Multi-phase power converters

Abstract

A multi-phase power converter, comprising: a regulator; a value supply system arranged to collect at least one operating point of the power converter; and a predictor for determining an updated phase state to activate or deactivate each phase (111, 112, 113, …) during further operation of the power converter. The updated phase state is determined using a process based on at least one collected operating point and predictor parameters obtained from machine learning processing.

Description

Technical Field

[0001] The present disclosure relates to multi-phase power converters. Background Art

[0002] Electrical power converters (particularly DC-DC power converters) are widely used due to their high efficiency and the small amount of area / volume consumed by such converters. Among them, multi-phase DC-DC power converters are particularly suitable for providing large currents at low voltages, such as the currents required by highly integrated electronic components such as microprocessors, graphics processors, and network processors.

[0003] Generally, a conventional multi-phase power converter generally includes a plurality of converter branches referred to as phases. These phases are generally connected in parallel to provide corresponding phase currents to a common load. Therefore, the output current provided to the load by the multi-phase power converter is the sum of the phase currents.

[0004] The current generation of processors (such as CPUs or GPUs) have a very high ratio of peak power to idle power requirements. The DC-DC power converters used to power such processors then need to perform well under a variety of load profiles. To this end, it is known to dynamically adapt the number of phases of the power converters that are simultaneously activated according to the output power desired at one time, so as to optimize the performance of the converters regardless of the level of output power. This adaptation of the number of activated phases is sometimes referred to in the art as phase addition and drop, or more generally as phase shedding.

[0005] So-called automatic stabilization involves this change in the number of activated phases controlled by comparing the output current with a threshold. But automatic stabilization may not be adapted to some changing behaviors of the output current, such as switching the output current. In fact, the number of activated phases can be reduced, while energy is still provided from one phase that has just dropped. In this way, the system may become unstable. In addition, automatic stabilization usually generates a delay (called rebalancing time), requiring the output current to drop below the threshold before reducing the number of activated phases. This avoids phase drops for very short periods of time, but this reduces the benefits of phase shedding and may even prevent adaptation to the number of activated phases if the output current changes too quickly.

[0006] When certain requirements apply to the output current (such as large voltage slew rates), automatic phase stabilization is not applicable. It may be necessary to increase the number of active phases to meet such requirements, although consideration of power savings alone would result in operation with the minimum number of active phases. Summary of the invention

[0007] In contrast to conventional techniques, embodiments herein include novel multi-phase power converters that are capable of operating in an optimized manner over a greater number of output conditions and requirements, including at least some of those cited above.

[0008] To achieve this or any other purpose, embodiments of the present invention include an electric power converter including multiple phases for providing output current and output voltage to a load, the output current and output voltage being derived from phase output current and phase output voltage provided by phases (phase, also referred to as "phase units"), respectively. An embodiment of the power converter also includes:

[0009] - a regulator for selectively activating each phase during operation of the power converter;

[0010] a value supply system for collecting at least one operating point, each operating point being related to an operating instant of the power converter and each operating point comprising measured values ​​of one or more input parameters of the power converter among an input current, an input voltage, a phase input current, a phase input voltage and / or one or more output parameters of the power converter among an output current, an output voltage, a phase output current, a phase output voltage for the corresponding operating instant, and / or each operating point further comprising at least one value of a target output voltage assigned to the same operating instant of the power converter; and

[0011] - a predictor (implemented as hardware and / or software) for providing, for each phase of the plurality of phases, a respective updated phase state indicating whether the phase is activated, the updated phase state controlling activation of the respective phase.

[0012] According to embodiments herein, the predictor operates to determine an update phase state using processing based on at least one operating point collected by a value provisioning system (ie, monitor and value generator) and based on predictor parameters obtained from a machine learning process.

[0013] Thus, embodiments herein include new ways to implement phase shedding in an electric power converter by controlling the number of active phases using a machine learning process.

[0014] Since the updated phase state is determined based on the measured values ​​of one or more of the input current, input voltage, output current, output voltage and the measured values ​​associated with each phase, the actual state of the converter input power supply and the converter load is taken into account during the phase shedding. In addition, when these states change over time, the interlocking operation of the value supply system and the predictor allows the phase state to be automatically modified to fit the number of activated phases to the new conditions. In particular, the implementation of a machine learning process for updating the phase state allows for improved fitting of the number of phases over a wide range of operating scenarios of the load, including rapid switching of the output current between values ​​corresponding to the number of active phases with a phase difference greater than 2 (if each of these output current values ​​is continuously maintained).

[0015] Furthermore, implementing a machine learning process also allows for automatic, rapid, and possibly periodic updating of the number of phases used to provide the desired power to the load.

[0016] According to one embodiment, the predictor is operative to provide, for each phase (when the corresponding updated phase state indicates that the phase is to be activated), an updated value of the operating parameter of the phase. The updated value of the phase operating parameter is transmitted to the regulator so that the updated value activates the phase of interest in a manner that matches the updated value of the phase operating parameter. With this improvement, optimal operation of each activated phase can be obtained.

[0017] According to a further embodiment, the predictor is operative to determine an updated phase state for each phase based on a plurality of operating points associated with successive operating instants of the power converter, the plurality corresponding to a fixed number of operating points. In other words, in one embodiment, the predictor provides an updated phase state based on a history comprising a fixed number of operating points. By this improvement, the predictor optimizes the operation of the power converter to a greater extent, in particular by predicting changes occurring in the operating scenario of the load.

[0018] When the phase state is determined based on multiple consecutive operating points, the predictor can be configured to implement a recurrent neural network, in particular a long short-term memory network, so that each time another operating point is provided to the predictor by the value supply system, the further operating point is added to the multiple operating points just used before in a FIFO queue manner to obtain an updated multiple operating points, which will be used to publish another updated phase state for each phase.

[0019] In one embodiment, one or more of the following additional features may be advantageously implemented, respectively, in combination with several of these additional features:

[0020] -The power converter can be configured as a DC-DC power converter or an AC-DC power converter;

[0021] - the predictor may be configured to include a look-up table for storing labeled training data and to include means for selecting one of these labeled training data as the closest neighbor to at least one operating point;

[0022] - the predictor may be configured to implement at least one calculation step of regression type in a calculation sequence for issuing an updated state for each phase from at least one operating point;

[0023] -The predictor can be configured to operate in a feed-forward artificial intelligence manner;

[0024] - the predictor may be configured to operate as a neural network, in particular to operate as a single layer neural network; and

[0025] -The predictor can be configured as a neuromorphic chip.

[0026] Other embodiments herein enable a rapid increase in the output current of a power converter while avoiding activating too few phases or too few sets of phases to produce the converter output current for powering a load. To this end, the power converter may further include a module operable to determine a minimum number of phases to be activated based on the measured values ​​of the converter output current for each of at least one operating point collected. The power converter may then be adapted so that the updated phase state to be implemented by the regulator results in a number of active phases at least equal to the minimum number of phases to be activated.

[0027] According to a further embodiment, a second aspect of the present disclosure provides a method for performing electric power conversion using a multi-phase power converter, for providing an output current and an output voltage to a load, the output current and the output voltage being derived from phase output currents and phase output voltages respectively provided by phases, the method comprising:

[0028] - Using a regulator to selectively activate each phase during power conversion;

[0029] - collecting at least one operating point occurring during the power conversion, each collected operating point being related to an operating instant during the power conversion and comprising measured values ​​of one or more input parameters of the power converter among an input current, an input voltage, a phase input current, a phase input voltage and / or one or more output parameters of the power converter among an output current, an output voltage, a phase output current, a phase output voltage for the operating instant, and at least one value of a target output voltage of the power conversion at the operating instant assigned to the operating point; and

[0030] - Using the predictor, providing an updated phase state for each respective phase of the plurality of phases, the updated phase state indicating whether the respective phase is activated.

[0031] According to further embodiments, each updated phase state is determined using a process based on at least one collected operating point and based on predictor parameters that have been obtained from a machine learning process.

[0032] Embodiments herein may also include the following preliminary steps / 1 / to / 3 / performed during the machine learning process:

[0033] / 1 / Gathering labeled training data, the labeled training data includes a training operating point for each phase and a corresponding associated phase state;

[0034] / 2 / using the labeled training data to train a machine learning model of the predictor in order to obtain predictor parameters to be used by the predictor for inferring each new updated phase state; and

[0035] / 3 / Transfer predictor parameters to the predictor.

[0036] The power conversion is then operated using the predictor parameters transmitted in step / 3 / .

[0037] Preferably, step / 2 / may be performed using a computing device external to the power converter. The computing device may then be disconnected from the power converter, so that the power converter can perform power conversion without being connected to the computing device.

[0038] Other method embodiments herein include the following additional steps:

[0039] - determining a minimum number of phases to be activated based on the collected measurements of the converter output current for each of the at least one operating point; and

[0040] - If the minimum number of phases to be activated is greater than the number of valid phases resulting from the updated phase states issued by the predictor, correcting at least one of the updated phase states issued by the predictor so as to meet the minimum number of phases to be activated.

[0041] Then, further updated phase states of the plurality of phases implemented by the regulator for performing the power conversion results in the number of active phases being at least equal to the minimum number of phases to be activated.

[0042] According to further embodiments, the power conversion performed according to the embodiments herein may be implemented for supplying electrical energy to a load forming part of a data center or server farm. It may be implemented for supplying electrical energy to a microprocessor, a graphics processor, or a memory bank. Such a microprocessor or graphics processor may itself form part of a data center or server farm supplied with electrical energy according to the embodiments herein. Alternatively, the power conversion performed according to the embodiments herein may be a first power conversion stage for supplying electrical energy to a downstream power converter.

[0043] It should be noted that any resource discussed herein (such as a predictor, regulator, etc.) may include one or more computerized devices, mobile communication devices, servers, base stations, wireless communication devices, communication management systems, controllers, workstations, user equipment, handheld or laptop computers, etc., to perform and / or support any or all method operations disclosed herein. In other words, one or more computerized devices or processors may be programmed and / or configured to operate as explained herein to perform the different embodiments described herein.

[0044] Other embodiments herein include software programs to perform the steps and operations outlined above and disclosed in detail below. One such embodiment includes a computer program product, including a non-transitory computer-readable storage medium (i.e., any computer-readable hardware storage medium), on which software instructions are encoded for subsequent execution. When executed in a computerized device (hardware) having a processor, the instructions program and / or cause the processor (hardware) to perform the operations disclosed herein. This arrangement is typically provided as software, code, instructions and / or other data (e.g., data structures) arranged or encoded on such as optical media (e.g., CD-ROM), floppy disks, hard disks, memory sticks, storage devices, etc., or other media (such as firmware in ROM, RAM, PROM, etc.), or provided as an application specific integrated circuit (ASIC), etc. Software or firmware or other such configurations may be installed on a computerized device to enable the computerized device to perform the techniques explained herein.

[0045] Accordingly, embodiments herein are directed to methods, systems, computer program products, etc., that support the operations discussed herein.

[0046] One embodiment includes a computer readable storage medium and / or system having instructions stored thereon to provide power conversion. When executed by computer processor hardware, the instructions cause the computer processor hardware (such as one or more processor devices located in the same location or located in different locations) to: receive a currently collected sample of an operating setting of a power converter; convert the currently collected sample of the operating setting of the power converter into control information, the control information being a machine learning control response assigned to a previously stored sample of the operating setting of the power converter; and output the control information to control a plurality of phases.

[0047] For the sake of clarity, the order numbers of the above steps are added. It should be noted that any process steps discussed herein can be performed in any suitable order.

[0048] Other embodiments of the present disclosure include software programs and / or corresponding hardware to perform any method embodiment steps and operations summarized above and disclosed in detail below.

[0049] It should be understood that the systems, methods, apparatus, instructions on computer-readable storage media, and the like discussed herein may also be strictly implemented as software programs, firmware, as a mixture of software, hardware, and / or firmware, or as separate hardware (such as within a processor (hardware or software) or within an operating system or within a software application).

[0050] As discussed herein, the techniques herein are well suited for providing more efficient use of wireless services to communication devices. However, it should be noted that the embodiments herein are not limited to use in such applications, and the techniques discussed herein are also well suited for other applications.

[0051] In addition, it should be noted that although each of the different features, techniques, configurations, etc. herein may be discussed in different places of the present disclosure, each concept may be optionally implemented independently of each other or combined with each other where appropriate. Therefore, one or more inventions described herein may be implemented and viewed in many different ways.

[0052] In addition, it should be noted that the preliminary discussion of the embodiments herein (a brief description of the embodiments) purposefully does not specify every embodiment and / or incremental novel aspect of the invention disclosed or claimed. Instead, this brief description only presents general embodiments and corresponding points of novelty relative to conventional techniques. For other details and / or possible viewpoints (combinations) of the invention, the reader will be directed to the detailed implementation section (which is an overview of the embodiments) discussed further below and the corresponding drawings of the disclosure.

[0053] These and other features of the invention will now be described with reference to the accompanying drawings, which relate to preferred but non-limiting embodiments of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a diagram illustrating elements of a multi-phase power converter according to embodiments herein.

[0055] Figure 2 It is shown that according to the embodiment of this article, Figure 1 An example diagram of a calculation sequence implemented in a predictor within a multi-phase power converter.

[0056] Figure 3 is an example diagram illustrating the operation of a multi-phase power converter according to embodiments herein.

[0057] Figure 4 is an example diagram illustrating a mapping of current operating settings of a power converter to appropriate control information to achieve a desired control response according to embodiments herein.

[0058] Figure 5is an example diagram illustrating mapping of current operating settings of a power converter to multiple sets of control information and deriving control information from the multiple sets to achieve a desired control response according to embodiments herein.

[0059] Figure 6 is an example diagram illustrating the use of logic to derive control information to control a power converter according to embodiments herein.

[0060] Figure 7 is a diagram illustrating an example computer architecture for performing one or more operations according to embodiments herein.

[0061] Figure 8 is an example diagram illustrating a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] For the sake of clarity, components and elements that are not directly related to the embodiments herein are not described below, and it is assumed that those skilled in the art know how to implement such components and elements.

[0063] For the purpose of illustration but not limitation to this type of embodiment, the embodiments herein include a DC-DC power converter. However, it should be understood that the power converter circuit and corresponding components can be implemented in any type of power converter, such as AC-DC, AC-AC, DC-AC, etc.

[0064] The DC-DC power converter according to the embodiments of the present invention can be used to provide power to one or more loads such as a computer motherboard, but preferably provides power to a processor in a load point configuration, for example. For this configuration, one converter is dedicated to one processor and is located next to it on a common printed circuit board.

[0065] The converter according to the embodiments of the present invention comprises a plurality of phases connected in parallel between the input of the converter and the load (i.e., in this example, the processor to which power will be provided). Each phase may comprise one or more switching devices, each of which produces a connected state during an on-cycle and an isolated state during an off-cycle. Each switching device is operated by a control signal (e.g., a PWM (pulse width modulation) signal or a PFM (pulse frequency modulation) signal issued by a regulator). Preferably, one regulator is common to all switching devices of one converter.

[0066] The regulator can be any type of regulator, including a proportional regulator, an integral regulator, a differential regulator, a proportional-integral regulator, an integral-differential regulator, a proportional-differential regulator, a proportional-integral-differential regulator known as a PID regulator, a regulator implementing at least one higher-order component for controlling power conversion, a delta regulator, a delta-sigma regulator, a differential regulator, etc. For the present invention, it is only necessary that the regulator can selectively activate or deactivate each phase during power conversion. Such phase activation or deactivation can be implemented by the regulator through a suitable control signal transmitted to the switching device of the phase. For example, such a control signal can alternately specify a connection state and an isolation state for the switching device of the current valid phase, and can specify an isolation state only for all switching devices of the current invalid phase.

[0067] During operation, the converter receives an input current and an input voltage from an external DC power source, denoted as I input and V input To implement the embodiments of the present invention, the input current I may be collected repeatedly (eg, every n-cycle operation cycle of the converter, where n is a non-zero fixed integer). input and input voltage V input The measured sample values ​​(different sample times).

[0068] The converter converts the input current I input and input voltage V input Converted into output current I transmitted to the load output and output voltage V output The output current I can also be collected for the same operating instant (same time point) output and output voltage V output The measured value.

[0069] Other values ​​may also be measured, or alternatively, depending on the design of the converter, only some of the above mentioned converter input / output voltages / currents may be measured. These other values ​​may be related to the current supplied to or sourced by one of the phases, respectively, referred to as phase input / output currents and denoted as I phase_input or I phase_output Similarly, the voltage supplied to or generated by one of the phases may be referred to as the phase input / output voltage and denoted as V phase_input or V phase_output This phase input / output current / voltage value can also be compared with the converter input value I input and Vinput and the converter output value I output and V output Some or all of the combinations can be used for any phase.

[0070] The collection of some or all of these measured values ​​is performed by a value provision system. The value provision system collects the measured values ​​relating to the same operating instant of the converter into a set of values ​​called operating points. Each operating point is further completed by the value provision system with a target output voltage also relating to the same operating instant as the measured value of the operating point. The regulator uses a voltage denoted as V target The target output voltage is used to generate a control signal so that the output voltage V actually generated by the converter output Close to the target output voltage V target . Target output voltage V target The continuous values ​​of allow controlling the variation of the instantaneous output voltage supplied to the load, in particular depending on the active or idle cycle of the module inside this load. They also allow controlling the converter output during transient cycles (between active and idle cycles).

[0071] The value supply system transmits each operating point to the predictor, which determines therefrom the state to be implemented by the regulator for each phase. The state indicates whether the associated phase is to be activated or deactivated by the regulator. For example, the predictor can transmit the phase state to the regulator in the form of a binary word, each bit of which is associated with a phase, a value of 0 for the bit indicating that the phase is deactivated, and a value of 1 indicating that the phase is activated. Thus, such a control word has the following form: (0 / 1)1, (0 / 1)2, (0 / 1)3, ..., where (0 / 1)1 is a bit value dedicated to specifying the state of the first phase, (0 / 1)2 is a bit value dedicated to specifying the state of the second phase, (0 / 1)3 is a bit value dedicated to specifying the state of the third phase, etc. The predictor transmits the determined phase states to the regulator of the converter so that the regulator implements these phase states. For example, this implementation of the phase states can be performed by controlling the switching devices of all phases from the moment after the state is received. In this way, the predictor sets the number of phases to be activated at a time during power conversion.

[0072] According to a refinement of the invention, the predictor may also determine one or more phase operating parameters for each phase that is activated. The phase operating parameters may specify the operation of the active phase. For example, the phase operating parameter may be a target value of a phase output current to be generated by the phase, a target output current, etc. The regulator then controls the switching device of the phase so that its phase output current matches the target value of the corresponding operating parameter associated therewith. Specifically, the use of a phase operating parameter indicating a reduced target value of the phase output current may be used to gradually reduce the phase of interest. In this way, the phase operating parameter may be used to indicate a phase state, with an intermediate value between a state indicating that the phase is active and an opposite state indicating that the phase is inactive.

[0073] exist Figure 1, reference numeral 10 denotes a power converter, reference numeral 20 denotes a power source connected to the input of the converter 10, and reference numeral 30 denotes a load connected to the output of the converter 10. In the specific example embodiment described herein, the power source 20 is of a DC type, and the load may be a microprocessor, a memory, a laptop computer, a smart phone, a tablet computer, an LED bulb, a TV, etc. Reference numerals 111, 112, 113, ... denote respective phases of the converter 10, whether or not their numbers are greater than 1, and each reference numeral 12 denotes a switching device within each phase 111, 112, 113, ... The internal structure of each phase 111, 112, 113, ... is shown in FIG. Figure 1 1 and 2. The switching device 12 is not shown in the figure and can be of any type known in the art. For example, it can be a buck converter type. For the clarity of the drawings, only one switching device 12 is shown per phase. Other reference numerals are:

[0074] 13: Regulator

[0075] 14: Predictor

[0076] 15: Value provisioning system 15 (sensors or monitors that generate data), although the corresponding components are distributed in several places in the figure

[0077] The regulator 13 , the predictor 14 and the value supply system 15 are part of the power converter 10 together with the phases 111 , 112 , 113 , . . .

[0078] The value supply system 15 may include one or more voltage sensors and / or one or more current sensors, such as conventional voltage and / or current sensors, for example DC resistors for sensing current. These sensors may be combined with a sample holding unit and an analog-to-digital converter to emit at least some measurement values ​​V corresponding to the usual operating instants of the converter 10. input ,I input 、V output ,I output 、V phase_input ,I phase_input 、V phase_output ,I phase_output Advantageously, the sample period may be a multiple of the switching period of the phases 111, 112, 113, ..., but the sample period may also be selected depending on the converter application, for example to update the phase state fast enough with respect to load variations. The sample period may also be selected depending on the power consumption caused by each value measurement and each update of the phase state.

[0079] V input ,I input 、V output ,I output 、V phase_input ,Iphase_input 、V phase_output ,I phase_output At least some of the target output voltage V target The measured value of is transmitted by the value supply system 15 to the regulator 13, so that the regulator operates in a manner known before the invention.

[0080] According to the embodiments of this article, compared with the conventional technology, V input ,I input 、V output ,I output 、V phase_input ,I phase_input 、V phase_output ,I phase_output At least some of the target output voltage V target The measured values ​​of are also transmitted to a predictor 14 which determines the phase state to be implemented by the regulator 13 .

[0081] The operation of predictor 14 will now be described.

[0082] In one embodiment, the predictor 14 is a memory set similar to a FIFO queue for storing a fixed number of operating points related to consecutive operating instants of the converter 10. For example, at the end of each sample time, a further operating point is issued by the value supply system 15. This further operating point is stored in the entry cell of the memory set similar to the FIFO queue, and all previously stored operating points are offset by one cell in the queue towards the last storage cell. One of the operating points stored in the last storage cell of the queue is discarded. The entire content of the memory set (data collected at multiple time instants) is used to determine the next phase state. This allows anticipating events such as load changes, voltage changes, and any event that may occur, by implementing in advance multiple valid phases suitable for such events.

[0083] In order to predict the phase state in a manner suitable for each application, the predictor 14 implements an algorithm called a machine learning model. This machine learning model can operate within the predictor 14 as embedded software or directly in hardware, or any combination of the two. This allows the same silicon chip to be used for any application of the converter 10. In particular, the use of a neuromorphic chip that implements a spiking neural network for the predictor 14 enables a very energy-efficient hardware implementation of the machine learning model.

[0084] In one embodiment, a simple machine learning model (such as a lookup table model) for the predictor 14 includes storing within the predictor a plurality of operating points with associated phase states for all phases of the converter 10. A series of consecutive operating points may be stored with associated phase states for all phases. Then, each time the value supply system 15 provides a series of actual operating points, an algorithm (such as a nearest neighbor algorithm) determines which of the stored series of operating points (of the plurality of series of previously stored sample data associated with the power converter circuit) is closest to the series of actual operating points. More specifically, in one embodiment, the predictor matches (and / or maps) the currently measured operating settings for the multiple phases to the closest matching previously measured set of operating settings (identified in the plurality of operating sets of measured settings for the multiple phases) to determine how to generate an appropriate machine learning control response. The embodiments herein assume a certain degree of repeatability in the conditions associated with the multiple phases, and the desired response may be predicted based on the previously learned settings and the corresponding mapped machine learning control response to produce an output voltage within a desired voltage range.

[0085] The difference between the actual operating point sequence (the currently collected data set or operating settings 210) and any of the stored operating point sequences may be calculated using any specification known in the art. Figure 2 As shown, for example, a criterion may be selected to weight the operating points differently in different calculations (e.g., applying a weighting value W11 to V input(0) , apply weighted value W12 to V input(1) , apply weighted value W13 to V input(2) , ...apply weighted value W21 to I input(0) , apply the weighted value W22 to I input(1) , weighted value W23 is applied to I input(2) , ...apply weighted value W31 to V phase_output(0) , apply the weighted value W32 to V phase_output(1) , apply weighted value W13 to V phase_output(2) , and so on).

[0086] Then, the phase state to be implemented (such as a suggested control setting indicating the number of phases to be activated) is the state associated with the closest one of the stored operating point sequences. For this implementation, the stored operating point sequences with associated phase states can be recorded in a lookup table (a plurality of previously stored operating settings of the power converter circuit and the corresponding control responses) stored inside the predictor 14. They constitute the so-called labeled training data, as well as the predictor parameters used by the predictor 14 to infer each new set of updated phase states. This implementation of the invention is more suitable when the converter 10 has to consider a small number of operating scenarios.

[0087] Another possible machine learning model may be based on regression, classification methods and / or may use neural networks. Such a regression-based implementation may allow avoiding significant jumps in the number of active phases (e.g., reducing the number of phases from, for example, 8 active phases to 4 active phases too quickly) that may result from the above-described nearest neighbor data implementation. Figure 2 The minimum computational structure to be implemented in the predictor 14 for this regression-based implementation is shown in FIG. In one embodiment, it is generally referred to as a perceptron of the linear classifier type. To obtain the next state to be implemented by the regulator 13 for each of the phases 111, 112, 113, ..., for some or all of the phases V input ,I input 、V output ,I output and V phase_input ,I phase_input 、V phase_output ,I phase_output All measured values ​​of at least some of the target output voltages V for all operating points stored in a FIFO queue memory set (such as storing the current operating setting 210) target Multiplied by a predetermined weight and added together to form a predetermined deviation.

[0088] The result of this combination is then used as an input parameter for an activation function specific to each phase. The result of the activation function is the next state to be implemented for this phase by the regulator 13. In one embodiment, each computational structure of this type is a so-called feedforward neuron, and a separate neuron is dedicated to the activation of each phase controlling the corresponding current operating state of the power converter circuit.

[0089] exist Figure 2 In the example, weights1 and bias1 are the combinations of the measured value and the target output voltage associated with phase 111 (V input Value, I input Value, V phaseoutput Value, etc. Figure 2weights1 and bias2 are respectively predetermined weights and biases for the combination of measured values ​​and target output voltage associated with phase 112; weights3 and bias3 are respectively predetermined weights and biases for the combination of measured values ​​and target output voltage associated with phase 113, and so on.

[0090] In this example embodiment, for the current control cycle, f1 is the activation function of phase 111 (indicating whether the phase is activated); f2 is the activation function of phase 112 (indicating whether the phase is activated); f3 is the activation function of phase 113 (indicating whether the phase is activated), and so on. Similar meanings apply to weights2, bias2, f2 and weights3, bias3, f3, ... relative to phases 112 and 113, ..., respectively.

[0091] Hidden layers can be added in a known manner within each neuron for determining each phase state in a more abrupt manner relative to the operating point. The number of hidden neural layers, the number of operating points for which combinations are determined for each phase state, and the frequency of determination are selected relative to the amount of computation applied to each converter, the balance between prediction accuracy and specific characteristics (particularly load-related).

[0092] exist Figure 2 In , n is the number of operating points involved in each determination of the phase state, i.e. the number of operating points in each series. For the predictor 14 described previously, n is the length (the number of samples obtained over a number of consecutive sample times) of a FIFO queue memory set (such as storing the current operating settings 210). However, the amount of storage necessary for a multi-phase power converter may become significant as n increases. A way to reduce this amount of storage is then to store at least part of the historical information, such as the operating point before the last one was transmitted to the predictor 14 by the value supply system 15, directly in the neural network instead of in an entry FIFO queue of the memory set. Such a neural network configuration is known in the art as a recurrent neural network. In such a recurrent neural network, long short-term memories may be preferred because they avoid vanishing or exploding gradients.

[0093] When implemented, the phase operating parameters for each phase 111, 112, 113, ... may be determined by the predictor 14 using any suitable algorithm (including algorithms based on machine learning processes, especially those just described for the phase state), wherein the state indicates that the phase is to be activated by the regulator 13. The phase operating parameters are then transmitted to the regulator 13 by the predictor 14 together with the phase state. Figure 1In the figure, when valid, the phase operation parameter for phase 111 is represented as (op)1, the phase operation parameter for phase 112 is represented as (op)2, the phase operation parameter for phase 113 is represented as (op)3, ...

[0094] As mentioned in the general part of this specification, the weights and biases of all phase states and possibly all phase operating parameters are predictor parameters. They are provided to the predictor 14 by a preparatory phase called training. This training is preferably performed by a computing device 40 external to the predictor 14 (see Figure 1 ) is implemented as determining the predictor parameters from labeled training data may require considerable computer resources. The computing device 40 may be provided as a separate computer, or may be accessed via the cloud. Such a configuration of computing devices 40 for the training phase is advantageous as these computing devices may be shared between a large number of users, thereby allowing potentially expensive computing devices to be implemented in a cost-effective manner. Each user may have access to a computing device for an initial training phase of the predictor for their power converter, and their power converter may then operate for a long period of time without requiring the computing device again.

[0095] The training phase mainly includes the following three steps:

[0096] - forming sets of labeled training data, such as each set comprising a series of consecutive operating points of the converter, which has an associated state (valid or invalid) for each phase. In this way, each set of labeled training data describes a possible sequence of operation of the converter over time, including instantaneous values ​​of at least some of the input and output voltages and currents of the converter and of the phase input and output voltages and currents, and also including instantaneous values ​​of the target output voltage. A corresponding desired state for all phases is associated with each series of consecutive operating points. In the art, the desired phase states are referred to as labels. The labeled training data can be advantageously selected in an appropriate manner with respect to the intended application of the converter 10, in particular with respect to its load 30, for later obtaining optimized operation of the converter in its specific application;

[0097] - predictor parameters are then determined by the computing means 40 using one of the known machine learning processes such as gradient descent, in particular Newton's method, or the conjugate gradient algorithm, statistical optimization methods, in particular genetic algorithms, or any process implementing back propagation or the like; and

[0098] - The predictor parameters are transferred to the predictor 14, which is used to determine the phase state using the predictor parameters. The transfer of the predictor parameters to the predictor 14 can be performed by value transfer or by writing a corresponding firmware in the predictor 14 to be implemented.

[0099] Running the predictor 14 while the converter 10 provides power to the load 30 is then referred to as inference of the phase state. The updated phase state is communicated to the regulator 13 causing it to switch from the previously implemented set of phase states to the updated phase state.

[0100] Reference now Figure 3 Describe the embodiments of this article, Figure 3 It is ensured that the regulator 13 does not control one or several of the phases 111, 112, 113, ..., for generating a phase output current that would be too high. In fact, a too high phase output current would degrade the switching devices of the phase, or cause overheating of some components. To this end, the power converter 10 may additionally include a monitor module 16 and a comparator 17. For example, the monitor module 16 may be connected to receive the output current of the converter 10 that has been collected by the value supply system 15, for all operating points of the n series currently processed by the predictor 14 (i.e., I output (0),I output (1), ..., I output The supervisor module 16 then calculates the minimum number of phases to be activated by the regulator 13, denoted as #min phase , for example, equal to Max i＝0 , …, n-1 (I output (i)) / I single_phase_max , where I output (i) is the measured value of the converter output current at the i-th operating point in the FIFO queue, and I single_phase_max is a stored value representing the maximum current allowed to be generated by each phase. The comparator 17 receives the value #min from the monitoring module 16 phase , on the other hand determines the number of phases to be activated resulting from the phase state denoted #active_phase_14 output by predictor 14. If #active_phase_14 is higher than or equal to #min phase , the phase state output by predictor 14 is transmitted to regulator 13 for implementation. If #active_phase_14 is less than #min phase , another run of the predictor 14 may be triggered to determine a new set of phase states from the same series of operating points, but with the constraint that the number of activated phases is equal to or greater than #min phase .

[0101] Furthermore, in this example embodiment, the power converter 10 includes a module 16 and a comparator 17 to generate control information 120 output to the regulator 13. In one embodiment, in a similar manner as previously discussed, the monitor module 16 operates to: i) monitor the current provided by the output voltage to the load 30 (such as the most recent sample of the output current or multiple samples of the output current), and ii) generate an output value 350 (such as how many phases to activate) based on the monitored output current (Ioutput(current)). The output value 350 indicates the number of multiple phases in the power converter 10 to be activated to generate the output voltage to power the load 30. The comparator 17 further operates to: i) compare the output value 350 with the control information, and ii) derive the control setting (control information 120) from the comparison.

[0102] According to a further embodiment, the comparator 17 is operated to select between the output value 350 and the control information 320 based on which has a larger magnitude to derive the control information 120 provided to the regulator 13 .

[0103] Figure 4 is an example diagram illustrating a mapping of current operating settings of a power converter to appropriate control information to achieve a desired control response according to embodiments herein.

[0104] As previously discussed, the power converter 10 includes a plurality of phases; the regulator 13 controls the plurality of phases to convert an input voltage into an output voltage.

[0105] exist Figure 4 In an example embodiment of , an instance of predictor 14-1 may be used to receive currently collected samples of operating settings 210 of power converter 10. Operating settings 210 are indicated as data set 410_1, data set 410_2, data set 410_3, and so on.

[0106] The data set 410 - 1 (such as data stored in a plurality of FIFO buffers) is a first set of buffered samples obtained at different sample times for each of a plurality of parameters (such as Vinput, Iinput, etc.).

[0107] The data set 410 - 2 (such as data stored in a plurality of FIFO buffers) is a second set of buffered samples obtained at different sample times for each of a plurality of parameters (such as Vinput, Iinput, etc.).

[0108] Data set 410 - 3 (such as data stored in a plurality of FIFO buffers) is a third set of buffered samples obtained at different sample times for each of a plurality of parameters (such as Vinput, Iinput, etc.), and so on.

[0109] Data set 410-4 (such as stored in a FIFO buffer) is a fourth set of buffered samples taken at different sample times. The values ​​in data set 410-4 measure the magnitude of the plurality of monitored parameters over different sample times; and so on.

[0110] Thus, the collected data samples include a respective sequence of a plurality of data samples for each of a plurality of parameters of the power converter collected over time.

[0111] As further shown, the predictor operates to convert currently collected samples of operating settings 210 of power converter 10 into appropriate control information 120. In one embodiment, the generated control information 120 is a machine-learned control response assigned to a pattern of previously stored samples of operating settings of power converter 10.

[0112] In one embodiment, the currently collected samples of operating settings 210 of power converter 10 represent the current operating state of power converter 10. Previously stored samples of operating settings (such as data set 410-1 indicating a first prior operating state of power converter 10, data set 410-2 indicating a second prior operating state of power converter 10, data set 410-3 indicating a third prior operating state of power converter 10, and so on) are stored.

[0113] In this example embodiment, each of the different sets of prior detected conditions (operation settings 210) is mapped to a corresponding appropriate control response based on prior machine learning.

[0114] More specifically, for a condition of power converter 10 (such as a voltage / current setting) as indicated by data set 410 - 1 , control information 120 - 1 (such as indicating the number of phases to be activated to accommodate the current conditions) represents a corresponding appropriate control response for controlling power converter 10 .

[0115] For conditions (such as settings) of power converter 10 as indicated by data set 410-2, control information 120-2 (such as indicating the number of phases to be activated to adapt to the current conditions / settings as specified by data set 410-2) indicates a corresponding appropriate control response for controlling power converter 10.

[0116] For conditions (such as settings) of power converter 10 as indicated by data set 410-3, control information 120-3 (such as indicating the number of phases to be activated to adapt to the current conditions / settings as specified by data set 410-3) indicates a corresponding appropriate control response for controlling power converter 10.

[0117] For a condition (such as a setting) of power converter 10 as indicated by data set 410-4, control information 120-4 (such as indicating the number of phases to be activated to adapt to the current condition / setting as specified by data set 410-4) indicates a corresponding appropriate control response for controlling power converter 10.

[0118] In this example embodiment, it is assumed that the current operating settings 210 (for N samples) of the power converter 10 are most similar to the settings indicated by the data set 410-3. In other words, the current (most recently) collected samples of the operating settings 210 of the power converter 10 most closely match the pattern of previously stored samples of the operating settings of the power converter 10. In this case, the predictor 14-1 maps the data set 410-3 to the appropriate control response as indicated by the control information 120-3 for selection.

[0119] As previously discussed, in one embodiment, the generated control information 120 (obtained from control information 120-3) indicates the number of multiple phases activated in the power converter for the previous operating state (associated with data set 410-3). The activation of the number of phases in the power converter 10 specified by the control information 120 is used to maintain the output voltage of the power converter 10 within the desired voltage range.

[0120] After generating control information 120 (such as selected from control information 120 - 3 ), predictor 14 - 1 outputs control information 120 to regulator 13 or other suitable resources to control the plurality of phases.

[0121] Thus, in one embodiment, predictor 14-1 is further operative to map currently collected samples of operating settings 210 of power converter 10 to previously stored samples of operating settings of power converter 10 (such as data set 410-3) to identify and select control information 120-3. The previously stored samples of operating settings (as indicated by data set 410-3) are one of a plurality of sets of previously stored samples of operating settings of power converter 10 (data set 410).

[0122] Figure 5 is an example diagram illustrating mapping of current operating settings of a power converter to multiple sets of control information and deriving control information from the multiple sets to achieve a desired control response according to embodiments herein.

[0123] In this example embodiment, predictor 14-1 identifies that current operating settings 210 most closely match settings specified by data set 410-3 and settings specified by data set 410-4. In this case, predictor 14-1 applies interpolation and / or extrapolation techniques to derive control information 120 from a combination of control information 120-3 and control information 120-4.

[0124] Figure 6 is an example diagram illustrating the use of logic to derive control information to control a power converter according to embodiments herein.

[0125] In this example embodiment, similar to Figure 3 , the processing logic of the predictor 14 - 2 receives the current operating settings 210 of the power converter 10 (such as stored in the buffer 610 ) and derives the control information 120 based on this information.

[0126] Buffer 610 - 1 stores samples of Vinput; buffer 610 - 2 stores samples of Iinput; buffer 610 - 3 stores samples of Vphase_output; buffer 610 - 4 stores samples of Iphase_output, and so on.

[0127] The control information 120 indicates settings applied to the regulator 13 .

[0128] Figure 7 is an example block diagram of a computer system for implementing any of the operations as previously discussed according to embodiments herein.

[0129] Any of the resources discussed herein (such as the predictor 14, the regulator 13, etc.) may be configured to include computer processor hardware and / or corresponding executable instructions to perform the various operations discussed herein.

[0130] As shown, the computer system 750 of this example includes an interconnect 711 that couples a computer-readable storage medium 712 (such as a non-transitory type of medium (which can be any suitable type of hardware storage medium that can store and retrieve digital information)), a processor 713 (computer processor hardware), an I / O interface 714, and a communication interface 717.

[0131] I / O interface 714 supports connections to storage repository 780 and input resources 792 .

[0132] The computer readable storage medium 712 may be any hardware storage device, such as a memory, optical storage, a hard drive, a floppy disk, etc. In one embodiment, the computer readable storage medium 712 stores instructions and / or data.

[0133] As shown, the computer-readable storage medium 712 may be encoded with the communication management application 140-1 (eg, including instructions) to perform any of the operations discussed herein.

[0134] During operation of one embodiment, processor 713 accesses computer-readable storage medium 712 via use of interconnect 711 to launch, run, execute, interpret or otherwise perform instructions in management application 140-1 stored on computer-readable storage medium 712. Execution of communication management application 140-1 generates communication management process 140-2 to perform any of the operations and / or processes discussed herein.

[0135] Those skilled in the art will appreciate that the computer system 750 may include other processing and / or software and hardware components, such as an operating system that controls the allocation and use of hardware resources to execute the communication management application 140-1.

[0136] According to different embodiments, it should be noted that the computer system can reside in any of various types of devices, including but not limited to mobile computers, personal computer systems, wireless devices, wireless access points, base stations, telephone devices, desktop computers, laptops, notebooks, netbooks, mainframe computer systems, handheld computers, workstations, network computers, application servers, storage devices, consumer electronic devices (such as cameras), camcorders, set-top boxes, mobile devices, video game consoles, handheld video game devices, peripheral devices (such as switches), modems, routers, set-top boxes, content management devices, handheld remote control devices, any type of computing or electronic device, etc. The computer system 750 can be located in any location or included in any appropriate resource in any network environment to implement the functions discussed herein.

[0137] Now it will go through Figure 8 The flowcharts of discuss the functions supported by different resources. It should be noted that the steps in the following flowcharts can be performed in any suitable order.

[0138] Figure 8 is a flowchart 800 illustrating an example method according to an embodiment. It should be noted that, as discussed above, there is some overlap in concepts.

[0139] In processing operation 810 , predictor 14 receives a currently collected sample of operating settings 210 of power converter 10 .

[0140] In processing operation 820 , predictor 14 converts currently collected samples of operating settings 210 of power converter 10 into control information 120 , which is a machine learning control response assigned to previously stored samples of operating settings of power converter 10 , such as one or more data sets 410 .

[0141] In processing operation 830 (such as a sub-operation of processing operation 820), predictor 14 maps currently collected samples of operating setting 210 of power converter 10 to a pattern of previously stored samples of operating setting (such as 410-3) of power converter 10; predictor 14 then selects control information 120-3 assigned to the pattern (setting 210) as a basis to generate control information 120 transmitted to regulator 13.

[0142] In processing operation 840 , such as an alternative sub-operation of processing operation 820 , predictor 14 inputs the current sample of operating settings 210 to logic 14 - 1 , which generates control information 120 based on the received settings 210 .

[0143] In process operation 850 , predictor 14 outputs control information 120 to control the plurality of phases of power converter 10 .

[0144] Although the detailed description has focused on predictor embodiments suitable for implementing nearest neighbor or regression-based machine learning models, it should be understood that the embodiments herein are not limited to these particular models, and other embodiments may be used instead. In particular, any regression variant and any sequence based on a hidden Markov chain may be used.

[0145] It should also be understood that embodiments herein are applicable to any electrical power conversion, such as any electrical power conversion other than DC-DC, particularly AC-DC power conversion, although for illustrative purposes the detailed description focuses on DC-DC power conversion.

[0146] Finally, it should also be understood that embodiments herein are compatible with any phase management system, particularly if the phase management system is provided by training phases. For example, in the event that one of the selected phases is disconnected or fails, the phase management system may use limited selection among converter phases to retain the unselected phases as spare phases.

Claims

1. A power converter comprising a plurality of phases that collectively provide an output current and an output voltage to a load, the output current and the output voltage being derived from a phase output current and a phase output voltage provided by the phases, respectively, wherein the power converter further comprises: - a regulator for converting an input voltage into said output voltage; a value supply system for collecting operating points, each operating point being related to an operating instant of the power converter and each operating point comprising, for the corresponding operating instant, measured values ​​of one or more input parameters of the power converter among the input current, the input voltage, the phase input current, the phase input voltage and / or one or more output parameters of the output current, the output voltage, the phase output current, the phase output voltage, and each operating point also comprising at least one value of a target output voltage assigned to the same operating instant of the power converter; as well as a predictor for providing, for each respective phase of the plurality of phases, a respective updated phase state indicating whether the respective phase is activated, the updated phase state being transmitted to the regulator so that the regulator selectively activates each phase indicated by the respective updated phase state, the predictor being operative to determine the updated phase state using a process based on the at least one operating point collected by the value provisioning system and based on predictor parameters obtained from a machine learning process, wherein the predictor is operative to determine the updated phase state for each phase based on a plurality of operating points, the plurality of operating points being associated with successive corresponding operating instants of the power converter, the plurality of operating points being a fixed number of operating points, The predictor operation is used to implement a recurrent neural network, so that each time a further operating point is provided to the predictor by the value supply system, the further operating point is added to the plurality of operating points in a FIFO queue manner so as to obtain an updated plurality of operating points, which will be used to publish a further updated phase state for each phase. 2 . The power converter according to claim 1 , wherein the power converter is a DC-DC power converter or an AC-DC power converter.

3. A power converter according to claim 1, wherein the predictor is operative to: for each phase, when the corresponding updated phase status indicates that the phase is valid, provide an updated value of a phase operating parameter of the phase, the updated value of the phase operating parameter going to the regulator so that the regulator activates the phase in a manner matching the updated value of the phase operating parameter. 4 . The power converter of claim 1 , wherein the predictor comprises a lookup table for storing labeled training data, the predictor further operative to select one of the labeled training data as the closest neighboring data to the at least one operating point. 5 . The power converter of claim 1 , wherein the predictor is further operative to implement at least one calculation step of a regression type in a calculation sequence for issuing the updated phase state for each phase from the at least one operating point.

6. The power converter of claim 1 wherein the predictor operates in a feed-forward artificial intelligence manner.

7. The power converter of claim 1 wherein the predictor operates as a single layer neural network.

8. The power converter of claim 6, wherein the predictor comprises a neuromorphic chip.

9. The power converter of claim 1 further comprising an operation for determining a minimum number of phases to be activated based on collected measurements of the converter output current for each of the at least one operating point, and the power converter is adapted so that the updated phase state to be implemented by the regulator results in a number of active phases at least equal to the minimum number of phases to be activated.

10. A method for performing electric power conversion using a power converter, the power converter comprising a plurality of phases operative to provide an output current and an output voltage to a load, the output current and the output voltage being derived from phase output currents and phase output voltages respectively provided by the phases, the method comprising: -Selectively activate each phase during power conversion; - collecting at least one operating point occurring during the power conversion, each operating point being related to an operating instant during the power conversion and comprising: i) measured values ​​of one or more input parameters of the power converter (10) of the input current, the input voltage of the power converter, the phase input current, the phase input voltage and / or one or more output parameters of the output current, the output voltage, the phase output current, the phase output voltage at the operating instant; and ii) at least one value of a target output voltage of the power conversion at the operating instant assigned to the operating point; and - providing, via a predictor, for each phase of the plurality of phases an updated phase state for indicating in a further implementation of the regulator whether the phase is active or inactive, wherein each updated phase state is determined using processing based on the collected at least one operating point and based on predictor parameters obtained from machine learning, wherein the predictor is operative to determine the updated phase state for each phase based on a plurality of operating points, the plurality of operating points being associated with successive corresponding operating instants of the power converter, the plurality of operating points being a fixed number of operating points, The predictor operation is used to implement a recurrent neural network, so that each time a further operating point is provided to the predictor, the further operating point is added to the plurality of operating points in a FIFO queue manner to obtain an updated plurality of operating points, which will be used to publish a further updated phase state for each phase.

11. The method according to claim 10, further comprising: The following preliminary steps / 1 / to / 3 / are performed during said machine learning: / 1 / Gathering labeled training data, wherein the labeled training data includes a training operating point for each phase and a corresponding associated phase state; / 2 / using the labeled training data to train a machine learning model in order to obtain the predictor parameters to be used by the predictor for inferring each new updated phase state; as well as / 3 / transmitting the predictor parameters to the predictor, Wherein the power conversion is operated using the predictor parameters transmitted in step / 3 / .

12. The method according to claim 11, wherein step / 2 / is performed using a pass-through resource which is i) located outside the power converter and ii) disconnected from the power converter when the power converter performs the power conversion.

13. The method according to claim 10, further comprising: - determining a minimum number of phases to be activated based on the collected measurements of the converter output current for each of the at least one operating point; as well as - if said minimum number of phases to be activated is greater than the number of valid phases resulting from said updated phase states issued by said predictor, correcting at least one of said updated phase states issued by said predictor so as to satisfy said minimum number of phases to be activated, Wherein the updated phase states of the plurality of phases further implemented by the regulator for performing the power conversion results in a number of active phases being at least equal to the minimum number of phases to be activated.

14. The method of claim 10, wherein the power conversion provides electrical energy to a load forming part of a data center or server farm.

15. The method of claim 10, wherein the power conversion provides electrical energy to an electronic device.

16. An apparatus comprising: a power converter comprising a plurality of phases operative to convert an input voltage into an output voltage; a regulator operative to control the plurality of phases of the power converter; as well as Predictors, operations are used to: i) receiving a current sample of operational settings of the power converter; ii) converting the current sample of the operating settings of the power converter into control information that is a machine-learned control response assigned to a previous set of samples of the operating settings of the power converter to maintain the output voltage within regulation, and iii) outputting the control information to control the plurality of phases, wherein the control information is associated with a corresponding updated phase state indicating whether the corresponding phase is activated, wherein the predictor is operative to determine the updated phase state for each phase based on a plurality of operating points, the plurality of operating points being associated with successive corresponding operating instants of the power converter, the plurality of operating points being a fixed number of operating points, The predictor operation is used to implement a recurrent neural network, so that each time a further operating point is provided to the predictor, the further operating point is added to the plurality of operating points in a FIFO queue manner to obtain an updated plurality of operating points, which will be used to publish a further updated phase state for each phase.

17. The apparatus of claim 16, wherein the current sample of the operational setting of the power converter represents a current operational state of the power converter, and Wherein a previously stored sample of the operating settings of the power converter indicates a previous operating state of the power converter, and the machine learning control response is assigned to the previous operating state to maintain the output voltage within regulation.

18. The apparatus of claim 17, wherein the control information indicates a number of multiple phases activated in the power converter for a previous operating state, the activation operation of the number of phases in the power converter specified by the control information being used to maintain the output voltage within a desired voltage range.

19. The apparatus of claim 16, wherein the current sample of operational settings of the power converter samples comprises a respective sequence of a plurality of data samples of each of a plurality of parameters of the power converter collected over time.

20. The apparatus of claim 19, wherein the corresponding sequence of the plurality of data samples comprises: a first sequence of buffered samples for measuring the magnitude of said input voltage at a plurality of sample times; a second sequence of buffered samples for measuring the magnitude of an input current provided by said input voltage at said plurality of sample times; a third sequence of buffered samples for measuring the magnitude of the output voltage at the plurality of sample times; as well as A fourth sequence of buffered samples is used to measure the magnitude of the output current provided to the load by the output voltage at the plurality of sample times.

21. The apparatus of claim 16, further comprising: A monitor module is operative to: i) monitor an output current provided to a load by the output voltage; and ii) generating an output value based on the monitored output current, the output value indicating the number of the plurality of phases in the power converter circuit to be activated to generate the output voltage to power the load; as well as A comparator is operative to: i) compare the output value with the control information, and ii) derive a control setting from the comparison.

22. The apparatus of claim 21, wherein the comparator is operative to select based on which of the output value and the control information has a greater magnitude to derive the control information.

23. The apparatus of claim 16, wherein the predictor is further operative to map the current sample of the operating setting of the power converter to the previously stored sample of the operating setting of the power converter to identify the control information, the previously stored sample of the operating setting of the power converter being one of a plurality of sets of previously stored samples of the operating setting of the power converter, each previously stored sample of the operating setting being assigned a corresponding machine learning control response.

24. The apparatus of claim 16, wherein a currently collected sample of operating settings of the power converter substantially matches the previously stored sample of operating settings of the power converter.

Images