Chip dynamic temperature adjusting method based on PID (Proportion Integration Differentiation) control algorithm

Abstract

The invention discloses a chip dynamic temperature adjusting method based on a PID control algorithm, and the method comprises the following steps: S1, collecting and processing multi-point temperature data, and obtaining a unified temperature vector; s2, performing thermal mode extraction, and calculating a target temperature deviation to generate an error signal; s3, constructing a segmented polymorphic PI D controller, and performing mode switching to calculate a proportional item; s4, generating an integral term for each thermal mode, and performing change rate processing to generate a differential term; s5, the proportion, the integral and the differential are combined, and the distributor outputs an actuator instruction; and S6, issuing an executor instruction, and executing amplitude limiting and slope constraint to complete adjustment. According to the invention, through segmented multi-state PI D control and a mode perception distributor, rapid response and stable adjustment of the chip temperature are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chip temperature management and control, and particularly relates to a chip dynamic temperature regulation method based on a PID control algorithm. BACKGROUND

[0002] At present, with the continuous improvement of chip integration and operating frequency, the heat generated by chips in high-performance computing and complex application scenarios has significantly increased. Excessive temperature not only affects the stability and operation accuracy of the chip, but also shortens the service life of the chip and even causes functional failure. In order to ensure the reliable operation of the chip, the temperature control method is generally used for dynamic adjustment in the prior art, and the PID control algorithm is widely used in chip temperature control systems due to its simple implementation and stable regulation. The common method is to collect temperature data of the chip surface or core area through a temperature sensor, compare the real-time temperature with the preset target temperature, use the PID controller to output a control quantity, and act on the fan, dynamic voltage frequency regulation unit or thermoelectric cooling unit to realize stable control of the chip temperature.

[0003] However, the PID control method in the prior art generally has the following deficiencies. On the one hand, the traditional PID algorithm mostly uses a single sensor or a simple multi-point average temperature as input, which is difficult to reflect the complex thermal distribution characteristics inside the chip, resulting in a lag in control response and poor regulation effect in the case of multi-region thermal coupling and rapid migration of hot spots. On the other hand, the conventional PID control parameters are fixed, and when the chip operating state changes from static stability to sudden high load or continuous high load, the parameters of the controller cannot be adaptively switched, which easily causes temperature overshoot or slow response. In addition, the integral link is prone to saturation, and the differential link is prone to amplify interference when there is measurement noise, further reducing the control accuracy.

[0004] The prior art cannot simultaneously consider fast response and steady-state accuracy, and lacks a detailed control mechanism for different working modes. This leads to insufficient robustness of chip temperature regulation under complex working conditions, low energy utilization efficiency, and even problems such as frequent action of actuators, increased energy consumption and noise.

[0005] Therefore, how to provide a chip dynamic temperature regulation method based on a PID control algorithm is a problem that those skilled in the art need to solve. SUMMARY

[0006] One purpose of the present application is to propose a chip dynamic temperature regulation method based on PID control algorithm, the present application extracts thermal modalities from a multi-point temperature vector and switches control parameters in combination with different operating states by introducing a segmented multi-modal PID controller, an independent modal integrator, fractional order differential processing and a modal perception distributor, and the dynamic temperature regulation process under various operating modes such as static stability, sudden high load and continuous high load is described in detail, which has the advantages of fast response speed, high steady-state accuracy, smooth actuator action and optimized energy consumption and noise.

[0007] According to the chip dynamic temperature regulation method based on the PID control algorithm, the method comprises the following steps:

[0008] S1, collecting real-time temperature data of a plurality of temperature sensors on a chip, processing the real-time temperature data to obtain a multi-point temperature vector;

[0009] S2, extracting thermal modalities from the multi-point temperature vector to form a plurality of thermal modality variables, setting a target temperature value, calculating the deviation of the thermal modality variables relative to the target temperature value to obtain a modal error signal;

[0010] S3, constructing a segmented multi-modal PID controller, switching between static stability mode, sudden high load mode and continuous high load mode according to the operating state of the chip, calling proportional parameters corresponding to the current mode, performing proportional operation on the modal error signal to obtain a modal proportional term;

[0011] S4, calling integral parameters and differential parameters corresponding to the current mode in the segmented multi-modal PID controller, configuring an independent integrator for each thermal modality to generate a modal integral term, and performing fractional order differential processing on the rate of change of the modal error signal in the differential operation process to generate a modal differential term;

[0012] S5, combining the modal proportional term, the modal integral term and the modal differential term to generate a modal control output signal, and inputting the modal control output signal into a modal perception distributor, generating corresponding actuator control instructions according to the sensitivity and constraint conditions of the actuator to different thermal modalities;

[0013] S6, issuing the actuator control instructions to a fan control unit, a dynamic voltage frequency regulation unit and a thermoelectric cooling unit, implementing amplitude limiting and slope constraint on the output process to complete the dynamic regulation of the chip temperature.

[0014] Optionally, the processing of the real-time temperature data specifically comprises:

[0015] The collected real-time temperature data of the temperature sensors is denoised to eliminate abnormal fluctuation data;

[0016] The denoised real-time temperature data is filtered to suppress high-frequency interference signals;

[0017] The filtered real-time temperature data is fused to form a unified multi-point temperature vector.

[0018] Optionally, S2 specifically includes:

[0019] S21. Read the multi-point temperature vector and extract the thermal modes of the multi-point temperature vector according to the pre-established and saved thermal mode basis vector set to obtain a number of thermal mode variables arranged in order.

[0020] S22. The target temperature value is converted into a modal target value that corresponds one-to-one with the thermal modal variable according to the preset modal mapping rule, so as to obtain a modal target sequence arranged with the same number.

[0021] S23. For each thermal modal variable, first calculate the initial difference between the thermal modal variable and the corresponding modal target value, and then perform the following processing in sequence:

[0022] Within the hysteresis interval formed by the first deviation threshold and the second deviation threshold, the deviation is set to zero. When the deviation exceeds the hysteresis interval, asymmetric deduction is performed according to different thresholds for positive and negative directions.

[0023] When an upward trend appears within several consecutive sampling periods and the trend intensity exceeds a preset trend threshold, a forward compensation amount is added to the deviation; when a downward trend appears and the trend intensity exceeds a preset trend threshold, a mitigation compensation amount is added to the deviation.

[0024] The deviation is scaled based on the sensor consistency evaluation coefficient corresponding to the thermal modal variable;

[0025] The amplitude of the scaled deviation is constrained according to preset upper and lower limits to obtain the modal deviation signal of the thermal modal variable;

[0026] S24. Combine the modal deviation signals of all thermal modes in the order of the thermal modal variables to form a modal error signal sequence, and output the modal error signal sequence.

[0027] Optionally, S3 specifically includes:

[0028] S31. Obtain the modal error signal sequence, establish a one-to-one mapping relationship with the corresponding thermal modal variables according to their serial numbers, and use it as the input of the piecewise multi-mode PID controller;

[0029] S32, construct the structure of the segmented polymorphic PID controller, the structure includes a state manager, a parameter library, a parameter interpolator, a mode residence manager, a small amplitude jitter suppressor, an abnormal switch protector and a proportional operator, the state manager determines a static stable mode, a sudden high load mode and a sustained high load mode, the parameter library presets a proportional parameter base value for each thermal mode variable in the three modes respectively;

[0030] S33, calculate two state determination indexes based on the mode error signal sequence by the state manager, wherein the error amplitude index is the maximum of the absolute amplitudes of all mode deviation signals at the same sampling time, and the error change rate index is the maximum of the change amplitudes of all mode deviation signals between adjacent sampling times;

[0031] S34, mode determination is performed in the following order:

[0032] When the error change rate index reaches the error change rate threshold, enter the sudden high load mode;

[0033] When the error amplitude index continuously reaches the high error amplitude threshold and the duration reaches the minimum residence count, enter the sustained high load mode;

[0034] When the error amplitude index is lower than the low error amplitude threshold and the error change rate index is lower than the error change rate threshold, enter the static stable mode;

[0035] When any of the above conditions is not met, keep the mode of the last sampling time, and form upper and lower limit areas by hysteresis threshold setting to limit frequent switching;

[0036] S35, after mode determination, perform proportional parameter generation processing, specifically:

[0037] The parameter library provides a proportional parameter base value corresponding to the current mode for each thermal mode variable, and the mode residence manager records the sampling count since the last mode switch;

[0038] The parameter interpolator performs time-based smooth interpolation between the proportional parameter base value of the last mode and the proportional parameter base value of the current mode;

[0039] The small amplitude jitter suppressor sets the proportional parameter adjustment coefficient to zero when the mode deviation signal is in a preset dead zone, and modifies the interpolated proportional parameter according to the sensor consistency evaluation coefficient to obtain the final proportional parameter of each thermal mode variable;

[0040] S36, the proportional operator calculates the mode proportional term for each thermal mode variable according to the linear relationship between the final proportional parameter and the corresponding mode deviation signal, and combines them in order to form a mode proportional term sequence.

[0041] Optionally, the S4 specifically comprises:

[0042] S41, acquiring a modal error signal sequence, a modal proportional term sequence, and integral and differential parameters corresponding to a current mode;

[0043] S42, establishing an independent integrator for each thermal modal variable, setting an integral initial value, an integral upper limit, an integral lower limit, an integral step, and an integral dead zone, and performing integral operation on the modal error signal of the thermal modal variable in a sampling order to obtain a modal integral term;

[0044] S43, performing integral anti-saturation processing on each thermal modal variable during the integral operation, setting the modal integral term to the integral upper limit when the accumulated modal integral term exceeds the integral upper limit, setting the modal integral term to the integral lower limit when the accumulated modal integral term is lower than the integral lower limit, recording an integral freezing flag, and suspending the integral accumulation of the thermal modal variable in the sampling interval where the integral freezing flag is recorded;

[0045] S44, calculating a rate of change of the modal error signal of each thermal modal variable, normalizing the rate of change according to a sampling period, and setting a rate of change dead zone and a rate of change hysteresis threshold, recording the rate of change as zero when the absolute value of the rate of change is within the rate of change dead zone, and updating a rate of change valid flag when the absolute value of the rate of change crosses the rate of change hysteresis threshold;

[0046] S45, performing fractional order differentiation processing on the rate of change sequence with the rate of change valid flag according to a preset fractional order, combining the differential parameter to obtain a modal differential term of each thermal modal variable, and recording the modal differential term as zero for the sampling points without the rate of change valid flag;

[0047] S46, combining the modal integral terms of all thermal modal variables in the order of the serial numbers of the thermal modal variables to form a modal integral term sequence, and combining the modal differential terms of all thermal modal variables in the order of the serial numbers of the thermal modal variables to form a modal differential term sequence.

[0048] Optionally, the S5 specifically comprises:

[0049] S51, acquiring a modal proportional term sequence, a modal integral term sequence, and a modal differential term sequence arranged in the order of the serial numbers of the thermal modal variables, and performing weighted superposition on the three terms corresponding to each thermal modal variable according to a pre-stored weighting coefficient and a fixed combination order to obtain a modal control output value, and combining the modal control output values in the order of the serial numbers to form a modal control output signal sequence;

[0050] S52, constructing a modal perception allocator, the modal perception allocator comprising a static parameter set and a running parameter set, the static parameter set comprising a set of actuators, a sensitivity coefficient table of modal and actuators, a cost coefficient table of actuators, an amplitude upper and lower limit table of actuators, and a maximum change rate table of actuators, the running parameter set comprising an actuator control quantity at a previous sampling time and a modal control output signal sequence at a current sampling time;

[0051] S53, at each sampling time, calculating, by the modal perception allocator, a set of intermediate control quantities based on the modal control output signal sequence, the sensitivity coefficient table, and the cost coefficient table, so as to minimize a sum of a modal fitting error and an actuator cost, and outputting the set of intermediate control quantities in a one-to-one correspondence with an actuator channel;

[0052] S54, sequentially performing, on the set of intermediate control quantities, a change rate constraint and an amplitude constraint;

[0053] In the change rate constraint, a difference between a control quantity of each actuator at the current sampling time and a control quantity of each actuator at the previous sampling time is compared, and the difference is limited in a range of a maximum change rate allowed by the actuator;

[0054] In the amplitude constraint, the control quantity is projected within an amplitude upper and lower limit of each actuator, and when any actuator channel touches a boundary in any constraint, the set of intermediate control quantities of all actuator channels is scaled in a preset rule and a boundary touch flag and a scaling factor are recorded;

[0055] S55, converting the set of control quantities into a set of executable control instructions, wherein a control instruction of the fan control unit is a target rotating speed or a duty ratio, a control instruction of the dynamic voltage and frequency adjustment unit is a target operating state or a duty ratio, and a control instruction of the thermoelectric cooling unit is a target driving power.

[0056] Optionally, the S6 specifically comprises:

[0057] S61, reading the set of control instructions and the set of actuator control quantities at the previous sampling time, and loading the boundary touch flag and the scaling factor;

[0058] S62, setting a minimum residence count and a maximum switching frequency for the fan control unit, the dynamic voltage and frequency adjustment unit, and the thermoelectric cooling unit, and converting the set of control instructions into a target control quantity format, wherein the fan control unit adopts a duty ratio format, the dynamic voltage and frequency adjustment unit adopts a target operating state, and the thermoelectric cooling unit adopts a target driving power format;

[0059] S63, performing a slope constraint, comparing a difference between a current target control quantity of each actuator and a control quantity of each actuator at the previous sampling time, limiting the difference in a range of a maximum change rate allowed by the actuator, and obtaining a set of control quantities after the slope constraint;

[0060] S64, perform amplitude limiting, limit the control amount of each actuator within the corresponding amplitude upper and lower limit range, when any actuator touches the boundary, according to the boundary touch flag and the scaling factor, all actuator control amounts are scaled according to the preset rule, forming a set of executable control amounts;

[0061] S65, issue the set of executable control amounts to the fan control unit, dynamic voltage frequency adjustment unit and thermoelectric cooling unit, perform minimum residence time check on the dynamic voltage frequency adjustment unit, perform soft start and soft stop on the fan control unit and the thermoelectric cooling unit, collect execution feedback and update the actuator control amount set and mode residence count at the last sampling time, detect faults and set the fault flag, and output the degradation instruction set.

[0062] The beneficial effects of the present application are:

[0063] The present application introduces a chip dynamic temperature regulation method based on PID control algorithm, solves the problem of multi-point temperature control response lag and inability to adapt to different working modes in the prior art. By introducing segmented multi-state control and mode perception distribution mechanism in the PID controller, the present application can flexibly adjust the control parameters for different chip working states, ensuring the efficiency and accuracy of the temperature regulation process. Especially when the chip load switches from static stable state to burst high load or continuous high load, the controller can automatically switch parameters, avoiding the problem of temperature overshoot or slow response of traditional PID control when the load changes greatly.

[0064] The present application introduces an anti-saturation mechanism in the integral element, preventing system instability caused by excessive accumulation of the integral term in traditional PID control. Through fractional order differential processing, the influence of high-frequency noise in the temperature signal on control accuracy is reduced, further improving the stability of temperature regulation. The mode perception distributor intelligently distributes according to the relationship between thermal mode and actuator sensitivity, making the control amount of the actuator more accurate and avoiding the problem of excessive work or frequent switching of the actuator, significantly improving the energy efficiency ratio of the system.

[0065] The present application can realize accurate control of chip temperature, while improving the robustness and adaptability of the system, reducing excessive energy consumption and noise, and improving the operation reliability and life of the chip. BRIEF DESCRIPTION OF DRAWINGS

[0066] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0067] Figure 1 A flowchart of a chip dynamic temperature regulation method based on a PID control algorithm is proposed in the present application.

[0068] Figure 2 A sectional multi-state PID controller of a chip dynamic temperature regulation method based on a PID control algorithm. DETAILED DESCRIPTION

[0069] The application will be described in further detail with reference to the drawings. These drawings are simplified schematic diagrams and only show the basic structure of the application in a schematic manner, and thus only show the components relevant to the application.

[0070] REFERENCE Figure 1 and Figure 2 A chip dynamic temperature regulation method based on a PID control algorithm, comprising the following steps:

[0071] S1, collecting real-time temperature data of multiple temperature sensors on a chip, processing the real-time temperature data to obtain a multi-point temperature vector;

[0072] S2, extracting thermal modalities from the multi-point temperature vector to form a plurality of thermal modality variables, setting a target temperature value, calculating the deviation of the thermal modality variables relative to the target temperature value to obtain a modal error signal;

[0073] S3, constructing a sectional multi-state PID controller, switching between a static stable mode, a sudden high load mode and a continuous high load mode according to the running state of the chip, calling proportional parameters corresponding to the current mode, performing proportional operation on the modal error signal to obtain a modal proportional term;

[0074] S4, calling integral parameters and differential parameters corresponding to the current mode in the sectional multi-state PID controller, configuring an independent integrator for each thermal modality to generate a modal integral term, and performing fractional order differential processing on the rate of change of the modal error signal in the differential operation process to generate a modal differential term;

[0075] S5, combining the modal proportional term, the modal integral term and the modal differential term to generate a modal control output signal, and inputting the modal control output signal into a modal perception distributor to generate corresponding actuator control instructions according to the sensitivity and constraint conditions of the actuator for different thermal modalities;

[0076] S6, issuing the actuator control instructions to a fan control unit, a dynamic voltage frequency regulation unit and a thermoelectric cooling unit, implementing amplitude limiting and slope constraint on the output process to complete dynamic regulation of the chip temperature.

[0077] In this embodiment, the processing of the real-time temperature data specifically includes:

[0078] The collected real-time temperature data of the temperature sensors is denoised to eliminate abnormal fluctuation data.

[0079] The real-time temperature data after denoising is filtered to suppress high-frequency interference signals.

[0080] The filtered real-time temperature data is fused to form a unified multi-point temperature vector.

[0081] In the embodiment, S2 specifically comprises:

[0082] S21, reading a multi-point temperature vector, and performing thermal mode extraction on the multi-point temperature vector according to a set of thermal mode basis vectors, to obtain a plurality of thermal mode variables arranged in sequence;

[0083] S22, converting a target temperature value into a mode target value corresponding to the thermal mode variable according to a preset mode mapping rule, to obtain a mode target sequence arranged in the same sequence number;

[0084] S23, for each thermal mode variable, first calculating an initial difference between the thermal mode variable and the corresponding mode target value, and then sequentially performing the following processing:

[0085] Setting the deviation to zero within a hysteresis interval formed by the first deviation threshold and the second deviation threshold, and performing asymmetric deduction according to different thresholds in positive and negative directions when the deviation exceeds the hysteresis interval;

[0086] When an upward trend appears within a plurality of consecutive sampling periods and the trend intensity exceeds a preset trend threshold, adding a forward-looking compensation amount to the deviation, and when a downward trend appears and the trend intensity exceeds the preset trend threshold, adding a mitigation compensation amount to the deviation;

[0087] Scaling the deviation according to a sensor consistency evaluation coefficient corresponding to the thermal mode variable;

[0088] Amplitude-constraining the scaled deviation according to preset upper and lower limits, to obtain a mode deviation signal of the thermal mode variable;

[0089] S24, combining the mode deviation signals of all thermal modes in the sequence number order of the thermal mode variables to form a mode error signal sequence, and outputting the mode error signal sequence.

[0090] The application introduces hysteresis interval control, trend compensation, sensor consistency correction and amplitude constraint mechanism in the modal error signal generation process, effectively solves the problems of large noise, slow response and insufficient precision of the deviation signal in the prior art. Through thermal modal extraction and target value mapping, temperature control is expanded from single-point measurement to multi-point fusion, which can more truly reflect the thermal distribution state inside the chip. The introduction of hysteresis interval and asymmetric deduction makes small fluctuations not trigger invalid response, while ensuring that the positive and negative deviations are reasonably corrected at different amplitudes, avoiding misoperation. The trend compensation mechanism corrects the deviation in advance or delays by identifying continuous rising or falling trend, so that the controller has certain foresight and buffering, and the dynamic adaptability is improved. The scaling mechanism of the sensor consistency evaluation coefficient can weaken the influence of abnormal sensor data and improve the reliability of the modal deviation. The final amplitude constraint ensures that the deviation signal is always within a controllable range. The application realizes more smooth, stable and accurate modal error signal generation, and lays a foundation for accurate adjustment of the subsequent controller.

[0091] In the embodiment, the S3 specifically includes:

[0092] S31, a modal error signal sequence is obtained, a one-to-one mapping relationship corresponding to the thermal modal variable is established according to the serial number of the thermal modal variable, and is taken as the input of the segmented multi-modal PID controller;

[0093] S32, the structure of the segmented multi-modal PID controller is constructed, the structure includes a state manager, a parameter library, a parameter interpolator, a mode residence manager, a small amplitude jitter suppressor, an abnormal switching protector and a proportional operator, the state manager determines a static stable mode, a sudden high load mode and a continuous high load mode, and the parameter library presets a proportional parameter base value for each thermal modal variable in the three modes;

[0094] S33, two state determination indexes are calculated by the state manager based on the modal error signal sequence, wherein the error amplitude index is the maximum of the absolute amplitudes of all modal deviation signals at the same sampling time, and the error change rate index is the maximum of the change amplitudes of all modal deviation signals between adjacent sampling time;

[0095] S34, mode determination is performed in the following order:

[0096] When the error change rate index reaches the error change rate threshold, the sudden high load mode is entered;

[0097] When the error amplitude index continuously reaches the high error amplitude threshold and the duration reaches the minimum residence count, the continuous high load mode is entered;

[0098] When the error amplitude index is lower than the low error amplitude threshold and the error change rate index is lower than the error change rate threshold, the static stable mode is entered;

[0099] When any of the above conditions is not met, the mode of the last sampling time is maintained, and upper and lower limit zones are formed by hysteresis threshold setting to limit frequent switching;

[0100] S35, after the mode is determined, a proportional parameter generation process is performed, specifically:

[0101] The parameter library provides a proportional parameter base value corresponding to the current mode for each thermal modal variable, and the mode residence manager records the sampling count since the last mode switching;

[0102] The parameter interpolator performs time-based smooth interpolation between the proportional parameter base value of the last mode and the proportional parameter base value of the current mode;

[0103] The small amplitude jitter suppressor sets the proportional parameter adjustment coefficient to zero when the modal deviation signal is in the preset dead zone, and modifies the interpolated proportional parameter according to the sensing consistency evaluation coefficient to obtain the final proportional parameter of each thermal modal variable;

[0104] S36, the proportional operator calculates the modal proportional term according to the linear relationship between the final proportional parameter and the corresponding modal deviation signal for each thermal modal variable, and combines to form a modal proportional term sequence in the order of the serial number of the thermal modal variable.

[0105] The present application introduces state manager, parameter library, parameter interpolator, mode residence manager, small amplitude jitter suppressor and abnormal switching protector in the segmented multi-modal PID controller, realizes the dynamic adaptive adjustment of temperature control parameters in different working modes of the chip. Through the joint determination of error amplitude index and error change rate index, the static stability, sudden high load and continuous high load mode can be accurately identified, and the lag and instability problems of traditional PID control in state switching are avoided. The parameter interpolator combined with the mode residence manager makes the proportional parameter smooth transition and suppresses the mutation caused by frequent switching. The addition of the small amplitude jitter suppressor and the sensing consistency correction mechanism effectively eliminates noise interference and improves the reliability of multi-sensor fusion. The present application realizes accurate generation of modal proportional parameters and stable output of modal proportional terms, balances between fast response and steady state accuracy, and significantly improves the robustness and energy efficiency of chip dynamic temperature regulation.

[0106] In the embodiment, the S4 specifically includes:

[0107] S41, the modal error signal sequence, the modal proportional term sequence and the integral parameter and the differential parameter corresponding to the current mode are obtained;

[0108] S42, for each thermal modal variable, an independent integrator is established, the integral initial value, the integral upper limit, the integral lower limit, the integral step and the integral dead zone are set, the modal error signal of the thermal modal variable is integrated according to the sampling sequence, and the modal integral term is obtained;

[0109] S43, during the integral operation, the integral anti-saturation processing is performed on each thermal modal variable, when the accumulated modal integral term exceeds the integral upper limit, the modal integral term is set to the integral upper limit, when the accumulated modal integral term is lower than the integral lower limit, the modal integral term is set to the integral lower limit, the integral freezing flag is recorded, and the integral addition of the thermal modal variable is suspended in the sampling interval of the recorded integral freezing flag;

[0110] S44, the change rate of the modal error signal of each thermal modal variable is calculated, the change rate is normalized according to the sampling period, and the change rate dead zone and the change rate hysteresis threshold are set, when the absolute value of the change rate is in the change rate dead zone, the change rate is recorded as zero, and when the absolute value of the change rate crosses the change rate hysteresis threshold, the change rate effective flag is updated;

[0111] S45, according to the preset fractional order number of the current mode, the fractional order differential processing is performed on the change rate sequence with the change rate effective flag, the modal differential term of each thermal modal variable is obtained by combining the differential parameters, and the modal differential term is recorded as zero for the sampling point without the change rate effective flag;

[0112] S46, the modal integral terms of all thermal modal variables are combined to form a modal integral term sequence in the order of the serial number of the thermal modal variables, and the modal differential terms of all thermal modal variables are combined to form a modal differential term sequence in the order of the serial number of the thermal modal variables.

[0113] In the integral and differential links, the independent integrator, the integral anti-saturation mechanism and the fractional order differential processing are introduced, which breaks through the limitation of the control precision of the traditional PID in the complex thermal environment. By establishing an independent integrator for each thermal modal variable and setting the upper and lower limits, the step and the dead zone, the saturation problem caused by excessive accumulation of the integral term is avoided, and the system stability is maintained during the integral freezing. Combined with the change rate normalization and the hysteresis threshold determination, the false judgment caused by the sensor noise is effectively suppressed. Further, the fractional order differential is used instead of the conventional integer order differential, so that the processing of the change rate of the modal error signal is more flexible, the fast response is retained, and the risk of high-frequency noise amplification is reduced. Through these innovative measures, the present application can output smoother and more reliable modal integral terms and differential terms, which can exhibit excellent control effect in rapid thermal change and long-term stable operation, and significantly improve the stability and robustness of the chip temperature regulation.

[0114] In the embodiment, the S5 specifically comprises:

[0115] S51, acquire the modal proportional term sequence, the modal integral term sequence and the modal differential term sequence arranged according to the thermal modal variable serial number, and perform weighted superposition on the three terms corresponding to each thermal modal variable according to the pre-stored weighting coefficients and fixed combination sequence to obtain the modal control output value, and combine the modal control output values according to the serial numbers to form the modal control output signal sequence;

[0116] S52, construct a modal perception distributor, the modal perception distributor including a static parameter set and a running parameter set, the static parameter set including an actuator set, a sensitivity coefficient table of the modal and the actuator, a cost coefficient table of the actuator, an amplitude upper and lower limit table of the actuator and a maximum change rate table of the actuator, and the running parameter set including the actuator control quantity at the last sampling time and the modal control output signal sequence at the current sampling time;

[0117] S53, at each sampling time, calculate the intermediate control quantity set that minimizes the sum of the modal fitting error and the actuator cost weighted by the modal perception distributor based on the modal control output signal sequence, the sensitivity coefficient table and the cost coefficient table, and output the intermediate control quantity set according to the one-to-one correspondence of the actuator channels;

[0118] S54, sequentially perform change rate constraint and amplitude constraint on the intermediate control quantity set:

[0119] In the change rate constraint, compare the control quantity difference of each actuator at the current sampling time and the last sampling time, and limit it within the maximum change rate range allowed by the actuator;

[0120] In the amplitude constraint, project the control quantity within the amplitude upper and lower limit of each actuator, and when any actuator channel touches the boundary in any constraint, proportionally scale the intermediate control quantity of all actuator channels according to the preset rule and record the boundary touch flag and the scaling factor;

[0121] S55, convert the control quantity set into an executable control instruction set, wherein the control instruction of the fan control unit is the target rotating speed or duty ratio, the control instruction of the dynamic voltage frequency regulation unit is the target operating state or duty ratio, and the control instruction of the thermoelectric cooling unit is the target driving power.

[0122] The application introduces a modal perception distributor and a multiple constraint mechanism in the mapping process of modal control output to actuators, solves the problems of coarse actuator distribution, high energy consumption and excessive action in the prior art. By weighting and superimposing the proportional, integral and differential terms, a complete modal control signal sequence is formed, and the sensitivity coefficient and cost coefficient of the actuator are combined, so that the adjustment behavior of each actuator matches its contribution to the thermal modal and the energy consumption weight, thereby avoiding the overload or invalid action of a single actuator. The intermediate control quantity is dynamically modified under the rate constraint and amplitude constraint to prevent oscillation and over-limit problems caused by rapid adjustment, and through the introduction of boundary touch flags and scaling factors, the cooperative protection between actuators is realized. The final generated instruction set can realize division and complement between different actuators, taking into account fast response and energy optimization. The innovation of the application lies in the distribution strategy of modal perception and multiple constraints, which makes the temperature regulation process more stable, energy-saving and controllable, greatly improving the robustness of the system and the service life of the actuator.

[0123] In the embodiment, S6 specifically includes:

[0124] S61, read the control instruction set and the actuator control quantity set at the last sampling time, load the boundary touch flag and the scaling factor;

[0125] S62, set the minimum residence count and the maximum switching frequency for the fan control unit, the dynamic voltage frequency adjustment unit and the thermoelectric cooling unit, and convert the control instruction set into a target control quantity format, wherein the fan control unit adopts a duty cycle format, the dynamic voltage frequency adjustment unit adopts a target operating state, and the thermoelectric cooling unit adopts a target driving power format;

[0126] S63, perform slope constraint, compare the difference between the current target control quantity of each actuator and the control quantity at the last sampling time, limit the difference within the maximum change rate range allowed by the actuator, and obtain the control quantity set after slope constraint;

[0127] S64, perform amplitude limiting, limit the control quantity of each actuator within the corresponding amplitude upper and lower limit range, when any actuator touches the boundary, proportionally scale all actuator control quantities according to the boundary touch flag and the scaling factor according to the preset rule, and form an executable control quantity set;

[0128] S65, issue the executable control quantity set to the fan control unit, the dynamic voltage frequency adjustment unit and the thermoelectric cooling unit, perform minimum residence time check on the dynamic voltage frequency adjustment unit, perform soft start and soft stop on the fan control unit and the thermoelectric cooling unit, collect execution feedback and update the actuator control quantity set and the mode residence count at the last sampling time, detect faults and set the fault flag, and output the degradation instruction set.

[0129] Example 1

[0130] To verify the feasibility of the application in practice, the application is applied to an information technology laboratory. Researchers conduct temperature management experiments on a 32-core server chip. The chip has a peak power consumption of nearly 240 watts when performing mixed load tasks (database queries and AI inference running in parallel), and the internal temperature of the chip rises significantly. The traditional PID method often has problems such as large temperature overshoot, frequent fan switching, and high noise when dealing with such rapidly changing loads, which seriously affects the stability and energy efficiency performance in the experimental environment.

[0131] To solve the above problems, researchers use the chip dynamic temperature regulation method based on the PID control algorithm proposed in the application. Twelve sensors are arranged inside the chip to collect real-time temperature data. After filtering and data fusion, a multi-point temperature vector that accurately reflects the local hot spots is obtained. The temperature vector is mapped to multiple thermal modal variables and compared with the target temperature value to generate a modal error signal.

[0132] The system calls a segmented polymorphic PID controller, which dynamically switches between static stability, sudden high load, and sustained high load modes according to the chip's running state. When the temperature rise rate exceeds the threshold, the controller enters the sudden high load mode, increases the proportional and derivative action, and achieves rapid response; when the temperature remains high, it enters the sustained high load mode, and the integrator is turned on to limit the amplitude to avoid overshoot.

[0133] Each thermal modal variable is configured with an independent integrator, and the integration process is limited by a boundary to avoid integration saturation. Meanwhile, fractional order differentiation is used in the differential section to smooth the trend of the error signal and reduce the differential amplification effect caused by sensor noise.

[0134] The modal perception distributor reasonably allocates the control signal to the fan control unit, dynamic voltage frequency adjustment unit, and thermoelectric cooling unit according to the sensitivity relationship between the thermal modal variable and the actuator. Compared with the traditional method, the application avoids overloading a single actuator and achieves coordinated regulation of multiple actuators. For example, when a hot spot appears at the edge of the chip, the distributor moderately increases the fan speed, while when the core area temperature rises rapidly, the dynamic voltage frequency adjustment unit intervenes to reduce power consumption.

[0135] The actuator control instructions are sent to each actuator after slope constraint and amplitude limiting processing. In actual experiments, the fan speed is limited between 2800 and 3500 revolutions per minute, the dynamic voltage frequency adjustment unit is reduced by about 0.2 GHz in the high load stage, and the thermoelectric cooling unit is enabled with a power not exceeding 75% of the rated value. After 20 minutes of pressure test, the chip temperature is stabilized below 83℃, the fluctuation is less than ±2℃, the noise is reduced by about 5 decibels, and the system power consumption is reduced by about 15 watts. The results show that the application effectively improves the energy efficiency and user experience while maintaining stable operation of the chip.

[0136] Table 1 Comparison of experimental data of the method of the application and the traditional PID method

[0137]

[0138]

[0139] As can be seen from the data in Table 1, in the static stable running scene, the average temperature of the traditional PID control is maintained at 72℃, the temperature fluctuation range is ±3.0℃, the maximum overshoot is 5.0℃, the average fan speed is about 2800 revolutions per minute, the system power consumption is 230 watts, and the noise is 44 decibels. After using the method of the application, the average temperature is reduced to 70℃, the fluctuation is reduced to ±1.2℃, the maximum overshoot is only 2.0℃, the fan speed is reduced to 2500 revolutions per minute, the system power consumption is reduced to 220 watts, and the noise is reduced to 39 decibels, indicating that the application can provide smoother temperature control and lower energy consumption and noise in the static stable state.

[0140] In the sudden high load scene, the chip temperature rapidly rises under the traditional PID control, the average temperature reaches 91℃, the fluctuation range is ±4.2℃, the maximum overshoot value is as high as 8.0℃, the average fan speed needs to be increased to 4000 revolutions per minute, the power consumption increases to 245 watts, and the noise reaches 56 decibels. In contrast, the method of the application controls the average temperature at 83℃, reduces the temperature fluctuation to ±1.8℃, reduces the maximum overshoot to 3.2℃, reduces the fan speed to 3200 revolutions per minute, reduces the system power consumption to 230 watts, and controls the noise within 50 decibels. This shows that when dealing with rapid load changes, the application can effectively suppress temperature overshoot and excessive fluctuation, and reduce excessive dependence on actuators.

[0141] In the continuous high load scenario, the average temperature of the traditional PID control is kept at 86℃, the fluctuation is ±3.8℃, the maximum overshoot is 7.0℃, the fan speed reaches 3700 rpm, the system power consumption is 240 watts, and the noise is 54 decibels. The method of the present application reduces the average temperature to 82℃, reduces the fluctuation range to ±1.6℃, controls the maximum overshoot to 2.9℃, reduces the fan speed to 3100 rpm, reduces the system power consumption to 225 watts, and reduces the noise to 48 decibels. This shows that the present application can also maintain stable temperature control under long-time high-load operation, reduce energy consumption and noise accumulation, and prolong the service life of the chip and the actuator.

[0142] The method of the present application is superior to the traditional PID method in terms of average temperature control, temperature fluctuation suppression, overshoot reduction, actuator speed optimization, and energy consumption and noise reduction, improves the precision and stability of chip dynamic temperature regulation, and verifies the practicability and superiority of the method.

[0143] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A chip dynamic temperature regulation method based on a PID control algorithm, characterized in that, The method comprises the following steps: S1, collecting real-time temperature data of multiple temperature sensors on a chip, processing the real-time temperature data to obtain a multi-point temperature vector; S2, extracting thermal modalities from the multi-point temperature vector to form a plurality of thermal modality variables, setting a target temperature value, calculating the deviation of the thermal modality variables relative to the target temperature value to obtain a modality error signal; S3, constructing a segmented multi-modal PID controller, switching between a static stable mode, a sudden high load mode and a continuous high load mode according to the running state of the chip, calling proportional parameters corresponding to the current mode, performing proportional operation on the modality error signal to obtain a modality proportional term; S4, calling integral parameters and differential parameters corresponding to the current mode in the segmented multi-modal PID controller, configuring an independent integrator for each thermal modality to generate a modality integral term, and performing fractional order differential processing on the rate of change of the modality error signal in the differential operation process to generate a modality differential term; S5, combining the modality proportional term, the modality integral term and the modality differential term to generate a modality control output signal, and inputting the modality control output signal into a modality-aware distributor to generate corresponding actuator control instructions according to the sensitivity and constraint conditions of the actuators for different thermal modalities; S6, issuing the actuator control instructions to a fan control unit, a dynamic voltage frequency adjustment unit and a thermoelectric cooling unit, implementing amplitude limiting and slope constraint on the output process to complete dynamic adjustment of the chip temperature.

2. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, The processing of the real-time temperature data specifically includes: performing denoising processing on the collected real-time temperature data of the temperature sensors to eliminate abnormal fluctuation data; performing filtering processing on the denoised real-time temperature data to suppress high-frequency interference signals; performing fusion operation on the filtered real-time temperature data to form a unified multi-point temperature vector.

3. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, The S2 specifically includes: S21, reading the multi-point temperature vector, and extracting thermal modalities from the multi-point temperature vector according to a set of thermal modality basis vectors to obtain a plurality of thermal modality variables arranged in order; S22, converting the target temperature value into a modality target value corresponding to the thermal modality variables according to a preset modality mapping rule to obtain a modality target sequence arranged in the same sequence number; S23, for each thermal modality variable, first calculating the initial difference between the thermal modality variable and the corresponding modality target value, and then sequentially performing the following processing: zeroing the deviation within a hysteresis interval formed by a first deviation threshold and a second deviation threshold, and performing asymmetric deduction according to different threshold values in positive and negative directions when the deviation exceeds the hysteresis interval; when an upward trend appears within a plurality of consecutive sampling periods and the trend intensity exceeds a preset trend threshold, adding a forward compensation amount to the deviation, and when a downward trend appears and the trend intensity exceeds the preset trend threshold, adding a mitigation compensation amount to the deviation; scaling the deviation according to the sensor consistency evaluation coefficient corresponding to the thermal modality variable; performing amplitude constraint on the scaled deviation according to the preset upper and lower limits to obtain a modality deviation signal of the thermal modality variable; S24, combining the modality deviation signals of all thermal modalities in the sequence number order of the thermal modality variables to form a modality error signal sequence, and outputting the modality error signal sequence.

4. The chip dynamic temperature regulation method based on PID control algorithm of claim 1, wherein, The S3 specifically comprises: S31, acquiring a modal error signal sequence, establishing a one-to-one mapping relationship corresponding to the sequence number of the thermal modal variable, and taking it as the input of the segmented multi-modal PID controller; S32, constructing the structure of the segmented multi-modal PID controller, the structure comprising a state manager, a parameter library, a parameter interpolator, a mode residence manager, a small amplitude jitter suppressor, an abnormal switching protector, and a proportional operator, the state manager determining a static stability mode, a sudden high load mode, and a sustained high load mode, the parameter library pre-setting proportional parameter base values for each thermal modal variable in the three modes; S33, calculating two state judgment indexes based on the modal error signal sequence by the state manager, wherein the error amplitude index is the maximum of the absolute amplitude of all modal deviation signals at the same sampling time, and the error change rate index is the maximum of the change amplitude of all modal deviation signals between adjacent sampling times; S34, performing mode judgment in the following order: When the error change rate index reaches the error change rate threshold, enter the sudden high load mode; When the error amplitude index continuously reaches the high error amplitude threshold and the duration reaches the minimum residence count, enter the sustained high load mode; When the error amplitude index is lower than the low error amplitude threshold and the error change rate index is lower than the error change rate threshold, enter the static stability mode; When none of the above conditions is met, maintain the mode of the previous sampling time, and form upper and lower limit zones by hysteresis threshold setting to limit frequent switching; S35, after mode determination, performing proportional parameter generation processing, specifically: The parameter library provides proportional parameter base values corresponding to the current mode for each thermal modal variable, and the mode residence manager records the sampling count since the last mode switch; The parameter interpolator performs time-based smooth interpolation between the proportional parameter base value of the previous mode and the proportional parameter base value of the current mode; The small amplitude jitter suppressor sets the proportional parameter adjustment coefficient to zero when the modal deviation signal is in a preset dead zone, and modifies the interpolated proportional parameter according to the sensing consistency evaluation coefficient to obtain the final proportional parameter of each thermal modal variable; S36, the proportional operator calculates the modal proportional term for each thermal modal variable according to the linear relationship between the final proportional parameter and the corresponding modal deviation signal, and combines them in the order of the sequence number of the thermal modal variable to form a modal proportional term sequence.

5. The chip dynamic temperature regulation method based on PID control algorithm of claim 1, wherein, The S4 specifically comprises: S41, acquiring a modal error signal sequence, a modal proportional term sequence, and integral and differential parameters corresponding to the current mode; S42, for each thermal modal variable, establishing an independent integrator, setting the integral initial value, the integral upper limit, the integral lower limit, the integral step, and the integral dead zone, and performing integral operation on the modal error signal of the thermal modal variable in the sampling order to obtain the modal integral term; S43, performing integral anti-saturation processing on each thermal modal variable during integral operation, when the accumulated modal integral term exceeds the integral upper limit, setting the modal integral term to the integral upper limit, when the accumulated modal integral term is lower than the integral lower limit, setting the modal integral term to the integral lower limit, recording the integral freezing flag, and pausing the integral accumulation of the thermal modal variable in the sampling interval where the integral freezing flag is recorded; S44, calculating the rate of change of the modal error signal of each thermal modal variable, normalizing the rate of change according to the sampling period, and setting the rate of change dead zone and the rate of change hysteresis threshold, when the absolute value of the rate of change is within the rate of change dead zone, recording the rate of change as zero, and when the absolute value of the rate of change crosses the rate of change hysteresis threshold, updating the rate of change valid flag; S45, according to the preset fractional order selected according to the current mode, performing fractional order differentiation processing on the rate of change sequence of each thermal modal variable, and combining the differential parameters to obtain the modal differential term of each thermal modal variable, and setting the modal differential term to zero for the sampling points without the rate of change valid flag; S46, combining the modal integral terms of all thermal modal variables in the order of the serial numbers of the thermal modal variables to form a modal integral term sequence, and combining the modal differential terms of all thermal modal variables in the order of the serial numbers of the thermal modal variables to form a modal differential term sequence.

6. The chip dynamic temperature regulation method based on PID control algorithm of claim 1, wherein, The S5 specifically comprises: S51, obtaining the modal proportional term sequence, the modal integral term sequence and the modal differential term sequence arranged according to the serial numbers of the thermal modal variables, and performing weighted superposition on the three terms corresponding to each thermal modal variable according to the pre-stored weighting coefficients and fixed combination order to obtain a modal control output value, and combining the modal control output values according to the serial numbers to form a modal control output signal sequence; S52, constructing a modal perception distributor, the modal perception distributor comprising a static parameter set and a running parameter set, the static parameter set comprising an actuator set, a modal and actuator sensitivity coefficient table, an actuator cost coefficient table, an actuator amplitude upper and lower limit table and an actuator maximum rate table, and the running parameter set comprising an actuator control quantity at the last sampling time and a modal control output signal sequence at the current sampling time; S53, at each sampling time, calculating, by the modal perception distributor, an intermediate control quantity set that minimizes the sum of the modal fitting error and the actuator cost weighted based on the modal control output signal sequence, the sensitivity coefficient table and the cost coefficient table, and outputting the intermediate control quantity set according to the one-to-one correspondence of the actuator channels; S54, sequentially performing rate of change constraint and amplitude constraint on the intermediate control quantity set: In the rate of change constraint, comparing the control quantity difference of each actuator at the current sampling time and the last sampling time, and limiting within the maximum rate of change allowed by the actuator; In the amplitude constraint, projecting the control quantity within the amplitude upper and lower limit of each actuator, when any actuator channel touches the boundary in any constraint, proportionally scaling the intermediate control quantities of all actuator channels according to a preset rule and recording the boundary touch flag and the scaling factor; S55, converting the control quantity set into an executable control instruction set, wherein the control instruction of the fan control unit is target speed or duty ratio, the control instruction of the dynamic voltage frequency regulation unit is target operating state or duty ratio, and the control instruction of the thermoelectric cooling unit is target driving power.

7. The chip dynamic temperature regulation method based on PID control algorithm of claim 1, wherein, The S6 specifically comprises: S61, reading the control instruction set and the actuator control quantity set at the previous sampling time, loading the boundary touch flag and the scaling factor; S62, setting the minimum residence count and the maximum switching frequency for the fan control unit, the dynamic voltage frequency regulation unit and the thermoelectric cooling unit, and converting the control instruction set into a target control quantity format, wherein the fan control unit adopts a duty ratio format, the dynamic voltage frequency regulation unit adopts a target operating state, and the thermoelectric cooling unit adopts a target driving power format; S63, performing slope constraint, comparing the difference between the current target control quantity of each actuator and the control quantity at the previous sampling time, and limiting the difference within the maximum change rate range allowed by the actuator to obtain a control quantity set after slope constraint; S64, performing amplitude limiting, limiting the control quantity of each actuator within the upper and lower limits of the corresponding amplitude, when any actuator touches the boundary, performing equal proportion scaling on all actuator control quantities according to the boundary touch flag and the scaling factor according to a preset rule to form an executable control quantity set; S65, delivering the executable control quantity set to the fan control unit, the dynamic voltage frequency regulation unit and the thermoelectric cooling unit, performing minimum residence time check on the dynamic voltage frequency regulation unit, performing soft start and soft stop on the fan control unit and the thermoelectric cooling unit, collecting execution feedback and updating the actuator control quantity set and the mode residence count at the previous sampling time, detecting faults and setting a fault flag, and outputting a degradation instruction set.

Images