CN114936632A - Hardware resource dynamic multiplexing neural network controller for digital power supply - Google Patents

Hardware resource dynamic multiplexing neural network controller for digital power supply Download PDF

Info

Publication number
CN114936632A
CN114936632A CN202210439324.XA CN202210439324A CN114936632A CN 114936632 A CN114936632 A CN 114936632A CN 202210439324 A CN202210439324 A CN 202210439324A CN 114936632 A CN114936632 A CN 114936632A
Authority
CN
China
Prior art keywords
neural network
controller
network controller
tuning controller
fine tuning
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210439324.XA
Other languages
Chinese (zh)
Inventor
武嘉瑜
刘建富
魏廷存
陈楠
肖培磊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
CETC 58 Research Institute
Original Assignee
CETC 58 Research Institute
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by CETC 58 Research Institute filed Critical CETC 58 Research Institute
Priority to CN202210439324.XA priority Critical patent/CN114936632A/en
Publication of CN114936632A publication Critical patent/CN114936632A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02M—APPARATUS FOR CONVERSION BETWEEN AC AND AC, BETWEEN AC AND DC, OR BETWEEN DC AND DC, AND FOR USE WITH MAINS OR SIMILAR POWER SUPPLY SYSTEMS; CONVERSION OF DC OR AC INPUT POWER INTO SURGE OUTPUT POWER; CONTROL OR REGULATION THEREOF
    • H02M3/00—Conversion of dc power input into dc power output

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Neurology (AREA)
  • Power Engineering (AREA)
  • Dc-Dc Converters (AREA)
  • Control Of Voltage And Current In General (AREA)

Abstract

The invention discloses a hardware resource dynamic multiplexing neural network controller for a digital power supply, which belongs to the field of power electronics and comprises a neural network controller and a control module, wherein the neural network controller is used for generating a digital duty ratio; the neural network controller simultaneously realizes a coarse tuning controller and a fine tuning controller, and the coarse tuning controller is used for adjusting the output voltage to a reference voltage as soon as possible after the digital power supply is disturbed so as to improve the transient response performance; the fine tuning controller is mainly used for eliminating steady-state errors and stabilizing the output voltage at a reference voltage. The invention can reduce the structure and realize the same control performance, and save the hardware resource needed by the hidden layer node.

Description

Hardware resource dynamic multiplexing neural network controller for digital power supply
Technical Field
The invention relates to the technical field of power electronics, in particular to a hardware resource dynamic multiplexing neural network controller for a digital power supply.
Background
The structure of a digitally controlled DC-DC switching converter (hereinafter referred to simply as a digital power supply) is shown in fig. 1. Analog output voltage V of load end out (t) conversion to digital output V by ADC out [k]Then V is added out [k]And a reference voltage V ref Error signal e k between]Sent to a digital voltage compensator. In a digital voltage compensator (e.g., a neural network controller), a specific control algorithm is used to calculate the digital duty cycle signal d k]Then, the Digital duty ratio signal d [ k ] is modulated by DPWM (Digital Pulse Width Modulation)]Converting into analog duty ratio signal d (t), and finally driving power level switch S by Gate driver P And S N To regulate the output voltage V out (t) stabilizing the voltage at a reference voltage V ref 。
For a digital power supply, a neural network controller has the advantages of high adaptivity, strong robustness, good transient performance and the like, but the neural network controller usually consumes a large amount of hardware resources. The traditional method for reducing hardware resource consumption of the neural network controller is to carry out 'pruning' on the neural network, namely carrying out weight value zero setting on a node with low sensitivity (less influence on an output result) or directly deleting the node. Although the pruning can effectively reduce the scale, the calculation amount and the consumption of hardware resources of the network, the problem of reduced control precision caused by a digital power supply is solved, so that the steady-state error is increased, the selection of the pruning node needs repeated iterative verification, the workload is large, and the adaptivity of the pruned network is reduced.
The document "dynamic Channel planning by Conditional access Change for Deep Neural Networks", IEEE Transactions on Neural Networks and Learning Systems, Vol.32, No.2, February 2021, pp: 799-. The method provided in the text effectively reduces the scale and the calculation amount of the network while obtaining higher precision, but the method still realizes the reduction of hardware resources and the operation acceleration based on the principle of network 'pruning', and the dynamically pruned network structure still causes partial precision loss due to the loss of partial nodes or channels, and more importantly, leads to the reduction of adaptivity.
Disclosure of Invention
The invention aims to provide a hardware resource dynamic multiplexing neural network controller for a digital power supply, which aims to solve the problem of high hardware resource consumption of the current neural network controller.
In order to solve the technical problem, the invention provides a hardware resource dynamic multiplexing neural network controller for a digital power supply, which comprises a neural network controller and a control module, wherein the neural network controller is used for generating a digital duty ratio, the control module comprises a lookup table and a dynamic multiplexing controller, the lookup table is used for storing the weight and the bias of the neural network controller, and the dynamic multiplexing controller dynamically controls the switching of the weight and the bias of the neural network controller and forbids partial hidden nodes;
the neural network controller simultaneously realizes a coarse tuning controller and a fine tuning controller, and the coarse tuning controller is used for adjusting the output voltage to a reference voltage as soon as possible after the digital power supply is disturbed so as to improve the transient response performance; the fine tuning controller is mainly used for eliminating steady-state errors and stabilizing the output voltage at a reference voltage.
Optionally, the neural network controller is obtained by offline training of data lines, and includes the following steps:
establishing error bands on two sides of the reference voltage, grouping training data of the neural network controller according to the error values, wherein the data with the error within the error bands are the training data of the fine tuning controller, and otherwise, the data are the training data of the coarse tuning controller;
and respectively training two groups of training data to two neural networks with the same input and output and different structures by adopting a neural network training method to respectively obtain the weight and the offset of the fine tuning controller and the weight and the offset of the coarse tuning controller.
Optionally, in the training data of the fine tuning controller, the output voltage is close to an ideal value; in the training data of the coarse tuning controller, various differences exist between the output voltage and an ideal value, including output voltage changes under different disturbances.
Optionally, the range of the error band is within ± 10% of the reference voltage.
Optionally, the number of implicit layers or implicit layer nodes of the coarse tuning controller is more, the number of implicit layers or implicit layer nodes of the fine tuning controller is relatively less, and the number of implicit layers or implicit layer nodes of the coarse tuning controller and the number of implicit layers or implicit layer nodes of the fine tuning controller are all smaller than the number of implicit layers or implicit layer nodes of the neural network controller obtained through direct data training; if the coarse tuning controller and the fine tuning controller adopt the neural network with the same hidden layer number, the coarse tuning controller can obtain the same structure with the fine tuning controller by forbidding part of hidden layer nodes, and if the weight and the bias are switched into the weight and the bias of the fine tuning controller at the same time, the coarse tuning controller can be used for realizing the function of the fine tuning controller.
Optionally, based on a time division multiplexing principle, selecting a structure of the neural network controller as a structure consistent with the coarse tuning controller, dynamically disabling part of hidden layer nodes and switching weights and biases according to a position where an error is located in an error band, so as to realize functions of the two controllers, obtain the neural network controller with dynamic multiplexing of hardware resources, and save hardware resources required by part of the hidden layer nodes; when the neural network controller works as a fine tuning controller, forbidding part of hidden layer nodes used by the coarse tuning controller independently, and switching the weight and the bias of the fine tuning controller, wherein the structure of the neural network controller is equivalent to that of the fine tuning controller; when the neural network controller works as a coarse tuning controller, all hidden layer nodes are activated, and the weight and the bias of the coarse tuning controller are switched to use, so that the structure of the neural network controller is the structure of the coarse tuning controller.
Optionally, the neural network controller is divided into two or more sub-controllers, and then based on a time division multiplexing principle, the functions of the two or more sub-controllers are realized by switching weights and biases and dynamically activating and deactivating part of hidden layer nodes by using one or more neural networks, so that the neural network controller is accelerated or hardware resource consumption is reduced.
In the hardware resource dynamic multiplexing neural network controller for the digital power supply, the complete neural network controller is divided into two sub-controllers: and then based on the idea of time division multiplexing, a small-scale neural network is adopted to realize the functions of two sub-controllers by dynamically switching weights and biases and forbidding part of hidden layer nodes. Therefore, the hardware resource dynamic multiplexing neural network controller for the digital power supply provided by the invention can not delete the trained hidden layer nodes, thereby not causing the reduction of the control performance and saving part of hardware resources required by the hidden layer nodes.
Drawings
FIG. 1 is a schematic diagram of a digitally controlled DC-DC switching converter;
FIG. 2 is a schematic diagram of a hardware resource dynamic multiplexing neural network controller for a digital power supply according to the present invention;
FIG. 3 is a schematic diagram of the operation of a neural network controller provided by the present invention;
FIG. 4 is a schematic diagram of the switching timing of the coarse and fine tuning controllers;
FIG. 5(a) is a graph comparing the output voltage response curves of a digital power supply at steady state using the techniques of the present invention and a conventional neural network controller;
FIG. 5(b) is a graph comparing the output voltage response curves of a digital power supply at load current jumps of + -1A using the present technique and a conventional neural network controller;
FIG. 5(c) is a graph comparing the output voltage response curves of a digital power supply at + -1V input voltage transitions using the present technology and a conventional neural network controller.
Detailed Description
The hardware resource dynamic multiplexing neural network controller for digital power supply according to the present invention is further described in detail with reference to the accompanying drawings and specific embodiments. Advantages and features of the present invention will become apparent from the following description and from the claims. It is to be noted that the drawings are in a very simplified form and are not to precise scale, which is merely for the purpose of facilitating and distinctly claiming the embodiments of the present invention.
The invention provides a hardware resource dynamic multiplexing neural network controller for a digital power supply, which structurally mainly comprises a neural network controller and a control module as shown in figure 2, wherein the control module comprises a lookup table (for storing weight values and bias values of a fine tuning controller and a coarse tuning controller) and a dynamic multiplexing controller. The neural network controller is used for calculating digital duty ratio, and the control module is used for controlling dynamic switching of weight and bias and dynamically disabling or activating nodes of partial hidden layers. The neural network controller functionally realizes a coarse tuning controller and a fine tuning controller, as shown in fig. 3, the coarse tuning controller is used for pulling an output voltage back to a reference voltage as soon as possible after a digital power supply is disturbed, so as to reduce stabilization time and overshoot and improve transient performance, and the fine tuning controller is mainly used for eliminating steady-state errors, maintaining a stable state and stabilizing the output voltage at the reference voltage. The input signal of the neural network controller is { e } o [k],e oc [k],I ind [k],V in [k]In which e is o [k]Is the voltage error of the current cycle, I ind [k]Is the inductor current of the current cycle, V in [k]Is the input voltage of the current cycle, e oc [k]＝e[k]-e[k-1]Is the voltage error rate of change for the current cycle. The output signal of the neural network controller is a digital duty cycle d nn [k]The specific design process is as follows:
(1) and establishing an error band and classifying the data. Firstly, establishing error band E on two sides of reference voltage b Usually within 10% of the reference voltageAs an error band, as shown in fig. 4. And then, classifying the training data of the neural network controller according to the error value, namely, the data with the error within the error band is the training data of the fine tuning controller, and otherwise, the data is the training data of the coarse tuning controller. In the training data of the fine tuning controller, the output voltage is close to an ideal value; in the training data of the coarse tuning controller, various differences exist between the output voltage and an ideal value, and the output voltage changes under different disturbances are included.
(2) And obtaining a coarse adjustment controller and a fine adjustment controller. Training the classified data with two neural networks with the same input and output but different structures by a specific training mode (such as a gradient descent method) to obtain the weights and offsets of the fine tuning controller and the coarse tuning controller respectively, wherein the number of hidden layer nodes of the coarse tuning controller and the fine tuning controller is N respectively c And N f As shown in fig. 3, the conventional neural network controller is obtained by directly training a neural network with training data, and the number of hidden layer nodes is N n Inevitably present N n ≤N c +N f And N is f ≤N c <N n 。
(3) And dynamically multiplexing hardware resources with the neural network controller. Coarse tuning controller disable N c -N f The function of the fine tuning controller can be realized by hiding the layer nodes and adopting the weight and the bias of the fine tuning controller. According to the classification principle of training data, the fine tuning controller and the coarse tuning controller can not work synchronously, namely when the fine tuning controller works, the coarse tuning controller is in an idle state; otherwise, when the fine tuning controller is in the idle state, the coarse tuning controller is in the working state. Therefore, based on the idea of time division multiplexing, a neural network (the structure of which is consistent with that of the coarse control) can be adopted to dynamically disable or activate N by dynamically switching the weight and bias c -N f The hidden layer node realizes the functions of two sub-controllers, specifically, N is forbidden when the controller works as a fine-tuning controller c -N f The hidden layer nodes are equivalent to the structure of the fine tuning controller, and are switched into the weight and the bias of the fine tuning controller; when inWhen the controller works as a coarse tuning controller, all hidden layer nodes are activated, and at the moment, the structure of the neural network is the structure of the coarse tuning controller and is switched into the weight and the bias of the coarse tuning controller.
(4) The switching sequence for the fine and coarse controllers is shown in FIG. 4, where the output voltage is within the steady state error band for three consecutive cycles, while the rate of change of error e is detected for the second of the three consecutive cycles c [k]0, as follows:
{(e oc [k]≠0)&(|e o [k]|≤|E b |)} (k＝1,3) &{(e oc [k]＝0)&(|e o [k]|≤|E b |)} (k＝2) (1)
the controller is switched from coarse to fine, with the part (N) disabled c -N f ) And (4) the nodes of the hidden layer are switched to fine tuning the weight and the bias of the controller at the same time.
When the output voltage is outside the steady state error band for three consecutive cycles, it is expressed as follows:
{|e o [k]|＞|E b |} (k＝1,2,3) (2)
the controller is converted from a fine tuning controller to a coarse tuning controller, at which time all hidden layer nodes are activated and switched to the weights and biases of the coarse tuning controller.
Aiming at the Buck type digital power supply, fig. 5(a), 5(b) and 5(c) are respectively the comparison of the output voltage response curves of the digital power supply when the technology of the invention is adopted and the traditional neural network controller is in a stable state, the load current jumps to +/-1A and the input voltage jumps to +/-1V, and it can be seen that the digital power supply adopting the hardware resource dynamic multiplexing neural network controller provided by the invention is almost overlapped with the output voltage response curve of the digital power supply adopting the traditional neural network controller, which shows that the performance of the two digital power supplies is similar, and the technology of the invention has no obvious performance reduction.
Compared with the traditional neural network controller, the hardware resource dynamic multiplexing neural network controller for the digital power supply provided by the invention has the advantages that the structure is reduced, the same control performance is almost realized, and the hardware resource dynamic multiplexing neural network controller for the digital power supply canSaving N n -N c The hardware resources required by the individual hidden layer nodes.
The above description is only for the purpose of describing the preferred embodiments of the present invention, and is not intended to limit the scope of the present invention, and any variations and modifications made by those skilled in the art based on the above disclosure are within the scope of the appended claims.

Claims (7)

1. A hardware resource dynamic multiplexing neural network controller for a digital power supply is characterized by comprising a neural network controller and a control module, wherein the neural network controller is used for generating a digital duty ratio, the control module comprises a lookup table and a dynamic multiplexing controller, the lookup table is used for storing a weight and an offset of the neural network controller, and the dynamic multiplexing controller dynamically controls the switching of the weight and the offset of the neural network controller and disables a part of hidden nodes;
the neural network controller simultaneously realizes a coarse tuning controller and a fine tuning controller, and the coarse tuning controller is used for adjusting the output voltage to a reference voltage as soon as possible after the digital power supply is disturbed so as to improve the transient response performance; the fine tuning controller is mainly used for eliminating steady-state errors and stabilizing the output voltage at a reference voltage.
2. The hardware resource dynamic multiplexing neural network controller for a digital power supply of claim 1, wherein the neural network controller is obtained by offline training of data, comprising the steps of:
establishing error bands on two sides of the reference voltage, grouping training data of the neural network controller according to the error values, wherein the data with the error within the error bands are the training data of the fine tuning controller, and otherwise, the data are the training data of the coarse tuning controller;
and (3) respectively training two groups of training data to two neural networks with the same input and output and different structures by adopting a neural network training method to respectively obtain the weight and the offset of the fine tuning controller and the weight and the offset of the coarse tuning controller.
3. The hardware resource dynamic multiplexing neural network controller for a digital power supply of claim 2, wherein in training data of the fine tuning controller, an output voltage is close to an ideal value; in the training data of the coarse tuning controller, various differences exist between the output voltage and an ideal value, including output voltage changes under different disturbances.
4. The hardware resource dynamic multiplexing neural network controller for a digital power supply of claim 3, wherein the error band ranges within ± 10% of a reference voltage.
5. The hardware resource dynamic multiplexing neural network controller for digital power supply of claim 4, wherein the number of implicit layers or implicit layer nodes of the coarse tuning controller is more, the number of implicit layers or implicit layer nodes of the fine tuning controller is relatively less, and the number of implicit layers or implicit layer nodes of the coarse tuning controller and the number of implicit layers or implicit layer nodes of the fine tuning controller are all less than the number of implicit layers or implicit layer nodes of the neural network controller directly trained by data; if the coarse tuning controller and the fine tuning controller adopt the neural network with the same hidden layer number, the coarse tuning controller can obtain the same structure with the fine tuning controller by forbidding part of hidden layer nodes, and if the weight and the bias are switched into the weight and the bias of the fine tuning controller at the same time, the coarse tuning controller can be used for realizing the function of the fine tuning controller.
6. The hardware resource dynamic multiplexing neural network controller for digital power supply of claim 5, wherein based on the time division multiplexing principle, the structure of the neural network controller is selected to be the same as that of the coarse tuning controller, and according to the position of the error in the error band, part of hidden layer nodes are disabled dynamically and the weight and the bias are switched, the functions of the two controllers are realized, and the hardware resource dynamic multiplexing neural network controller is obtained, so as to save the hardware resources required by part of hidden layer nodes; when the neural network controller works as a fine tuning controller, forbidding part of hidden layer nodes used by the coarse tuning controller independently, and switching the weight and the bias of the fine tuning controller, wherein the structure of the neural network controller is equivalent to that of the fine tuning controller; when the neural network controller works as a coarse tuning controller, all hidden layer nodes are activated, and the weight and the bias of the coarse tuning controller are switched to use, so that the structure of the neural network controller is the structure of the coarse tuning controller.
7. The hardware resource dynamic multiplexing neural network controller for digital power supply of claim 6, wherein all the neural network controller is divided into two or more sub-controllers, and then based on the time division multiplexing principle, one or more neural networks are adopted to realize the functions of two or more sub-controllers by switching weights and biases and dynamically activating and deactivating part of hidden layer nodes, thereby realizing the acceleration of the neural network controller or the reduction of the hardware resource consumption.
CN202210439324.XA 2022-04-25 2022-04-25 Hardware resource dynamic multiplexing neural network controller for digital power supply Pending CN114936632A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210439324.XA CN114936632A (en) 2022-04-25 2022-04-25 Hardware resource dynamic multiplexing neural network controller for digital power supply

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210439324.XA CN114936632A (en) 2022-04-25 2022-04-25 Hardware resource dynamic multiplexing neural network controller for digital power supply

Publications (1)

Publication Number Publication Date
CN114936632A true CN114936632A (en) 2022-08-23

Family

ID=82861918

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210439324.XA Pending CN114936632A (en) 2022-04-25 2022-04-25 Hardware resource dynamic multiplexing neural network controller for digital power supply

Country Status (1)

Country Link
CN (1) CN114936632A (en)

Similar Documents

Publication Publication Date Title
CN103516216B (en) Digital switch mode voltage regulators
CN103051177B (en) Quick response control circuit and control method thereof
US7902800B2 (en) Adaptive power supply and related circuitry
EP1909379B1 (en) Method and apparatus for pulse width modulation
TW201905623A (en) Envelope-tracking power supply modulator
US11594971B2 (en) Control circuit and control method for switching regulator
CN102055332A (en) Hysteretic controlled buck-boost converter
US20130162228A1 (en) Single Inductor Multiple Output Power Converter
GB2311629A (en) Constant on-time architecture for switching voltage regulator
CN106452369B (en) Efficient envelope tracking power supply and method based on power amplifier output signal control
US20080238380A1 (en) Hierarchical control for an integrated voltage regulator
US20050225376A1 (en) Adaptive supply voltage body bias apparatus and method thereof
WO2010083202A1 (en) Sensor-less operation and detection of ccm and dcm operation modes in synchronous switching power converters
US9312766B2 (en) Digital serializer based pulsewidth modulator controller
CN103546034B (en) A kind of compounding feedforward control type Hysteresis control system
CN103023327B (en) A kind of quick Hysteresis control circuit of self-adaptative adjustment ring width
US20140159679A1 (en) Target voltage generator for a dc to dc converter, a combination of a target voltage generator and a dc to dc converter, and a method of operating a dc to dc converter
KR102453380B1 (en) Boost dc-dc converter using dsm, duty ratio controller for the boost dc-dc converter, and a method for configuring the duty ratio controller
CN114936632A (en) Hardware resource dynamic multiplexing neural network controller for digital power supply
CN115001274B (en) Input-output co-modulated adaptive ramp voltage type pulse width control BUCK converter
CN109375693A (en) A kind of voltage adjuster
Xu et al. Pulse train‐controlled CCM boost converter with suppression of low‐frequency oscillation
CN114900039A (en) Digital peak current mode control method, system and extended pulse width modulation method
CN115037157A (en) ISOS system output voltage balance control method based on model prediction
KR102193264B1 (en) DC-DC Converter with Intelligent Controller

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination