered is disconnected.

[0185] Record the timestamp of power transfer completion and update the cumulative switching count counter for each power supply switch.

[0186] Furthermore, based on the efficiencies of multiple apparent systems, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination, which also includes:

[0187] Obtain real-time loss characteristics of multiple power supply on / off switches;

[0188] Based on the real-time loss characteristics and the preset nonlinear mapping function, the life penalty coefficient of each power supply switching switch is calculated.

[0189] Based on the lifetime penalty coefficient, lifetime penalties are applied to the efficiency of multiple apparent systems to obtain the apparent corrected system efficiency.

[0190] Based on multiple apparent corrections of system efficiency, the on / off combination corresponding to the highest efficiency value is selected and updated as the target on / off combination.

[0191] In addition, based on the real-time loss characteristics and a preset nonlinear mapping function, the lifetime penalty coefficient for each power supply switching switch is calculated, which also includes:

[0192] Based on the confidence failure probability corresponding to different feature terms in the real-time loss characteristics, the first lifetime penalty factor and the second lifetime penalty factor are determined.

[0193] The first coefficient term and the second coefficient term corresponding to the first lifetime penalty factor and the second lifetime penalty factor are calculated using a nonlinear mapping function.

[0194] Historical fault data for each power supply switch is obtained, and the lifetime contribution of different feature items in the real-time loss characteristics is determined by combining statistical analysis methods.

[0195] Using the normalized lifespan contribution as the weight, the first coefficient term and the second coefficient term are weighted and summed to obtain the lifespan penalty coefficient.

Claims

1. A charging control system for a Level 1 energy efficiency charging host, characterized in that, include: The combined enumeration module is used to obtain the power parameters of the current charging demand, and based on the power parameters and the intrinsic state parameters of the target charging host, enumerate all feasible on / off combinations of power supply switches. An efficiency prediction module is used to predict the apparent system efficiency of the target charging host under each of the on / off combinations based on a pre-built efficiency prediction model, wherein the apparent system efficiency is the efficiency under the relative load balancing strategy. The on / off control module is used to select the on / off combination corresponding to the highest apparent system efficiency as the target on / off combination, and to control the on / off state of multiple power supply on / off switches in the target charging host in response to the target on / off combination.

2. The charging control system of the first-level energy efficiency charging host as described in claim 1, characterized in that, Obtain the power parameters of the current charging demand, and based on the power parameters and the intrinsic state parameters of the target charging host, enumerate all feasible on / off combinations of power supply switches. The execution steps of the combination enumeration module include: Obtain the current charging demand and extract the corresponding power parameters; The intrinsic state parameters of the target charging host are obtained, wherein the intrinsic state parameters include at least the number of power modules, the rated power value of each power module, and the current on / off state of each power supply switch. Based on the intrinsic state parameters, determine the consistency of the multiple power supply on / off switches; If they are consistent, then the number of connected switches is used as the only variable to perform simplified combination enumeration on multiple power supply on / off switches to obtain multiple on / off combinations; If they are inconsistent, then based on the combination enumeration method, heterogeneous combination enumeration is performed on multiple power supply on / off switches to obtain multiple on / off combinations.

3. The charging control system of the first-level energy efficiency charging host as described in claim 1, characterized in that, The construction of the efficiency prediction model and the execution steps of the efficiency prediction module include: By combining the intrinsic state parameters of the target charging host with the prior knowledge base, a physical efficiency model corresponding to each power supply on / off switch is constructed and calibrated. Obtain the historical charging management logs of the target charging host; Based on the historical charging management logs, the physical efficiency model is introduced to calculate the sample physical predicted efficiency, and the sample efficiency residual between the sample physical predicted efficiency and the sample actual efficiency in the historical charging management logs is calculated. Based on the historical charging management logs, sample load features and sample loss features of the power supply on / off switch are extracted respectively. Using the sample load features and the sample loss features as inputs, and the sample efficiency residuals as supervision, an efficiency correction model based on regression learning is constructed and trained. The physical efficiency model and the efficiency correction model are correlated and connected in series to form the efficiency prediction model; The physical efficiency model includes at least the conduction loss component, the switching loss component, the core loss component, and the fixed loss component.

4. The charging control system of the first-level energy efficiency charging host as described in claim 3, characterized in that, Based on a pre-built efficiency prediction model, the apparent system efficiency of the target charging host is predicted under each of the on / off combinations, wherein the apparent system efficiency is the efficiency under the relative load balancing strategy. The execution steps of the efficiency prediction module further include: Based on the relative load balancing strategy, load allocation calculations are performed for each on / off combination, taking into account the intrinsic state parameters, to obtain the load allocation results. Based on the load allocation results, the data is matched and input into the corresponding physical efficiency model in the efficiency prediction model to obtain the physical prediction efficiency; Based on the load allocation results and the intrinsic state parameters, the load characteristics and loss characteristics of each power supply switch are extracted and input into the corresponding efficiency correction model in the efficiency prediction model to obtain the efficiency residual. The efficiency residual is superimposed on the physical predicted efficiency to obtain the apparent switching efficiency of each power supply switch. Based on the mapping relationship between the apparent switching efficiency and the on / off combination, the apparent switching efficiency is systematically combined and integrated to obtain multiple apparent system efficiencies.

5. The charging control system of the first-level energy efficiency charging host as described in claim 1, characterized in that, Based on the apparent system efficiency of multiple systems, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination. The execution steps of the on / off control module include: Stability conditions are determined sequentially, including: Determine whether the difference between the apparent system efficiency corresponding to the highest one and the current system efficiency under the current on / off combination exceeds a preset efficiency improvement threshold; If it exceeds, then determine whether the time interval from the last on / off state control to the current time exceeds the preset minimum dwell time; If all stability conditions are met, the on / off combination corresponding to the highest value is selected as the target on / off combination.

6. The charging control system of the first-level energy efficiency charging host as described in claim 1, characterized in that, In response to the target on / off combination, the on / off state control of multiple power supply on / off switches in the target charging host is performed. Prior to this, the execution steps of the on / off control module further include: If the consistency of multiple power supply switching switches in the target on / off combination is inconsistent, then, in conjunction with the efficiency prediction model, the optimization problem is defined with the highest apparent system efficiency as the optimization objective and the power allocation value of each power module as the independent variable. Based on the intrinsic state parameters, a set of boundary constraints corresponding to the optimization problem is defined, wherein the set of boundary constraints includes at least power capacity constraints, minimum number of modules constraints, maximum number of modules constraints, and single-module rated power constraints for each power module. Based on the boundary constraint set, the optimization problem is solved using an iterative numerical optimization method to obtain the optimal power allocation value for each power module; The system retains the one with the highest apparent system efficiency between the optimal power allocation value and the load allocation result under the relative load rate balancing strategy, and updates the target on / off combination accordingly.

7. The charging control system of the first-level energy efficiency charging host as described in claim 1, characterized in that, In response to the target on / off combination, the on / off state control of multiple power supply on / off switches in the target charging host is performed, and the execution steps of the on / off control module further include: Identify the difference between the current on / off combination and the target on / off combination; For newly added power modules that need to be connected in the difference module set, first turn on the corresponding power supply switch, and use the online power module as the power transfer source to control the output of the newly added power module to be increased to be consistent with the load distribution result in the target on / off combination at a preset first power transfer rate; For the power modules to be removed from the set of differential modules, the online power modules are used as the power receiving source. The output of the power modules to be removed is controlled to be reduced to the same as the load distribution result in the target on / off combination at a preset second power transfer rate. After the power transfer is completed and the total output parameters of the target charging host are verified to meet the preset steady-state discrimination conditions, the power supply switch corresponding to the power module to be de-powered is disconnected. Record the timestamp of power transfer completion and update the cumulative switching count counter for each power supply on / off switch.

8. The charging control system of the first-level energy efficiency charging host as described in claim 1, characterized in that, Based on the apparent system efficiency of multiple systems, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination. The execution steps of the on / off control module further include: Obtain the real-time loss characteristics of multiple power supply on / off switches; Based on the real-time loss characteristics and the preset nonlinear mapping function, the lifetime penalty coefficient of each power supply switching switch is calculated. Based on the lifetime penalty coefficient, lifetime penalties are applied to the apparent system efficiencies of the plurality of systems to obtain the apparent corrected system efficiency. Based on the apparent correction system efficiency, the on / off combination corresponding to the highest efficiency value is selected and updated as the target on / off combination.

9. The charging control system of the first-level energy efficiency charging host as described in claim 8, characterized in that, Based on the real-time loss characteristics and the preset nonlinear mapping function, the lifetime penalty coefficient of each power supply switching switch is calculated. The execution steps of the switching control module further include: Based on the confidence failure probability corresponding to different feature terms in the real-time loss characteristics, the first lifetime penalty factor and the second lifetime penalty factor are determined. The first coefficient term and the second coefficient term corresponding to the first lifetime penalty factor and the second lifetime penalty factor are calculated using the nonlinear mapping function. Historical fault data for each of the power supply switching switches are obtained, and the lifetime contribution of different feature items in the real-time loss characteristics is determined by statistical analysis methods. Using the normalized lifetime contribution as the weight, the first coefficient term and the second coefficient term are weighted and summed to obtain the lifetime penalty coefficient.

10. A charging control method for a Level 1 energy efficiency charging host, characterized in that, A module for implementing the charging control system of the Level 1 energy efficiency charging host according to any one of claims 1-9, comprising: Obtain the power parameters of the current charging demand, and based on the power parameters and the intrinsic state parameters of the target charging host, enumerate and obtain all feasible on / off combinations of power supply switches. Based on a pre-built efficiency prediction model, the apparent system efficiency of the target charging host under each of the on / off combinations is predicted, wherein the apparent system efficiency is the efficiency under the relative load balancing strategy. Based on the apparent system efficiency, the on / off combination corresponding to the highest efficiency is selected as the target on / off combination, and the on / off state control of the multiple power supply on / off switches in the target charging host is performed in response to the target on / off combination.

Images