Safety control method for hot plug of power supply

Abstract

According to the power supply hot plug safety control method provided by the invention, through non-contact module identity recognition and a distributed cooperative pre-charging process considering the health state of the module, the risks of inrush current, electric arc and circulating current in the hot plug process are eliminated, real non-disturbance online replacement is realized, and the hot plug safety is improved; and an intelligent unit multi-party compensation learning algorithm is adopted, so that the system has prospective load tracking capability and autonomous negotiation optimization capability, and the dynamic response speed, the voltage stability and the overall energy efficiency of the system for power distribution are greatly improved. Through health state and global optimization distribution, overload or over-discharge of local modules is effectively avoided, and the overall service life and reliability of the power supply system are remarkably prolonged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power supply system safety, in particular to a power supply hot plug safety control method. BACKGROUND

[0002] A power supply system is usually composed of three core modules: a lithium battery module (responsible for energy storage and discharge), a voltage stabilizing module (such as an AC / DC or DC / DC converter, responsible for providing a stable bus voltage), and a power distribution module (responsible for distributing and managing bus power to various loads). Ideally, an operator can perform hot plug operations on any of the three modules without affecting the power supply of the entire power supply system to the load or critical load.

[0003] Current hot plug technology still stays in the passive level of "perception-reaction", and its technical idea can be summarized as follows: the type of the module is identified by detecting a pin, such as setting different resistance values of pull-down resistors on the module side for different types of modules. When the module is plugged in, the detection pin reads a specific voltage value, which is determined by the resistance value of the pull-down resistor on the module side, and the type of the module is identified by the resistance value. Then the control charges the capacitor in the pre-charge branch, and when the capacitor voltage is charged to the bus voltage, the output power is redistributed to the plugged-in module. When the module is unplugged, the power redistribution process is also based on this principle.

[0004] However, the above existing method has the following problems: 1. Only the resistance value is used to identify the type of the module, and the specific model, batch, and even health status of the module cannot be distinguished, so the subsequent control strategy cannot be refined; 2. The pre-charge control is only for the newly plugged-in module itself, ignoring the impact of its access on the overall network of the power supply system, which may pose a safety hazard; 3. The priority of the power that each module needs to increase or decrease is fixed, and it cannot adapt to the dynamic changes of the system state, and the power distribution strategy lacks self-learning and optimization capabilities; 4. The power transfer process lacks strict stability theory support, and it is difficult to avoid system oscillation in extreme cases. SUMMARY

[0005] The present application provides a power supply hot plug safety control method to achieve refined identification of the type of the plugged-in module, consider the impact of the newly plugged-in module on the overall network of the system, and dynamically optimize the power distribution strategy after the module is plugged in through self-learning, to avoid system oscillation in extreme cases, and to improve the safety of power supply hot plug.

[0006] To achieve this purpose, the present application adopts the following technical solutions: Provided is a power supply hot plug safety control method, applied to at least one of a lithium battery module, a power distribution module, and a voltage stabilizing module, plugged into a power supply system, comprising the steps of: S1, when the module is close to the system backboard to a preset threshold distance, the system backboard non-contact perceives the resonant frequency and internal equivalent impedance spectrum of the module, so as to serve as a physical unclonable identity representing the module type; S2, according to the physical unclonable identity, matching the type information and historical distributed cooperative pre-charging record of the module from the module characteristic database, and generating an adaptive pre-charging nonlinear control curve, and then diagnosing the health status of the module to be plugged in; S3, the power supply system takes the pin contact of the module to be plugged in to the backboard pin as an instruction, starts a distributed cooperative pre-charging process based on the adaptive pre-charging nonlinear control curve for all online modules, and optimizes the instantaneous impact kinetic energy and electromagnetic interference spectrum peak of the entire network, and solves and distributes the optimal driving signal of the pre-charging MOS tube in each online module in real time; S4, the optimal driving signal is applied to the gate of the pre-charging MOS tube of the corresponding online module, the plug-in action based on the module to be plugged in is completed, and the power supply safety control of all online modules is completed.

[0007] Preferably, the power supply hot plug safety control method further comprises the steps of: S5, according to the real-time maximum power capability function provided by the access module and the predicted total load demand of the system, all online modules are autonomously negotiated and quickly converged to a globally optimal and stable power distribution point through a smart unit multi-party compensation learning algorithm, and then each load in the network is powered.

[0008] Preferably, in step S2, the method for diagnosing the health status of the module to be plugged in comprises the steps of: A1, according to the data perceived by the system backboard and the historical distributed cooperative pre-charging record matched to the module to be plugged in, calculating the resonant frequency drift 、 value drift , low-frequency impedance real part drift and energy dissipation ratio slope of the module to be plugged in; A2, inputting 、 、 and to the health degree evaluation model bound to the module to be plugged in, and the model predicts the health status of the module to be plugged in.

[0009] Preferably, The calculation method of the health degree of the module includes steps of: A11, for each historical pre-charge event of the module to be plugged, calculating an energy dissipation ratio ,

[0010] represents the total energy dissipated in the form of heat energy through the pre-charge resistance and MOS on-resistance in the module to be plugged in the historical pre-charge event; represents the total energy transferred from the system bus to the input capacitor of the module to be plugged in the historical pre-charge event; A12, fitting the slope of each health feature factor of the N latest hot plug events of the module to be plugged with time as the energy dissipation ratio slope .

[0011] Preferably, the model prediction process in step A2 includes steps of: A21, calculating a capacitance aging factor , a connection resistance growth factor , a dynamic performance degradation factor according to the model input; ; A22, solving the health score corresponding to each health feature factor calculated in step A21, then weighting and fusing after obtaining the weight parameters associated with the module to be plugged, and taking the fusion result as the health degree of the module to be plugged; A23, judging whether the health degree is greater than a preset health degree threshold, if yes, further calculating the confidence degree of the health degree, and then turning to step A24; if no, triggering an alarm and terminating the power hot plug safety control process; A24, judging whether the confidence degree is consistent, if yes, taking the health degree as the model prediction output result; if no, triggering an alarm and terminating the power hot plug safety control process.

[0012] Preferably, in step S2, the method of generating the adaptive pre-charge nonlinear control curve includes steps of: B1, calling the voltage-charge nonlinear relationship model of the input capacitor of the module to be plugged from the module feature database; B2, combining the real-time monitored system bus voltage fluctuation range, and using the voltage-charge nonlinear relationship model constructed in step B1 to construct a target charge integral function taking time as the variable;​ B3, taking the target charge integral function as the boundary condition of the state space equation of the pre-charge circuit of the pseudo-plug-in module, and inversely solving the optimal gate voltage time sequence function of the pre-charge MOS tube according to the state space equation, and performing function fitting to obtain the adaptive pre-charge nonlinear control curve.

[0013] Preferably, in step B3, the method of inversely solving the optimal gate voltage time sequence function comprises the steps of: B31, discretizing the pre-charge period into sampling periods, and solving the optimal gate voltage of the pre-charge MOS tube at each sampling time point, the solving method comprising the steps of: B311, calculating the input capacitance charge amount at each sampling time point ; , adding the parasitic inductance in the pre-charge circuit to the current at the moment , and integrating the current in the time step; B312, taking as the dependent variable of the voltage-charge nonlinear relationship model constructed in step B1, and inversely solving the independent variable as the optimal gate voltage applied to the gate of the pre-charge MOS tube at the moment ; B32, constructing each at the moment as the optimal gate voltage time sequence function with the time point as the independent variable, as the dependent variable.

[0014] Preferably, the distributed collaborative pre-charge process in step S3 comprises the steps of: S31, incorporating the adaptive pre-charge nonlinear control curve generated for the pseudo-plug-in module into a global objective function ; S32, taking minimizing the global objective function as the optimization goal, and solving the optimal gate voltage control sequence of the optimal gate voltage of each online module at each sampling time in time sequence; S33, at the moment , applying the optimal gate voltage solved at the moment to the gate of the pre-charge MOS tube in the corresponding online module. ​

[0015] Preferably, in step S5, the method for predicting the total power demand of the system includes the following steps: C1, continuously monitor and record the total load power demand time series. , It is a one-dimensional data sequence, and then the one-dimensional data sequence is reconstructed into dimensional phase space sequence ; C2, tracking the point with the smallest initial distance. and The evolution index of distance over time And extract the largest evolution index. , recorded as ; C3, judgment Is it greater than 0? If so, the total power demand of the system is determined to be unstable, and then proceed to step C4; If not, then terminate step S5; C4, predict the total load power demand of the system, and use the prediction result as the input of the intelligent unit multi-party compensation learning algorithm.

[0016] Preferably, the operation process of the intelligent unit multi-party compensation learning algorithm includes the following steps: D1, build the current... for each online module The power compensation function at any given time incorporates the predicted total load power demand of the system, the maximum power capacity function, and the health status into the power compensation function. D2, each of the online modules aims to maximize the global compensation in the power compensation function, with local compensation in the power compensation function as a constraint. At each adjacent time step... Adjust its own power output to adjust the amount of adjustment; D3, judge The first global compensation calculated below is compared Is the deviation of the second global compensation calculated below less than the preset deviation threshold? If so, then for the current Moment With the online module Output power at time The summation is performed, and the summation result serves as the final power allocation point that the online module converges to the globally optimal and stable point. If not, then adjust. Find the value of , and then return to step D2; .

[0017] This application has the following beneficial effects: 1. By employing contactless module identification and a distributed collaborative pre-charging process that considers module health status, the risks of inrush current, arcing, and circulating current during hot-swapping are eliminated, achieving truly uninterrupted online replacement and improving the safety of hot-swapping. The use of a smart unit multi-party compensation learning algorithm enables the system to possess forward-looking load tracking capabilities and autonomous negotiation optimization capabilities, significantly improving the dynamic response speed, voltage stability, and overall energy efficiency of the system's power allocation. Through health status and global optimized allocation, local module overload or over-discharge is effectively avoided, significantly extending the overall lifespan and reliability of the power system.

[0018] 2. and dynamic performance degradation factor The introduction of this feature enables the model to have predictive capabilities. It can not only assess the current state of the module to be plugged in, but also predict future attenuation trends. As a result, in the safe control of hot-plugging of the module to be plugged in, it can more effectively and accurately solve and allocate the optimal drive signal of the pre-charge MOSFET in each online module. At the same time, it makes the autonomously negotiated power allocation point more accurate, realizing flexible and safe control of the power supply system.

[0019] 3. By incorporating the adaptive pre-charge nonlinear control curve generated for the newly plugged-in module into the global objective function. In this process, while striving to maintain the current stability of the existing online modules and avoid disturbances to them, the current of these modules can be allowed to undergo a controlled, small, and opposite offset to compensate for the disturbances brought about by the access of new modules. This allows the entire pre-charging process to achieve a globally optimal balance between "tracking accuracy" and "smoothness". Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments of this application will be briefly described below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a diagram illustrating the implementation steps of the power supply hot-swap safety control method provided in the embodiments of this application. Detailed Implementation

[0022] The technical solution of this application will be further described below with reference to the accompanying drawings and specific embodiments.

[0023] In the drawings, only for example, the representation is a schematic diagram, not a physical diagram, and cannot be understood as a limitation of the present application; in order to better illustrate the embodiments of the present application, some components of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0024] The same or similar reference numerals in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that if the terms "upper", "lower", "left", "right", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the positional relationship described in the drawings is only for example, and cannot be understood as a limitation of the present application, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0025] In the description of the present application, unless otherwise explicitly specified and limited, if the term "connection" and the like appear to indicate the connection relationship between components, the term should be broadly understood, for example, it can be fixedly connected, or it can be detachably connected, or it can be integrated; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication or interaction relationship between two components. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0026] The power supply hot plug safety control method provided by the embodiments of the present application is applied to at least one of the lithium battery module, the power distribution module and the voltage stabilizing module which is plugged into the power supply system, as shown in Figure 1 The method comprises the following steps: S1, when the module is close to the system backboard to reach a preset threshold distance, the system backboard non-contact perceives the resonant frequency and the internal equivalent impedance spectrum of the module, which is used as a physical unclonable identity tag representing the module type; Specifically, behind each slot (for plugging module pins) of the system backboard of the power supply system, a miniature antenna array (such as a microstrip patch antenna or a spiral inductor) driven by a monolithic microwave integrated circuit is integrated, and the working frequency range of the array covers 1GHz to 3GHz. The monolithic microwave integrated circuit can emit a scanning signal with fine frequency steps and accurately measure the reflection coefficient and transmission coefficient of the module.

[0027] Each PCB board of the template features a passive, non-contact LC resonant circuit. This circuit is composed of distributed switching parameters, including high-frequency transmission lines, ceramic capacitors, the dielectric constant of the PCB board itself, parasitic capacitance of component pads, and parasitic inductance between traces. The final resonant frequency and Q-factor (quality factor) of the LC circuit are determined by microscopic variations during manufacturing processes, such as PCB layout, component tolerances, and board characteristics. Even on the same production line, no two modules will have completely identical electromagnetic characteristics. This forms the basis for the physically unclonable function of the modules in this application, thus providing anti-counterfeiting capabilities.

[0028] The monolithic microwave integrated circuit sends the reflection and transmission coefficients acquired by the module to the system main control (such as an FPGA or a processor with a dedicated DSP core). The processor runs an existing impedance extraction algorithm to convert the reflection and transmission coefficients into complex impedance data that varies with frequency, thus obtaining the impedance spectrum of the module.

[0029] After obtaining the physical unclonable identity of the module near the system backplane through step S1, such as Figure 1 As shown, the power supply hot-swap safety control method provided in this embodiment proceeds to the following steps: S2, based on the module's physical non-clonable identity, matches the module's type information and historical distributed collaborative pre-charge records from the module feature database, generates an adaptive pre-charge curve, and then diagnoses the health status of the module to be plugged in. Specifically, the matching process includes the following parts: 1. Feature Extraction Based on the extracted impedance spectrum Calculate the corresponding feature vector The vector includes: Main resonant frequency The frequency point in the impedance spectrum where the imaginary part is 0 and the real part is a local minimum.

[0030] Q value at resonant point: ; The preferred bandwidth is 3dB.

[0031] Impedance real and imaginary parts of multiple non-resonant points: Select several fixed frequency points (such as 1.5GHz, 2.0GHz, 2.5GHz) and record their impedance values ​​to characterize the overall impedance characteristics of the module.

[0032] The phase angle curve of the impedance spectrum: The relationship between the phase angle and frequency also contains rich feature information, which can reflect the physical non-cloning identity characteristics of the module.

[0033] 2. Database matching Power system maintenance hierarchical module feature database The first level is: factory record database Each module is scanned for its electromagnetic features under the same conditions when it is tested at the factory, and the feature vector obtained is stored in the factory record database in association with its serial number, model number, production batch, rated parameters, etc.

[0034] The matching process is: when the module is scanned by the system backplane to obtain the feature vector , the system performs a nearest neighbor search in the database, calculates the cosine similarity or Euclidean distance of each in the database with the feature vector obtained by the module. Finally, the associated information of the record with the highest similarity is extracted as the matching result. For example, the matching result is: "this is a 3kW rated power lithium battery module with serial number SN12345678, model number XH123, produced by company A".

[0035] The second level: historical control database After the type information of the module to be inserted is obtained through the matching of the first level, in the second level, the historical control data of the module to be inserted, including the historical distributed cooperative pre-charging record of the module, are further matched according to the matching result of the first level.

[0036] For example, the matching result of the first level is: "this is a 3kW rated power lithium battery module with serial number SN12345678, model number XH123, produced by company A". In the matching of the second level, the record of the hot plug safety control action made by the power system according to the plug-in behavior of "a 3kW rated power lithium battery module with serial number SN12345678, model number XH123, produced by company A" is further matched from the historical control database, including the historical distributed cooperative pre-charging record of the lithium battery module.

[0037] After the matching results of the first level and the second level are obtained, in step S2, the health status of the current module to be inserted also needs to be diagnosed, and the diagnosis method includes the following steps: A1, according to the data sensed by the system backplane and the historical distributed cooperative pre-charging record matched to the module to be inserted, the resonance frequency drift , value drift , low-frequency impedance real part drift and energy dissipation ratio slope of the module to be inserted are calculated; The resonance frequency drift , indicates the current measured value of the main resonance frequency; ​This indicates the factory-set main resonant frequency of the module. A positive value indicates a positive frequency. This indicates that the capacitance value of the capacitor in the pre-charge circuit of the module has decreased.

[0038] A ratio less than 1 indicates increased losses. This indicates the current measured value of the quality factor for the module to be plugged in; This indicates the factory value of the quality factor for this module.

[0039] On the right This indicates an increase in the resistance of the connector (used to connect to the system backplane), fuse, or wiring in the module to be plugged in. This represents the current measured value of the real-valued low-frequency impedance of the module to be plugged in; This indicates the factory-set value of the low-frequency impedance of the module under the same low-frequency conditions. The low frequency referred to here is, for example, the non-resonant frequency point of 1.5GHz.

[0040] Based on the historical distributed collaborative pre-charging records of the module to be plugged in, the current energy dissipation ratio slope of the module to be plugged in is calculated. The method is as follows: For each historical precharge event of the module to be plugged in, calculate the energy dissipation ratio. The calculation formula is:

[0041] in, This indicates the total energy dissipated as heat through the pre-charge resistor and MOS on-resistance in the intended plug-in module during this historical pre-charge. The preferred calculation method is to solve it by integrating the product of the current flowing through the resistor and the resistance value over time. Since... The calculation method is not within the scope of the rights claimed in this application, and therefore will not be specifically explained. This represents the total energy transferred from the system bus to the input capacitor of the module to be plugged in during this historical pre-charge. Similarly, due to... The calculation process is not within the scope of the rights claimed in this application, nor will it be specifically explained.

[0042] This reflects the efficiency of the pre-charging process, the pre-charging process of a new module, its It is close to the ideal calculated value, but as the module ages, It will gradually increase, deviating from the ideal calculated value. Therefore, The changing trend is a direct reflection of the changes in internal losses within the module.

[0043] In this application, in step A2, the health assessment model does not use a single [test / test]. Instead, it analyzes the most recent N (e.g., 10) hot-plug events of the module to be plugged in. Values, and then fit each The slope of the trend over time is used as the slope of the energy dissipation ratio calculated for the proposed plug-in module. A positive, ever-increasing... This is a strong signal that the module's health is deteriorating.

[0044] A2, will , , as well as The data is input into the health assessment model bound to the module to be plugged in, and the model predicts and outputs the health status of the module to be plugged in. The model prediction process includes the following steps: A21, Calculate the capacitor aging factor based on the model input. growth factors Connection resistance growth factor Dynamic performance degradation factor ; In this application, Mainly composed of The calculation yielded:

[0045] growth factors It is a fused feature value. In this application, and They all directly reflect (The increase of parasitic resistance existing inside the capacitor in the pre-charge circuit, which leads to energy loss and heat generation). In this embodiment, by... and We perform a weighted summation to obtain the growth factor after feature fusion. Growth factors The calculation formula is as follows:

[0046] The weighting coefficients were determined through extensive experimental data. The process of determining the nature of the matter is not within the scope of the rights claimed in this application, and therefore will not be described in detail.

[0047] Connection resistance growth factor Directly by After normalization, we get:

[0048] Dynamic performance degradation factor As a comprehensive indicator, by The absolute value and its rate of change together determine whether the ability of the module to be plugged in to respond to system control commands is continuously decreasing.

[0049] In this application, the dynamic performance degradation factor The calculation method is briefly described below: For example, take the 10 most recent time points. Slope value, then for these 10 The slope value is then subjected to linear regression again to fit a new trend line, the slope of which is the slope of the trend line. The rate of change of this value is the acceleration of the degradation of the module's ability to respond to system control commands. The larger the value of this acceleration, the faster the degradation rate and the worse the module's ability to respond to system control commands.

[0050] This will reflect the rate of basic degradation. The acceleration reflecting the rate of degradation is combined into a comprehensive index through a weighted summation method. This comprehensive index is the dynamic performance degradation factor. The weighting coefficients were also obtained through extensive experimental verification and calculation; the specific process will not be explained here.

[0051] In addition, to improve model prediction efficiency, dynamic performance degradation factor With capacitor aging factor growth factors Connection resistance growth factor Similarly, a normalized value is preferred. The preferred normalization method is to set a degradation tolerance threshold and then calculate... The ratio of the degradation tolerance threshold to the degradation tolerance threshold is used as the basis for... The normalized value.

[0052] After step A21, the model prediction process in step A2 transitions to step: A22, calculate the health score corresponding to each health characteristic factor calculated in step A21, then obtain the weight parameters associated with the module to be plugged in and perform weighted fusion. The fusion result is used as the health score of the module to be plugged in. Among them, capacitor aging factor Health score of this health characteristic factor The calculation method is as follows:

[0053] growth factors Health score of this health characteristic factor The calculation method is as follows:

[0054] Connection resistance growth factor Health score of this health characteristic factor The calculation method is as follows:

[0055] Dynamic performance degradation factor Health score of this health characteristic factor The calculation method is as follows:

[0056] , , , This indicates the failure threshold set according to the model of the module to be plugged in.

[0057] Then to , , , Weighted fusion is performed to obtain the health score of the module to be plugged in. .

[0058]

[0059] It's important to note that the weights are not fixed but dynamically adjusted based on the type of module to be plugged in. For example, for lithium-ion battery modules, the health of the parasitic resistance within the input capacitor and the capacitor in the pre-charge circuit—which leads to energy loss and heat generation—is crucial to battery life. , The weighting of this component is higher compared to voltage regulator modules and power distribution modules. For power distribution modules, connection reliability (contact resistance) and dynamic response under high current (performance degradation) are critical; therefore, , The weight of the voltage regulator module is higher than that of the lithium battery module. For the voltage regulator module, a balanced approach is needed, therefore, all weights are relatively evenly distributed. Since the specific weighting of different types of modules is not within the scope of this application, it will not be specifically detailed.

[0060] After calculating the health status of the module to be plugged in through step A22, the model prediction process in step A2 is transferred to step: A23, Assessing Health Is it greater than the preset health threshold? If so, then calculate the confidence level for the health status and proceed to step A24; If not, an alarm will be triggered and the power hot-swap safety control process will be terminated; In this application, for example, the health threshold is set to 80%, when the health level... At that time, the model calculates The confidence level.

[0061] The confidence level is calculated as follows: Set a confidence interval based on the health threshold, for example, set the confidence interval for an 80% health threshold as [70%, 80%]; Then determine , , , Whether all fall within the confidence interval. If so, then... , , , The confidence levels are set to "1" respectively, indicating that the confidence levels are consistent; If not, the confidence level of health scores that do not fall within the confidence interval will be set to "0", indicating that the confidence level is inconsistent.

[0062] After completing the confidence level calculation in step A23, proceed to the following step: A24, Determine whether the confidence level is consistent. If so, then the health level will be... As the output of the model prediction; If not, an alarm will be triggered and the power hot-plug safety control process will be terminated.

[0063] In step S2, the method for generating the adaptive pre-charge nonlinear control curve of the proposed plug-in module includes the following steps: B1, retrieve the voltage-charge nonlinear relationship model of the input capacitor of the module to be plugged in from the module feature database; For an ideal capacitor, the charge Q and the voltage V across it are linearly related. However, real capacitors have effects such as dielectric absorption and capacitance voltage dependence, and their relationship is non-linear. This means that the capacitance C of the input capacitor is not constant, but varies with the voltage.

[0064] In this application, the method for constructing the voltage-charge nonlinear relationship model of the input capacitor of the module to be plugged in is as follows: At the time of module shipment, a charge amplifier or current integrator circuit is used to apply a very slow, linearly increasing voltage to the module's input capacitor (ensuring no charging current surge), while simultaneously measuring the amount of charge Q flowing into the capacitor and the voltage V across the capacitor.

[0065] Plot the measured (V,Q) data points as a curve using a polynomial function such as... To fit the curve, , , The coefficients are denoted by . This polynomial function represents the voltage-charge nonlinear relationship model associated with the proposed plug-in module.

[0066] B2, combining the real-time monitored system bus voltage fluctuation range, constructs a target charge integral function with time as the variable; In this application, the purpose of constructing the target charge integral function is to define an "ideal" pre-charge process to ensure that the start and end of the charging process are very smooth and without any shocks.

[0067] Specifically, it is desirable for the capacitor voltage to rise smoothly from 0V to the system bus voltage without overshoot. Based on the voltage-charge nonlinear relationship model constructed in step B1, with the target voltage... Using y as the independent variable, the value of the voltage-charge nonlinear relationship model can be solved, i.e. Corresponding total charge Therefore, the target charge integral function to be constructed in step B2 is... It's a charge quantity from 0 to... A smooth time function.

[0068] In this embodiment, the target charge integral function is expressed as follows:

[0069] This indicates the initial time 0 at which the pre-charging of the input capacitor begins; Indicates the time when the pre-charging of the input capacitor ends. . This indicates that at time 0, the initial charge of the input capacitor is 0. Indicates termination At that moment, the charge on the input capacitor reaches the target value. . This indicates that at time 0, the initial current of the input capacitor is 0. Indicates termination At that moment, the termination current of the input capacitor is 0. This indicates that at time 0, the initial rate of change of the input capacitor current is 0. Indicates termination At time t, the rate of change of the input current at the termination current is 0.

[0070] This application uses mandatory constraints at time 0 and By taking the zeroth, first, and second derivatives at time points, a smooth charging trajectory with minimal impact was designed. Subsequently, an optimal control algorithm was used to force the actual system behavior to precisely track this ideal trajectory.

[0071] B3. Substitute the target charge integral function into the state space equation of the pre-charge circuit of the proposed plug-in module, solve in reverse the optimal gate voltage timing function of the pre-charge MOS transistor, and perform function fitting to obtain the adaptive pre-charge nonlinear control curve.

[0072] In this embodiment, the state equation is expressed as follows:

[0073]

[0074] The differential representing the state of the current; express The parasitic inductance flowing through the pre-charge circuit at all times The current; The on-resistance of the pre-charge MOSFET in the pre-charge circuit is a non-linear function determined by the gate voltage. When... When the value is small, the pre-charge MOSFET operates in the linear region. Larger; when After exceeding the threshold, It becomes very small and essentially constant.

[0075] This indicates the pre-charge resistor, which is a fixed value.

[0076] This indicates that the circuit resistance opposes changes in current.

[0077] express The input capacitor voltage at any given time is not a measured value, but rather a value derived from the voltage-charge nonlinear relationship model constructed in step B1. ( The voltage-charge nonlinear relationship model is used as the dependent variable in this model, and the independent variable is solved in reverse to obtain the capacitor voltage. In this application, the voltage-charge nonlinear relationship model is embedded into the system dynamics expressed by the state-space equations, so that the model can accurately describe the actual physical characteristics of the input capacitor.

[0078] This illustrates the reaction force of the capacitor voltage on the charging current. The higher the capacitor voltage, the greater the resistance to continued charging.

[0079] Expressing control input Its driving effect on the system. This is the gain coefficient related to the transconductance of the precharged MOSFET.

[0080] It is the differential of the charge state.

[0081] Step B3, the method for reverse-engineering the optimal gate voltage timing function, includes the following steps: B31, pre-charging period Discretized The method involves calculating the optimal gate voltage for the pre-charged MOSFET at each sampling time point within a sampling period (e.g., 1 microsecond per sampling point). The solution method includes the following steps: B311, calculate each sampling time point Input capacitor charge , For the previous Input capacitor charge at time point In addition to the parasitic inductance in the pre-charge circuit exist Current at any moment Integral over the time step; B312, will As the dependent variable of the voltage-charge nonlinear relationship model constructed in step B1, the independent variable of the model is solved in reverse. As in The optimal gate voltage applied to the gate of the pre-charged MOSFET at all times; B32, will each Moment Constructed as a time point As the independent variable, The optimal gate voltage timing function is the dependent variable. Preferably, the optimal gate voltage timing function is a quadratic function.

[0082] The adaptive precharge nonlinear control curve is obtained by curve fitting the optimal gate voltage timing function.

[0083] After completing step S2, diagnosing the health status of the module to be plugged in and generating the adaptive pre-charge nonlinear control curve, as follows: Figure 1 As shown, the power supply hot-swap safety control method provided in this embodiment proceeds to the following steps: S3, the power system takes the contact of the pin of the module to be plugged in with the backplane pin as the instruction, and starts the distributed cooperative pre-charge process based on the adaptive pre-charge nonlinear control curve generated in step S2 for all online modules. The process aims to minimize the instantaneous impact kinetic energy and electromagnetic interference spectrum peak of the entire network, and solves and allocates the optimal drive signal of the pre-charge MOS transistor in each online module in real time. Specifically, the method for initiating a distributed collaborative pre-charging process for all online modules includes the following steps: S31 incorporates the adaptive pre-charge nonlinear control curve generated for the intended plug-in module into the global objective function. ; In this embodiment, the global objective function "0", "These represent the start and end times of the pre-charging period, respectively; Indicates the current deviation term; This represents the electromagnetic interference energy term; These represent the optimization weights for the current deviation term and the electromagnetic interference energy term, respectively. ; , Indicates the first One online module (including the currently plugged-in module); Indicates the number of online modules; Indicates the first The online modules are Real-time current at any given moment; Indicates the first The online modules are The expected current at any given moment. For the newly plugged-in module, It is important to emphasize here that The input capacitor charge obtained through step B31 is used to incorporate the adaptive pre-charge nonlinear control curve generated for the newly plugged-in module into the global objective function. In the middle. As for older modules that are already online, This is the stable operating current instant before the new module is plugged in. The purpose of performing the distributed collaborative pre-charge process is to maintain the current stability of the old modules as much as possible, to avoid disturbances to them, and at the same time to make their currents undergo a controlled, small, and opposite offset to compensate for the disturbances brought about by the new module connection.

[0084] S32, to minimize the global objective function To optimize the objective, the solution for each online module at each sampling time is obtained. The optimal gate voltage control sequence arranged in time order; Assuming the online modules include the newly plugged-in modules. Old modules , In this embodiment, the optimizer will consider countless possible scenarios. Searching and calculation are performed within the combinations.

[0085] The online solution module in this application solves at each sampling time. The method for determining the optimal gate voltage is as follows: For example, a slight increase Gate voltage of the precharged MOSFET This makes it more conductive, but slightly reduces Gate voltage of the precharged MOSFET This slightly reduces its conductivity while providing power to the module. The gate voltage of the precharge MOSFET is It is important to note here that The independent variable of the model is obtained by substituting the input capacitor charge obtained in step B31 as the dependent variable into the voltage-charge nonlinear relationship model constructed in step B1, and solving it in reverse.

[0086] Assuming the old module The stable operating current is 10A immediately before the new module is plugged in, and the old module... The stable operating current is 8A immediately before the new module is plugged in. Assuming... The corresponding current increases to 10.3A, The corresponding current decreases to 7.2A, at which point... The corresponding current can reach the ideal 2A. However, in another scheme, the current can be... The corresponding current is reduced to 9.4A, which will The corresponding current decreases to 7.5A, at which point... The corresponding current can only reach 1.8A, although the second solution is suitable for older modules. , The overall point flow rate is smaller, but because... If the corresponding current cannot reach the ideal 2A, then the second approach will be adopted. By trying various combinations, the goal is to minimize the global objective function. To optimize the objective, the optimal gate voltage control sequence, consisting of the optimal gate voltage of each online module at each sampling time, is finally obtained.

[0087] S33, in At that moment, The optimal gate voltage, calculated at each step, is applied to the gate of the pre-charged MOSFET in the corresponding online module.

[0088] After solving and allocating the optimal gate voltage (optimal drive signal) of the pre-charge MOSFET in each online module during the pre-charge period through step S3, as follows: Figure 1 As shown, the power supply hot-swap safety control method provided in this embodiment proceeds to the following steps: S4 applies the optimal drive signal to the gate of the pre-charge MOS transistor of the corresponding online module to complete the insertion action based on the module to be inserted, and to control the power supply safety of all online modules.

[0089] Preferably, after performing step S4, the following steps are performed: S5 takes the pre-charging of each online module as an instruction, and based on the maximum power capacity function that the access module can provide in real time and the predicted total system load demand, it uses a smart unit multi-party compensation learning algorithm to enable all online modules to autonomously negotiate and quickly converge to the globally optimal and stable power allocation point, and then supply power to each load in the network.

[0090] In this embodiment, the method for predicting the total system load demand includes the following steps: 1. Data preparation and preprocessing 1) Data Acquisition. The system continuously monitors and records the time series of total load power demand. The sampling time interval is, for example, 1 second, to form a one-dimensional data sequence.

[0091] 2) Data preprocessing. For the raw data... Data cleaning is performed to remove outliers and smooth the data using methods such as moving averages or low-pass filtering to suppress high-frequency noise, providing a clean data foundation for subsequent stability analysis.

[0092] 2. Spatial Reconstruction In this application, the purpose of spatial reconstruction is to reconstruct the topology of the potential multidimensional dynamic system that drives the signal from an observable one-dimensional signal (one-dimensional data sequence).

[0093] 1) Calculate key parameters. In this application, the key parameters for spatial reconstruction include time delay. and embedding dimension .

[0094] Time delay The calculation method is as follows: Calculate the autocorrelation function or mutual information function of the original one-dimensional data sequence, and select the delay time corresponding to the first zero crossing of the autocorrelation function or the first minimum value of the mutual information function as the starting point. .

[0095] Embedding dimension The calculation method is as follows: starting from a low-dimensional space, gradually increase the dimension. When "false" nearest neighbors (i.e., points that are not adjacent in the high-dimensional space) disappear as the dimension increases, the corresponding minimum... This is the final determined embedding dimension. There are many existing methods for searching for false nearest neighbors, which will not be detailed here.

[0096] 2) Reconstructing Space Using the time delay of the above solution and embedding dimension The original one-dimensional data sequence Reconstructed dimensional spatial sequence .

[0097] 3. Calculate the probability of prediction In this embodiment, not all signals are unstable; they may be purely random noise, thus requiring a clear prediction of the probability.

[0098] 1) First, calculate the index. The calculation method is as follows: In the reconstructed space, trace the two points with the smallest initial distance. and , and They represent , A spatial sequence of time, observing the evolution exponent of the distance between two points over time. and the largest evolutionary index , recorded as ;

[0099] This represents the distance (preferably the minimum absolute value of the difference) between two infinitely close or smallest points in the reconstructed space at the initial moment. In this application, the initial moment is, for example, set as: when the pin of the module to be plugged in contacts the backplane pin and is pushed forward. A hot-plug event, for example, the current time when the pin of the module to be plugged in contacts the backplane pin is t1, and before time t1, there were... The first hot-plug incident, spanning from the first hot-plug incident to the [number missing]th hot-plug incident. For example, if the cumulative time is T2, then at the initial time... For t1-T2. This represents the distance between two points after time t. For example, the distance between two points at time t1 after time T2. This indicates the magnification or reduction factor of the initial distance. To Take the natural logarithm.

[0100] pass The value of is used to assess whether the total power demand of the system is stable. If the system's total power demand is determined to be stable, then the prediction space for the system's future total power demand is very small. It is sufficient to simply shift the stable total power demand originally provided by the old module to be shared by all online modules after the new module is added. However, when... This indicates that the total power demand of the system is unstable, so it is necessary to predict the future total power demand of the system so that the power allocation of all online modules is no longer a post-event remedy, but a pre-event preparation.

[0101] In this application, the method for predicting the total load power demand of the system is as follows: The system at the current moment status Mapped into the reconstruction space, we get ; Reconstructing the historical trajectory of space middle( (Representing each historical time point, preferably the time point when a hot-plug event was performed), and searching for the current state point. The nearest neighbor (preferably with a value deviation less than a preset deviation threshold) A point, expressed as ; this The next state of each of the nearest neighbors is known, that is... These are known (because historical data is used). Based on these nearest neighbors, a limited prediction model is constructed. For example, fitting the data from the least squares method... arrive The local linear mapping relationship.

[0102] Then Substitute into the established local prediction model In the next step, calculate the predicted state:

[0103] It is 3D vector The first component in is the one in the current... The predicted value of the system load power at the next time step is denoted as . The algorithm ultimately outputs a load power prediction value with one or more prediction steps.

[0104] After obtaining the predicted value of the total power demand of the system load, this application guides all online modules to perform power allocation through a smart unit multi-party compensation learning algorithm. The specific method is as follows: 1. Set up smart units Each online module is considered a smart unit. Each smart unit The status includes the current output power Maximum allowable output power (This is a dynamic value, affected by module temperature, state of charge, health status, etc.) State of Charge Module temperature Health status One or more of them.

[0105] The intelligent unit's decision-making action is set as follows: adjust its own power output value. ; Configure intelligent unit power compensation: Adjust the power output compensation value based on the predicted total system load power demand. .

[0106] 2. Construct the power compensation function In this application, the constructed power compensation function is expressed as follows:

[0107] , These represent local compensation and global compensation, respectively. , These represent the proportion and weight of local compensation and global compensation in the overall compensation process, respectively.

[0108] In this application,

[0109] Indicates the current The total power that all online modules can provide at any given time; Indicates the predicted The total system load power requirement at any given time; express Real-time voltage value on the main power bus of the system; Indicates the rated value of the bus voltage; This represents the weighting coefficient. In this embodiment, Choose a moderate value, such as 1.5 or 1.2, to find a balance between the two goals of maintaining voltage stability and achieving forward-looking power balance. The method for determining the value requires extensive application verification of the performance of the multi-party compensation learning algorithm for intelligent units, which is a very complex process. The calculation is not within the scope of the rights claimed in this application, and therefore will not be specifically explained.

[0110]

[0111] Indicates the weight of motion smoothness; Indicates load rate weight; Indicates the weight of the state of charge; It should be noted here that... For steps A21-A24, the first The health status is calculated by an online module.

[0112] 3. Consensus and Convergence In this application, different online modules aim to maximize the global compensation in the power compensation function, while using the local compensation in the power compensation function as a constraint. At each adjacent time step... (This is a dynamically adjusted value, not a fixed value) The power output is adjusted by the adjustment amount; for example, for online modules... Output power at time Adjusted to ; Then, determine The first global compensation calculated below is compared Is the deviation of the second global compensation calculated below less than a preset deviation threshold? If so, then for this moment With this online module Output power at time The summation is performed to determine the final, globally optimal, and stable power allocation point for the online module; If not, then adjust. Find the value of , and then return to step D2; .

[0113] For example, Compared to the earlier time The power adjustment at time is The online module is in The power output at any given time is When the judgment is made in At all times The global compensation adjusted by the adjustment amount (defined as the first global compensation) and in At all times The deviation (preferably the absolute value of the difference) of the global compensation after adjustment (defined as the second global compensation) is considered to be the deviation of the second global compensation. Power compensation for each online module is used to achieve a globally optimal and stable allocation point.

[0114] This application, through steps D1-D3, combines the maximum power supply capability, health status, and prediction results of the total load power demand of each online module to achieve dynamic power allocation for each online module. This changes the system from a "passive response" to reduce disturbances when plugging and unplugging modules to a system where each online module, through collaborative autonomy, steadily achieves globally optimal and stable power allocation.

[0115] It should be stated that the above-described specific embodiments are merely preferred embodiments and technical principles applied in this application. Those skilled in the art should understand that various modifications, equivalent substitutions, and variations can be made to this application. However, such variations, as long as they do not depart from the spirit of this application, should be within the scope of protection of this application. Furthermore, some terminology used in this application's specification and claims is not limiting but merely for ease of description.

Claims

1. A power supply hot-swappable safety control method, applied in scenarios where at least one of a lithium battery module, a power distribution module, and a voltage regulator module is plugged into or unplugged into a power supply system, characterized in that... Including the following steps: S1, when the module approaches the system backplane and reaches a preset threshold distance, the system backplane non-contactly senses the module's resonant frequency and internal equivalent impedance spectrum, using this as a physical, non-cloning identifier to characterize the module type. S2, based on the physical unclonable identity identifier, match the module type information and historical distributed collaborative pre-charging records from the module feature database, generate an adaptive pre-charging nonlinear control curve, and then diagnose the health status of the module to be plugged in. S3, the power system initiates a distributed cooperative pre-charging process based on the adaptive pre-charging nonlinear control curve for all online modules when the pin of the module to be plugged in contacts the backplane pin. The distributed cooperative pre-charging process aims to minimize the instantaneous impact kinetic energy and electromagnetic interference spectrum peak of the entire network, and solves and allocates the optimal drive signal of the pre-charging MOS transistor in each online module in real time. S4, apply the optimal drive signal to the gate of the pre-charge MOS transistor of the corresponding online module to complete the insertion action based on the module to be inserted, and control the power supply safety of all online modules.

2. The power supply hot-swap safety control method according to claim 1, characterized in that, It also includes the following steps: S5 takes the pre-charging of each online module as an instruction, and based on the maximum power capacity function that the access module can provide in real time and the predicted total system load demand, it uses a smart unit multi-party compensation learning algorithm to enable all online modules to autonomously negotiate and quickly converge to the globally optimal and stable power allocation point, and then supply power to each load in the network.

3. The power supply hot-swap safety control method according to claim 1, characterized in that, In step S2, the method for diagnosing the health status of the module to be plugged in includes the following steps: A1. Based on the data sensed by the system backplane and the historical distributed cooperative pre-charging records matched with the intended plug-in module, calculate the resonant frequency drift of the intended plug-in module. , Value drift Low-frequency impedance real part drift and energy dissipation ratio slope ; A2, will , , as well as The data is input into the health assessment model bound to the proposed plug-in module, and the model predicts and outputs the health status of the proposed plug-in module.

4. The power supply hot-swap safety control method according to claim 3, characterized in that, The calculation method includes the following steps: A11, calculate the energy dissipation ratio for each historical pre-charge event of the module to be plugged in. , This indicates the total energy dissipated as heat through the pre-charge resistor and MOS on-resistance in the proposed plug-in module during this historical pre-charge. This represents the total energy transferred from the system bus to the input capacitor of the module to be plugged in during this historical pre-charge. A12, Fitting the N most recent hot-plug events of the proposed plug-in module. The slope of the value as a function of time is used as the slope of the energy dissipation ratio. .

5. The power supply hot-swap safety control method according to claim 3, characterized in that, The model prediction process in step A2 includes the following steps: A21, Calculate the capacitor aging factor based on the model input. growth factors Connection resistance growth factor Dynamic performance degradation factor ; A22, calculate the health score corresponding to each health characteristic factor calculated in step A21, then obtain the weight parameters associated with the proposed plug-in module and perform weighted fusion. The fusion result is used as the health degree of the proposed plug-in module. A23, determine whether the health status is greater than a preset health status threshold. If so, then calculate the confidence level of the health status and proceed to step A24; If not, an alarm will be triggered and the power hot-swap safety control process will be terminated; A24, Determine whether the confidence level is consistent. If so, the health status will be used as the model prediction output. If not, an alarm will be triggered and the power hot-plug safety control process will be terminated.

6. The power supply hot-swap safety control method according to claim 1, characterized in that, Step S2, the method for generating the adaptive pre-charge nonlinear control curve includes the following steps: B1, retrieve the voltage-charge nonlinear relationship model of the input capacitor of the module to be plugged in from the module feature database; B2. Combining the real-time monitored system bus voltage fluctuation range, and using the voltage-charge nonlinear relationship model constructed in step B1, construct a target charge integral function with time as the variable. B3. The target charge integral function is used as the boundary condition of the state space equation of the pre-charge circuit of the proposed plug-in module. Based on the state space equation, the optimal gate voltage timing function of the pre-charge MOS transistor is solved in reverse. The adaptive pre-charge nonlinear control curve is obtained by function fitting.

7. The power supply hot-swap safety control method according to claim 6, characterized in that, Step B3, the method for reverse-engineering the optimal gate voltage timing function, includes the following steps: B31, pre-charging period Discretized The method involves calculating the optimal gate voltage for the pre-charged MOS transistor at each sampling time point within a sampling period. The solution method includes the following steps: B311, calculate each sampling time point Input capacitor charge , For the previous Input capacitor charge at time point In addition to the parasitic inductance in the pre-charge circuit exist Current at any moment Integral over the time step; B312, will As the dependent variable of the voltage-charge nonlinear relationship model constructed in step B1, the independent variable of the model is solved in reverse. As The optimal gate voltage applied to the gate of the pre-charged MOSFET at all times; B32, will each Moment Constructed as a time point As the independent variable, The optimal gate voltage timing function is the dependent variable.

8. The power supply hot-swap safety control method according to claim 1, characterized in that, The distributed cooperative pre-charging process in step S3 includes the following steps: S31, incorporate the adaptive pre-charge nonlinear control curve generated for the proposed plug-in module into the global objective function. ; S32, to minimize the global objective function To optimize the objective, the solution is obtained for each online module at each sampling time. The optimal gate voltage control sequence arranged in time order; S33, in At that moment, The optimal gate voltage, calculated at each step, is applied to the gate of the pre-charged MOS transistor in the corresponding online module.

9. The power supply hot-swap safety control method according to any one of claims 2-8, characterized in that, In step S5, the method for predicting the total power demand of the system includes the following steps: C1, continuously monitor and record the total load power demand time series. , It is a one-dimensional data sequence, and then the one-dimensional data sequence is reconstructed into dimensional phase space sequence ; C2, tracking the point with the smallest initial distance. and The evolution index of distance over time And extract the largest evolution index. , recorded as ; C3, judgment Is it greater than 0? If so, the total power demand of the system is determined to be unstable, and then proceed to step C4; If not, then terminate step S5; C4, predict the total load power demand of the system, and use the prediction result as the input of the intelligent unit multi-party compensation learning algorithm.

10. The power supply hot-swap safety control method according to any one of claims 2-8, characterized in that, The computation process of the intelligent unit multi-party compensation learning algorithm includes the following steps: D1, build the current... for each online module The power compensation function at any given time incorporates the predicted total load power demand of the system, the maximum power capacity function, and the health status into the power compensation function. D2, each of the online modules aims to maximize the global compensation in the power compensation function, with local compensation in the power compensation function as a constraint. At each adjacent time step... Adjust its own power output to adjust the amount of adjustment; D3, judge The first global compensation calculated below is compared Is the deviation of the second global compensation calculated below less than a preset deviation threshold? If so, then for the current Moment With the online module Output power at time The summation is performed, and the summation result serves as the final power allocation point that the online module converges to the globally optimal and stable point. If not, then adjust. Find the value of , and then return to step D2; 。

Images