Intelligent load balancing method for double-input computing power supply

Abstract

The invention discloses a dual-input computing power supply intelligent load balancing method, and particularly relates to the technical field of fusion positioning, and the method comprises the following steps: collecting the voltage, current, power and load data of two input power supplies, and constructing a mathematical model reflecting the relation between the power supply output and the load; selecting calculation precision configuration according to the current output capability and the load state, and preliminarily calculating a power supply proportion; correcting the precision parameter based on the power supply deviation and carrying out iterative updating, collecting power supply temperature change data in the process of updating the precision parameter, if a temperature risk state is detected, bringing the temperature risk state into the updating process, and adjusting the precision parameter to inhibit performance deviation caused by temperature rise; according to the method, the power supply proportion calculation mathematical model is constructed, the precision parameters are dynamically matched, cycle-by-cycle updating is achieved in combination with load response, the precision parameters are adjusted on the basis of introducing temperature change data, the load adaptability, the calculation stability and the thermal safety are enhanced, and the accuracy and reliability of multi-source power supply control are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fusion positioning technology, more particularly, the present application relates to a dual-input computing power intelligent load balancing method. BACKGROUND

[0002] With the wide deployment of high-performance computing architectures and heterogeneous processing units in data centers, intelligent terminals and edge devices, the dynamic adaptability and load distribution efficiency of power management systems are put forward higher requirements. Dual-input computing power supply architecture has become the mainstream way to meet the needs of high load operation and redundancy backup, in which two input power supplies need to support the stable power supply of computing cores according to their respective capabilities. Therefore, how to reasonably allocate the power supply ratio according to the power supply performance, current load demand and power supply response capability has become the core link in the power load balancing control.

[0003] In the prior art, the power supply ratio scheduling method based on static model or rule triggering is generally adopted, which can realize basic load sharing function under most operating conditions. However, in the numerical calculation process of power supply ratio, due to the limitation of floating point algorithm or fixed point calculation in the calculation structure, the output result of power supply ratio is usually affected by bit width rounding, truncation processing or low bit omission. This precision loss problem is not easy to appear under normal ratio, but in the case of highly unbalanced power sharing ratio, such as the extreme ratio scene of ninety percent to ten percent, eighty-five percent to fifteen percent, the calculation error will be amplified due to the rounding direction and granularity limitation, causing the nonlinear bias of power supply ratio output.

[0004] More seriously, such nonlinear bias can not be identified in time through traditional backtracking correction mechanism. The reason is that in the initial calculation stage, the power supply state is equivalent to an ideal linear output unit, when the load suddenly rises, one power supply may continuously bear a high proportion of load, while the other power supply is suppressed due to the lower limit of the proportion, so that one power supply continuously runs in the overloading edge area. Such phenomenon is difficult to be detected by the system in a short period, once it continues to medium and long term, it will cause the accumulation of heat of high load power components, trigger device accelerated aging, and even the phenomenon of rapid decay of power output capacity, which seriously affects the stability and life cycle of the whole power architecture. Therefore, the present application proposes a dual-input computing power intelligent load balancing method to solve the above problems. SUMMARY

[0005] To achieve the above purpose, the present application provides the following technical scheme: A dual-input computing power intelligent load balancing method, comprising the following steps: The voltage, current, power and load change data of the two input power sources are collected, and a mathematical model for power supply ratio calculation is constructed based on the continuous collection results, so that the mathematical model reflects the functional relationship between the output capacity of the power source and the system load; According to the current output capacity of the power source and the system load state, a calculation precision configuration matched with the running scene is selected from the mathematical model, and the initial calculation of the power supply ratio is completed by adjusting the precision parameters for numerical calculation; The initial calculation of the power supply ratio is compared with the current load response of the input power source, and the precision parameters in the mathematical model are modified according to the power supply deviation. The modified precision parameters and the load trend of the next period are jointly used as inputs to perform a new round of precision parameter update on the mathematical model, so that the precision parameters can be improved period by period in the continuous load fluctuation environment, thereby realizing the continuous adaptation of the power supply ratio calculation path to the load change; In the process of updating the precision parameters, the temperature change data of the two input power sources are collected. When it is detected that the temperature of any power source is in a risk state, the temperature change data is included in the precision parameter update process, and the performance deviation caused by temperature rise is suppressed by adjusting the precision parameters, so that the power supply ratio remains stable and thermally safe in the continuous fluctuation environment.

[0006] In a preferred embodiment, constructing a mathematical model for power supply ratio calculation includes the following steps: The voltage, current and power parameters of the two input power sources are collected in multiple consecutive running periods, and the real-time change values of the system load are recorded synchronously. By classifying the power source response data under different load levels in time sequence, an original data matrix is established; Based on the original data matrix, a segmented linear fitting method is used to establish the functional relationship between the power source power change rate and the load change, and a change rate threshold is added to the model structure to describe the adaptability interval of the power source carrying capacity under sudden load.

[0007] In a preferred embodiment, establishing the functional relationship between the power source power change rate and the load change means: According to the collected data, an original data matrix is generated, the power change amount and the load change amount in each period are extracted, and a standardized change rate data set is formed; The load change amount is divided into multiple continuous intervals according to the change amplitude, and an independent linear fitting operation is performed in each interval to construct a local functional relationship; The linear expressions generated in each interval are combined into a segmented mapping table, and the segmented mapping table is used as one of the estimation bases of the power supply ratio calculation model.

[0008] In a preferred embodiment, the initial calculation of the power supply ratio by selecting a calculation precision configuration matched with the running scene means: After the real-time collection of the power supply output capability and the system load state is completed, the current load fluctuation level is determined according to the load change rate, and the precision parameter set corresponding to the fluctuation level is selected from the preset precision configuration set, and the selected precision parameter set is applied to the numerical processing structure in the power supply ratio calculation model.

[0009] In a preferred embodiment, the modification of the precision parameters in the mathematical model according to the power supply deviation means: The interval granularity parameter in the precision parameter set is applied to the load change interval division logic, and the interval number and boundary precision of the segmented mapping table are controlled by adjusting the minimum resolution unit of each interval; According to the lookup table mode control strategy set in the precision parameter set, it is selected whether to perform interpolation processing on the power supply ratio results between the mapping intervals, so that the power supply ratio output has continuity and flexible transition ability in the critical interval; After the power supply ratio calculation is completed, the output fine tuning factor in the precision parameter set is used as a multiplication coefficient to adjust the initial power supply ratio output value to compensate for the deviation caused by data quantization or model fitting residual error.

[0010] In a preferred embodiment, the new round of precision parameter update of the mathematical model includes the following steps: The adjusted initial power supply ratio output value, i.e. the calculated power supply ratio, is multiplied by the total system load power of the current period to obtain the target power supply power value of the two input power supplies, and the target power supply power value of each power supply is compared with its corresponding actual output power to calculate the power supply power deviation of the two power supplies; The power supply power deviations of the two power supplies are combined into a comprehensive power supply deviation value of the current period by weighted average method, and the average value is calculated by combining the comprehensive power supply deviation value of the last period to form the power supply offset index; According to the power supply offset index, the corresponding precision parameter set is selected in the mathematical model, and the power supply ratio calculation is re-executed based on the updated precision parameter set to generate the corrected power supply ratio output of the current period.

[0011] In a preferred embodiment, the collection of temperature change data of the two input power supplies means: The continuous temperature sampling values of the two input power supplies are collected, the temperature change rate of each in unit time is calculated, and the comparison with the set highest temperature threshold and temperature rise rate threshold is made to identify whether there is a risk trend.

[0012] In a preferred embodiment, the risk state means that the temperature of the power supply exceeds the set highest temperature threshold, or the change rate of the temperature of the power supply in unit time exceeds the set temperature rise rate threshold.

[0013] In a preferred embodiment, incorporating temperature change data into the precision parameter updating process comprises the following steps: When it is detected that any power supply temperature is in a risk state, the corresponding current power supply path is marked as a restricted path, a precision parameter set with a load limit is screened based on the mark, and a preset compression process is performed on the calculation interval associated with the restricted path in the segmented mapping table to limit the fluctuation range of the power supply ratio in a short period. The updated precision parameters are written into the power supply ratio calculation process, and the final output is the power supply ratio result after thermal adjustment.

[0014] The technical effects and advantages of the present application are: The present application converts the traditional static power supply control into a numerical calculation process with dynamic expression ability by constructing a mathematical model reflecting the functional relationship between power supply output capacity and system load. The model is established based on the continuous collection results of the voltage, current, power and load state of two input power supplies, and can match different calculation precision configurations as the running scene changes, so that the system can flexibly adjust the initial calculation strategy of the power supply ratio according to the actual load condition. This way of coupling the calculation process of the power supply ratio with the actual running environment not only improves the flexibility of the calculation, but also enhances the adaptation ability to complex running scenes, thereby significantly improving the scheduling efficiency and resource utilization between the power supply path and the load.

[0015] After completing the initial calculation of the power supply ratio, the present application compares the calculation result with the current load response of the input power supply, identifies the power supply deviation, and uses the deviation to correct the precision parameters in the mathematical model, forming a calculation path with feedback correction ability. The corrected precision parameters are not applied statically, but are updated together with the load trend of the next period as model inputs, so that the precision parameters evolve continuously in each scheduling period and gradually approach the dynamic needs of the actual load fluctuation. This periodic precision parameter updating mechanism enables the power supply ratio calculation path not only to have fast response ability, but also to form a coherent adaptation characteristic to load changes, significantly improving the power supply control stability and calculation precision in the scene of frequent fluctuating loads.

[0016] The application introduces the temperature change data of two input power supplies as the adjustment basis in the precision parameter updating process, and when any power supply temperature is in a risk state, the related temperature change data is included in the precision parameter updating process. Through this mechanism, the system can identify and respond in the early stage of temperature rise abnormality, dynamically adjust the precision parameters in the calculation process, suppress the performance deviation caused by temperature abnormality, and prevent the load distribution error caused by uneven heat. In the control environment with continuous fluctuation of power supply ratio, the mechanism effectively guarantees the thermal safety and performance stability of the power supply path, thereby improving the reliability, anti-interference ability and long-term running consistency of the entire control process under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to facilitate the understanding of those skilled in the art, the application will be further described below with reference to the accompanying drawings; Figure 1 A schematic diagram of a dual-input computing power supply intelligent load balancing method in the application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0019] Reference Figure 1 The following embodiments are obtained: Embodiment 1: A dual-input computing power supply intelligent load balancing method, comprising the following steps: Collecting voltage, current, power and load change data of two input power supplies, and constructing a mathematical model for power supply ratio calculation based on continuous collection results, so that the mathematical model reflects the functional relationship between power supply output capacity and system load; this step is used to obtain the dynamic interaction behavior between power supply and load in actual running environment, and the original data structure is established by continuous sampling of multi-period data, so as to ensure that the constructed mathematical model can fully express the response capability and power supply characteristics of different power supplies under different load scenarios, and provide basic function relationship support for subsequent power supply ratio calculation.

[0020] According to the current output capability of the power supply and the system load state, select the calculation precision configuration matching the running scene from the mathematical model, and complete the initial calculation of the power supply ratio by adjusting the precision parameters for numerical calculation; this step aims to match the precision configuration scheme suitable for the current fluctuation scene from the existing model structure by combining the real-time system state, and dynamically select the precision parameters to regulate the calculation path of the power supply ratio, so that the model has numerical resolution in different scenes such as load mutation and light load steady state, thereby generating a power supply ratio output with preliminary balancing ability.

[0021] Compare the initial calculation of the power supply ratio with the current load response of the input power supply, modify the precision parameters in the mathematical model according to the power supply deviation, and use the modified precision parameters and the load trend of the next period as inputs to perform a new round of precision parameter update on the mathematical model, so that the precision parameters can be improved period by period in the continuous load fluctuation environment, thereby realizing the continuous adaptation of the power supply ratio calculation path to the load change; this step forms a feedback adjustment mechanism by analyzing the deviation between the calculation result and the real running response, which is used to correct the potential numerical error or response delay in the model, and continuously optimize the precision parameters combined with the predicted load trend, to build a model regulation path evolving period by period, so that the output of the power supply ratio can respond stably and continuously to the actual change of the load, preventing error accumulation or power supply mutation caused by insufficient modeling accuracy.

[0022] In the process of updating the precision parameters, the temperature change data of the two input power supplies is collected, and when it is detected that the temperature of any power supply is in a risk state, the temperature change data is included in the precision parameter update process, and the performance deviation caused by temperature rise is suppressed by adjusting the precision parameters, so that the power supply ratio remains stable and thermally safe in a continuously fluctuating environment. This step introduces a thermal safety control mechanism, which synchronously monitors the power supply thermal state while updating the precision parameters, actively identifies the risk of power supply overheating caused by continuous power output or sudden load, and dynamically modifies the precision parameter structure based on this, thereby suppressing the power supply deviation behavior under high temperature operating conditions, ensuring that the power supply ratio not only has numerical accuracy, but also considers the thermal stability and reliability of system operation.

[0023] The mathematical model for power supply ratio calculation includes the following steps: collecting voltage, current and power parameters of two input power supplies in multiple consecutive operation cycles, and synchronously recording real-time change values of system load. Specifically, each sampling cycle is set to 20 milliseconds, and a total of not less than 1000 sampling cycles are selected to constitute a data window, forming an initial operation data segment with a length of 20 seconds. During the sampling process, the instantaneous voltage (unit: volt), current (unit: ampere) and output power (unit: watt) of input power supply A and input power supply B are collected, and the corresponding real-time load value (unit: watt) of the load end is recorded. All data are time-stamped and have a strict sampling interval to ensure the time consistency and response continuity of the sampling data.

[0024] By classifying the power supply response data under different load levels in time series, an original data matrix is established. In order to realize the structured expression of power supply output response characteristics, all collected samples are divided into multiple non-overlapping load level intervals according to the load value in the 0% to 100% full load interval, with each interval spanning 10% and a total of 10 load levels. In each level interval, the voltage, current and power data in the sampling time window are classified and stored respectively, and a five-dimensional data structure containing sampling point index, power supply number, instantaneous output power, load value and corresponding load level is established. The above original data matrix takes the running time slice as the row and the sampling attribute as the column, forming a basic modeling data framework and providing structured input for the subsequent fitting modeling process.

[0025] Based on the original data matrix, a segmented linear fitting method is used to establish the functional relationship between the power change rate and the load variation of the power supply. Specifically, the above 10 load levels are further refined into smaller continuous variation intervals, with a default 1% load change as the smallest fitting unit, and each interval contains at least 100 valid data samples. Within each variation interval, the least squares method is used to perform a linear fitting operation on the power change rate and load variation data of input power supply A and input power supply B, and the linear expression coefficients in the corresponding interval are obtained. All fitting results are numbered according to the interval index and saved in the fitting expression index table. To avoid fitting accuracy deviation, normalization is performed before each fitting, and sampling samples with a power change amplitude less than 5 watts are removed to improve the stability of modeling and the reliability of prediction.

[0026] A change rate threshold is added in the model structure to describe the adaptability range of the power supply under burst load. In the specific implementation, the threshold represents the maximum power variation amplitude that the input power supply can stably respond to in a unit time. When performing linear fitting in each range, if the power change rate of a large number of samples, for example, sixty percent of the samples, in a certain range exceeds the set threshold, it is considered that the range is in a high dynamic operation range, and this type of range will be marked as a high response range. In the modeling process, different residual tolerances can be used in this type of high response range to enhance the fitting stability and calculation sensitivity of the input power supply under burst load scenarios.

[0027] Establishing the functional relationship between the power change rate and the load variation means that the original data matrix is generated according to the collected data, the power change and the load change in each period are extracted, and the standardized change rate data set is formed. In multiple consecutive operation periods, the instantaneous output power values of the two input power supplies at the beginning and end of each period and the load power values in this period are recorded respectively. Taking each sampling period as a sample unit, the output power change value of the two input power supplies in the period is calculated, which is equal to the power value at the end of the period minus the power value at the beginning of the period, and the corresponding load power change value is calculated in the same way. Then, the power change value of each input power supply and the corresponding load power change value are divided by the period length to convert the change rate in a unit time, forming the power change rate data set and the load change rate data set corresponding to each input power supply respectively. In order to eliminate the influence of data scale, the above change rate values are normalized to adjust the values to the range of zero to one, forming the standardized change rate data set for subsequent fitting analysis.

[0028] The load change amount is divided into multiple continuous intervals according to the change range, and a linear fitting operation is performed in each interval to construct a local function relationship. In the normalized change rate data set, the numerical range of the load change rate is divided into several non-overlapping continuous intervals, each interval span is set to five percent, forming twenty intervals. Each interval contains multiple sample data pairs, that is, the combination of the load change rate and the corresponding power change rate within the interval range. For each input power source, all data samples belonging to each interval are extracted independently in each interval, and a linear fitting operation is performed once. Linear fitting determines the slope and intercept of the fitting function according to the principle of minimum error, so that the numerical relationship between the load change rate and the power change rate in the interval is closest to the straight line function change rule. The fitting result is expressed in words as: the power change rate in a certain interval is equal to a certain coefficient multiplied by the load change rate plus another constant term, where the coefficient represents the response strength of the input power source in the load fluctuation interval, and the constant term represents the power change offset of the input power source when there is no load change. In this way, the fitting operation is completed for all load change rate intervals to obtain a complete set of interval fitting functions.

[0029] The linear expressions generated in each interval are combined into a segmented mapping table, which is used as one of the estimation bases for the power supply ratio calculation model. In the processing flow of each input power source, the linear function result generated in each of the aforementioned intervals is established in a one-to-one correspondence with its corresponding load change rate interval to form an interval-function mapping entry. For example, the load change rate interval of zero to five percent corresponds to a certain linear expression; the interval of five to ten percent corresponds to another linear expression, and so on to build a complete mapping table. The segmented mapping table can be used in the power supply ratio calculation model. In the subsequent model calling stage, when the load change rate identified in the current period falls into a certain interval, the linear expression corresponding to the interval is automatically called to calculate the estimated power change rate of the input power source in response to the current load change. The estimated power change rates of the two input power sources are then weighted or normalized to obtain the corresponding power supply ratio. The mapping mechanism can improve the sensitivity and prediction accuracy of the load change response in actual execution, while having the processing efficiency advantages of fast table lookup and segmented calculation.

[0030] A logical judgment mechanism is introduced in the constructed segmented mapping table structure, which is used to dynamically select whether to perform interval boundary interpolation, so as to improve the fitting smoothness during the transition of continuous intervals. When the load change rate value is close to the boundary intersection of two intervals, the system can enable the transition weight calculation logic according to the configuration, so that the power change rate prediction value of the current period is weighted and calculated in a certain proportion in the linear expression of the adjacent two intervals, thereby avoiding sudden changes when crossing intervals and improving the continuity and stability of the power supply ratio calculation process. The interpolation calculation method can be expressed in words as follows: the current load change rate prediction value is equal to the product of the previous interval function calculation result and the previous weight, plus the product of the next interval function calculation result and the next interval weight, and the sum of the two weights is equal to one, and is inversely proportional to the distance from the current value to the center value of the two intervals. In this way, the function jump phenomenon caused by segmented linear fitting is effectively avoided, and smooth and controllable fitting support is provided for subsequent power supply ratio output.

[0031] The initial calculation of the power supply ratio by selecting the calculation precision configuration matching the running scene means that after the real-time collection of the power output capability and the system load state is completed, the current load fluctuation level is determined according to the load change rate. Specifically, the difference sequence of the system load power sampling values in the last ten consecutive running periods is calculated to obtain the load power change amount of each period, and then divided by the period length to convert it into a load change rate. After obtaining the load change rate, according to the preset level division rule, such as low fluctuation level, medium fluctuation level and high fluctuation level, three types of running scenes corresponding to load change rates less than 5%, between 5% and 20%, and greater than 20% are respectively determined, so as to determine the load fluctuation level of the current running period.

[0032] The precision parameter set corresponding to the fluctuation level is selected from the preset precision configuration set. The precision configuration set is a set of predefined parameter templates, each template consisting of three dimensions: interval granularity parameter, table lookup control parameter and output fine tuning factor. Among them, the interval granularity parameter represents the minimum resolution unit of the load change interval division in the power supply ratio calculation model, for example, five watts, ten watts or twenty watts; the table lookup control parameter indicates whether to enable the boundary interpolation mechanism between different interval fitting expressions; the output fine tuning factor is a dimensionless multiplication factor, which is used to compensate for the fitting model error at the output end. Taking the medium fluctuation level as an example, the system can select a precision parameter set with a granularity of ten watts, enable linear interpolation, and a fine tuning factor of 0.98.

[0033] The selected precision parameter set is applied to the numerical processing structure in the power supply ratio calculation model. Specifically, first, the interval granularity parameter in the precision parameter set is used to redivide the load change range, determine the number of fitting intervals and the boundary interval in the segmented mapping table; second, a linear fitting operation is performed in each interval to obtain the function expression of the interval, which is: the power change rate in the interval is equal to a numerical value representing the input power response strength multiplied by the load change rate, plus another numerical value representing the power offset of the input power without load change. The numerical precision of the response strength coefficient will be controlled by the interval granularity parameter in the selected precision parameter set. If the granularity is smaller, the slope floating range allowed by the model is finer, indicating that the local response of the fitted curve is more sensitive. Under the action of the table lookup control parameter, if the load change rate in two adjacent fitting intervals is located between the boundaries, interpolation calculation is enabled according to the set strategy, and weighted smoothing is performed between the output values of the two fitted straight lines to improve the continuity across intervals. Finally, the output fine tuning factor is multiplied by the power change rate value obtained by preliminary fitting calculation to obtain the corrected result for power supply ratio estimation.

[0034] According to the power supply deviation, the precision parameters in the mathematical model are modified, that is, after the power supply ratio calculation is completed, the target power supply power values of the two input power sources are compared with their actual output powers, the deviations of the two are calculated respectively, and the precision parameter set used in the current period is dynamically adjusted based on this. The modification process includes three aspects: first, the interval granularity parameter in the precision parameter set is applied to the load change interval division logic, if the power supply deviation in a certain interval exists for a long time, for example, the deviation exceeds five percent for more than three periods, the granularity value of the interval is automatically reduced according to the preset rule, for example, from ten watts to five watts, thereby improving the fitting sensitivity of the region; second, according to the table lookup mode control strategy set in the precision parameter set, it is selected whether to perform interpolation processing on the power supply ratio results between the mapping intervals, if it is found that the interpolation is missing, which leads to the cross-interval jump, the interpolation control strategy is adjusted to the enabled state; third, after the power supply ratio calculation is completed, the output fine tuning factor in the precision parameter set is used as a multiplication coefficient to adjust the initial power supply ratio output value, for example, the fitted output result is multiplied by zero point nine six to reduce the error caused by the fitting residual. The execution logic of the above modification steps and the power supply deviation value form a mapping relationship, and are dynamically iterated in continuous periods, gradually optimizing the stability and robustness of the power supply ratio calculation path.

[0035] The new round of precision parameter updating of the mathematical model comprises the following steps: multiplying the adjusted initial power supply ratio output value, i.e. the calculated power supply ratio, by the total system load power in the current period to obtain the target power supply power values of the two input power sources, and comparing the target power supply power value of each power supply with the actual output power corresponding thereto to calculate the power supply power deviation of each power supply; specifically, in the current period, the corresponding segmented mapping table of each input power source is used to find the interval in which the load change is located in the mapping table according to the real-time collected load change, and the linear expression corresponding to the interval is called to calculate the power change rate prediction value of each input power source in the current period. After the prediction value is multiplied by the historical reference power value of the corresponding power source, the power change value of each power supply is obtained, and the power change values of the two power supplies are added and normalized to obtain the initial power supply ratio output value of the two input power sources in the period.

[0036] To further improve the accuracy of the power supply ratio under the actual running load, the system corrects the initial ratio according to the output fine-tuning factor in the selected precision parameter set after the last correction step to form the adjusted initial power supply ratio output value. For example, if the mapping calculation output of the two power sources is sixty-five percent and thirty-five percent, and the fine-tuning factor is ninety-eight percent, then the final ratio is sixty-three point seven percent and thirty-four point three percent. Subsequently, the two power supply ratio values are multiplied by the total load power in the current period, for example, one kilowatt, to obtain the target power supply power values of the two power sources, which are six hundred thirty-seven watts and three hundred forty-three watts. Then, the target power values are compared with the real-time collected power output to obtain the difference values, forming the power supply power deviation of each power supply.

[0037] The power supply power deviations of the two power sources are integrated into the comprehensive power supply deviation value of the current period by weighted averaging, and the average value is calculated by combining the comprehensive power supply deviation value of the last period to form the power supply offset index; the power supply power deviations of the two power sources represent the error amplitude between the predicted power supply and the actual power supply, and the deviation values of the two power sources need to be integrated for accurate evaluation of the error of the overall power supply strategy. Therefore, the system calculates the weighting coefficient according to the proportion of the two power sources in the initial power supply ratio. For example, the first power source ratio is sixty-five percent, so its weight is zero point six five, and the second power source is thirty-five percent, so its weight is zero point three five. Multiplying each power supply power deviation by the corresponding weight and then adding them together can obtain the comprehensive power supply deviation value of the current period. To avoid model fluctuations caused by isolated period disturbances, the current comprehensive power supply deviation and the comprehensive power supply deviation of the last period are also averaged, for example, the last period is fifteen watts and the current period is twenty watts, so the final power supply offset index is seventeen point five watts. The power supply offset index serves as a systematic error index for measuring the difference between the current model calculation result and the actual output, and is used to guide the selection process of the next precision parameter.

[0038] According to the power supply offset index, the corresponding precision parameter set is selected in the mathematical model. The precision parameter set is predefined as multiple configuration profiles in the modeling stage, and each profile covers three items: interval granularity parameter, lookup table control parameter, and output fine-tuning factor. To facilitate mapping selection, the model presets a mapping rule table of power supply offset index and precision parameter group. The offset below ten watts is the low error zone, the offset from ten to thirty watts is the medium error zone, and the offset above thirty watts is the high error zone. Each error zone corresponds to a precision parameter set with higher optimization degree. For example, if the power supply offset index is seventeen and a half watts, matching the medium error zone configuration group, the currently selected precision parameters are: granularity parameter is five watts, interpolation is enabled, and fine-tuning factor is ninety-six percent. In the new round of calculation, the granularity parameter in the precision parameter set is used to update the interval distribution logic of the segmented mapping table, the lookup table control parameter is embedded into the interval boundary response judgment process, and the output fine-tuning factor is used for end multiplication adjustment of the power supply ratio result.

[0039] Based on the updated precision parameter set, the power supply ratio calculation is re-executed to generate the corrected power supply ratio output for the current period. According to the latest selected precision parameter set, the model re-executes the segmented mapping table update, linear expression fitting, power supply ratio output and fine-tuning compensation process, and completes the power supply ratio recalculation based on feedback correction. For example, the updated interval granularity is reduced to five watts, which means that the segmented mapping table will have a more intensive interval distribution, making the model more delicate in load fluctuation response. The recalculated power supply ratio output will have higher matching accuracy, so that the deviation between the target power supply power value and the actual output power value will tend to converge in the subsequent period. After completing this round of update, the new power supply ratio output value is used as the final scheduling basis for this period, and the relevant correction information is transmitted to the next period of collection-modeling-output process, ensuring the continuous closed loop of the load balancing control chain.

[0040] Collecting temperature change data of two input power sources refers to collecting temperature change data of two input power sources during the process of updating precision parameters. Specifically, temperature values of two input power sources are collected once every second, and 10 sets of continuous temperature sampling values are obtained within a continuous 10-second period. By calculating the temperature change amplitude between the front and back sampling points in these 10 sets of data, the temperature change rate per unit time is obtained. The unit time can be preset to 5 seconds, for example, if the temperature of a power source is 45°C at 0 seconds and 50°C at 5 seconds, the temperature change rate is 1°C / s; if another power source rises from 48°C to 54°C within the same time period, the temperature change rate is 1.2°C / s. The above temperature change rates are compared with the highest temperature threshold and temperature rise rate threshold set in the system to determine whether the power source is in normal working state.

[0041] Based on the real-time temperature rate calculation results in the preceding acquisition step, it is identified whether there is a temperature risk trend. This process is: "acquire continuous temperature sampling values of two input power supplies, calculate the temperature change rate of each in unit time, and compare it with the set maximum temperature threshold and temperature rise rate threshold to identify whether there is a risk trend." In actual application, the maximum temperature threshold can be set to 65°C, and the temperature rise rate threshold can be set to 1.5°C / s. For example, if the temperature change rate of any one of the two power supplies exceeds the threshold, or its temperature itself reaches or exceeds 65°C, it is judged that the power supply has entered a temperature risk state. Through this judgment logic, the power supply input end with a potential overheating trend can be identified in time to prevent error amplification or system protective shutdown caused by overheating. When any input power supply is detected to be in a risk state, the following operation is further performed: when any power supply temperature is detected to be in a risk state, the temperature change data is included in the precision parameter update process.

[0042] Including the following steps in the precision parameter update process: in the process of continuously monitoring the temperature sampling of the input power supply, if it is detected that any power supply temperature is in a risk state, i.e. the temperature value exceeds the preset maximum temperature threshold, or the temperature rise rate in unit time exceeds the temperature rise rate threshold, immediately perform the path marking operation to set the power supply path corresponding to the current power supply as a restricted path. The "restricted" attribute of the path will be stored as a state identifier in the state recording area of the power supply scheduling module for subsequent conditional judgment. In this embodiment, if the temperature of power supply A rises from 52°C to 61°C in 3 seconds, the temperature rise rate is 3°C / s, which exceeds the set threshold of 2°C / s, power supply A is marked as a high-temperature risk source, and its power supply path P1 is marked as a restricted path at the same time, which is used as a judgment basis for downstream precision parameter screening.

[0043] Based on the marking information of the restricted path above, a precision parameter set with limited load is screened out. The so-called "precision parameter set with limited load" means that for a certain risk path, a subset of parameters that are not suitable for high-temperature state is excluded from the original precision parameter set, and only parameter items that are still representative under certain limited conditions are retained. For example, the power supply path P1 has a wide range of precision parameter values such as load current, temperature coefficient, and voltage response time when it is in normal operation, but only a part of the parameter items that remain stable under high-temperature conditions are allowed to be used in the risk state, such as the precision point position that limits the voltage deviation to not more than ±0.2% under the restriction of temperature rise. In this way, a smaller precision parameter set with limited load is formed, thereby avoiding the risk path from participating in the wrong high-precision calculation, and ensuring the stability and safety of power supply regulation.

[0044] After the set of precision parameters with limited load is screened out, the calculation interval associated with the limited path in the segmented mapping table is subjected to a preset compression process to limit the fluctuation range of the power supply ratio in a short period. The "segmented mapping table" here refers to the dynamic correspondence table in the power supply adjustment module for managing the relationship between the power supply ratio of each path and parameters such as temperature, load, and voltage. The "compression process" refers to reducing the upper and lower limit range of the power supply ratio adjustment of the path in a short period (e.g., 30 seconds) when there is a limited path. In specific implementation, the fluctuation range of the power supply ratio of the original P1 path can be compressed from ±20% to ±8%, so that the power supply ratio remains stable in the high-temperature risk state and avoids frequent adjustments that cause thermal load oscillation. The purpose of this is to reduce the influence of the limited path during the high-temperature fluctuation period by "compressing" the response interval of the segmented mapping curve, thereby improving the robustness of the overall power supply balancing strategy.

[0045] After the screening of the set of precision parameters and the compression process of the segmented mapping table are completed, the updated precision parameters are written into the power supply ratio calculation process, and the final output of the power supply ratio result after thermal adjustment is obtained. In this process, the original calculation process will call the updated set of precision parameters with limited load, and reconstruct the power supply ratio of the current period within the set sampling period according to the compressed mapping rule. For example, if the original calculation result is path P1: 45%, path P2: 55%, and P1 is in a limited state, then under the action of the thermal adjustment mechanism, the output will be corrected to P1: 38%, P2: 62%. This dynamic adjustment method not only ensures that the load of the risk path is reasonably controlled, but also improves the adaptability of the power supply model to environmental thermal changes.

[0046] In actual operation, to enhance the stability and fault tolerance of the power supply adjustment process in extreme scenarios, the following redundancy protection and retreat mechanisms can be set. When it is detected that both input power supplies are in a temperature risk state at the same time, that is, either of the conditions "the temperature of the power supply exceeds the set maximum temperature threshold, or the rate of change of the temperature of the power supply in unit time exceeds the set temperature rise rate threshold" is met, the double-path limited state will be triggered. In this case, to avoid the loss of effective input path in the power supply proportion calculation process leading to output failure, a fault tolerance mechanism can be enabled for processing. Specifically, under the premise of double-path limitation, the input path with relatively slow temperature change rate and far away from the maximum temperature threshold is preferentially selected, and temporarily allowed to participate in power supply proportion calculation, while based on the deviation of its current temperature and temperature rise rate from the risk threshold, the maximum allowed value of its power supply proportion is limited, which is used to reduce its load contribution in the total power supply. Assuming that the power supply proportion of the path does not exceed 30%, to realize dynamic load control of the risk path, for example, if the temperature rise rate of input power supply A is 1.8°C / s, the current temperature is 64°C, and the temperature rise rate of input power supply B is 2.5°C / s, the current temperature is 66°C, then power supply A is selected as the "second limited path", and is allowed to participate in power supply proportion calculation with the limitation of maximum power supply proportion 30% without restoring all precision parameters, to ensure that the output is not interrupted. While for the other input path with higher temperature rise rate and closer or exceeding the set maximum temperature threshold, a strong restriction strategy is executed, which excludes its power supply proportion participation right in the power supply period, or limits its proportion in the power supply proportion to not more than 10% of the minimum fault tolerance output range, only retaining the basic redundancy ability to maintain the continuity of the adjustment process; at the same time, under the premise that power supply A and power supply B are in a limited state, a backup path such as an energy storage battery or a third input power supply channel is further started to dynamically make up for the remaining required power supply proportion, to ensure that the overall power supply capacity meets the load demand, and realize the thermal risk closed-loop control of power supply stability.

[0047] In order to prevent parameter update abnormality from causing regulation instability, a backoff and alarm mechanism can be set when an input path is continuously in a limited state for N cycles. The specific implementation is as follows: if a power supply path is continuously marked as a limited path for more than N sampling cycles (for example, N=6 and the cycle is 60 seconds), the path abnormality processing flow is activated. The flow includes the following operations: ① temporarily exclude the path from the power supply proportion calculation flow and enter a "cooling waiting" state, and the default duration is 30 seconds; ② start the system log recording module to record the start time, temperature data, change rate and precision parameter compression amplitude of the limited state completely, which is used for subsequent fault backtracking; ③ when entering the N+1 cycle, if the path state is not removed, an alarm signal is sent to the remote monitoring end or the operation terminal through the communication module, prompting that the current power supply path has a long-term risk abnormality; ④ if the continuous limitation exceeds the set upper limit cycle (for example, 12 cycles), the system switches to the minimum stable operation mode, only the core load power supply is reserved, and the standby path (such as a battery or a third input end) is started to maintain the regulation closed loop uninterrupted.

[0048] The above algorithms or formulas are all dimensionless values, and the results of the latest real situation are obtained by collecting a large amount of data and simulating software. The preset parameters are set by a person skilled in the art according to the actual situation.

[0049] It should be understood that the size of the sequence number of the above processes in various embodiments of the present application does not mean the order of execution, and the execution order of the processes should be determined according to their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0050] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0051] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0052] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A dual-input hash power supply intelligent load balancing method, characterized in that, The method comprises the following steps: Collecting voltage, current, power and load change data of two input power supplies, and constructing a mathematical model for power supply ratio calculation based on the continuous collection results, so that the mathematical model reflects the functional relationship between power supply output capacity and system load; According to the current output capacity of the power supply and the system load state, select the calculation accuracy configuration matched with the running scene from the mathematical model, and complete the initial calculation of the power supply ratio by adjusting the accuracy parameters for numerical calculation; Comparing the initial calculation result of the power supply ratio with the current load response of the input power supply, and modifying the accuracy parameters in the mathematical model according to the power supply deviation, and taking the modified accuracy parameters and the load trend of the next period as inputs to execute a new round of accuracy parameter update on the mathematical model; In the process of updating the accuracy parameters, collect the temperature change data of the two input power supplies, and when detecting that the temperature of any power supply is in a risk state, include the temperature change data in the accuracy parameter update process, and suppress the performance deviation caused by temperature rise by adjusting the accuracy parameters.

2. The dual-input power supply intelligent load balancing method of claim 1, wherein, The method for constructing a mathematical model for power supply ratio calculation comprises the following steps: Collecting voltage, current and power parameters of two input power supplies in multiple consecutive running periods, and synchronously recording the real-time change value of system load, establishing an original data matrix by classifying the power supply response data under different load levels in time sequence, and establishing a function relationship between the power supply power change rate and the load change based on the original data matrix, and adding a change rate threshold in the model structure to describe the adaptability interval of the power supply carrying capacity under sudden load. The function relationship between the power supply power change rate and the load change refers to:

3. The dual-input power supply intelligent load balancing method of claim 2, wherein, According to the collected data, generate an original data matrix, extract the power change and load change in each period to form a standardized change rate data set; Divide the load change into multiple continuous intervals according to the change amplitude, and perform independent linear fitting operation in each interval to construct a local function relationship; Combine the linear expressions generated in each interval into a piecewise mapping table, and take the piecewise mapping table as one of the estimation bases of the power supply ratio calculation model. The initial calculation of the power supply ratio by selecting the calculation accuracy configuration matched with the running scene refers to:

4. The dual-input power supply intelligent load balancing method of claim 3, wherein, After completing the real-time collection of the power supply output capacity and the system load state, determine the current load fluctuation level according to the load change rate, and select the accuracy parameter set corresponding to the fluctuation level from the preset accuracy configuration set, and combine the selected accuracy parameter set into the numerical processing structure in the power supply ratio calculation model. The modification of the accuracy parameters in the mathematical model according to the power supply deviation refers to:

5. The dual-input power supply intelligent load balancing method of claim 4, wherein, Apply the interval granularity parameter in the accuracy parameter set to the load change interval division logic, control the interval number and boundary accuracy of the piecewise mapping table by adjusting the minimum resolution unit of each interval; According to the lookup table mode control strategy set in the accuracy parameter set, select whether to perform interpolation processing on the power supply ratio results between the mapping intervals, so that the power supply ratio output has continuity and flexible transition ability in the critical interval; ​ After the power supply ratio calculation is completed, the output fine-tuning factor in the precision parameter set is used as a multiplication coefficient to adjust the initial power supply ratio output value to compensate for the deviation caused by data quantization or model fitting residual error.

6. The dual-input power supply intelligent load balancing method of claim 5, wherein, The new round of precision parameter updating for the mathematical model includes the following steps: The adjusted initial power supply ratio output value, i.e., the calculated power supply ratio, is multiplied by the total system load power of the current period to obtain the target power supply power values of the two input power sources. The target power supply power value of each path is compared with its corresponding actual output power, and the power supply power deviation of each power source is calculated. The power supply power deviations of the two power sources are combined into a comprehensive power supply deviation value for the current period through weighted averaging, and the average value is obtained by combining the comprehensive power supply deviation value of the previous period to form the power supply offset index. According to the power supply offset index, the corresponding precision parameter set is selected in the mathematical model, and the power supply ratio calculation is performed again based on the updated precision parameter set to generate the corrected power supply ratio output for the current period.

7. The dual-input power supply intelligent load balancing method of claim 6, wherein, Collecting temperature change data of the two input power sources means: Collecting continuous temperature sampling values of the two input power sources, calculating the temperature change rate of each in unit time, and comparing it with the set maximum temperature threshold and temperature rise rate threshold to identify whether there is a risk trend.

8. The dual-input power supply intelligent load balancing method of claim 7, wherein, The risk state refers to the temperature of the power source exceeding the set maximum temperature threshold, or the change rate of the temperature of the power source in unit time exceeding the set temperature rise rate threshold.

9. The dual-input power supply intelligent load balancing method of claim 8, wherein, Including the following steps in the precision parameter updating process: When detecting that the temperature of any power source is in a risk state, mark the corresponding current power supply path as a restricted path, filter the precision parameter set with limited load based on the mark, and perform a preset compression processing on the calculation interval associated with the restricted path in the segmented mapping table to limit the fluctuation range of its power supply ratio in a short period. Write the updated precision parameters into the power supply ratio calculation process, and finally output the power supply ratio result after thermal adjustment.

Images