Model inference dynamic frequency reduction temperature control safe operation system and method adapting to vehicle engine temperature threshold

Abstract

The application discloses a model inference dynamic frequency reduction temperature control safe operation system and method suitable for adapting to vehicle machine temperature threshold, belongs to the technical field of vehicle thermal management, and comprises a data acquisition module, which is used for acquiring temperature, including acquiring the temperature of the CPU of the vehicle machine, and performing relevant processing on the acquired temperature; a data processing module, the data processing module comprising a dynamic threshold calibration unit and an acquisition unit; the dynamic threshold calibration unit responds to the data acquisition module, and is used for performing dynamic threshold calibration processing on the acquired temperature, including: acquiring relevant data, the relevant data including the general properties of the CPU factory calibration. The application adopts a millisecond-level computing power-temperature linkage detection mechanism, and only triggers frequency reduction when two conditions of temperature variation exceeding the standard and computing power increase exceeding 10% are simultaneously met, so that the waste of computing power caused by false frequency reduction is eliminated, the temperature rise risk is predicted in advance, the threshold is automatically iterated and calibrated, the supply of computing power for high-safety intelligent driving reasoning tasks is ensured, and the driving function safety is met.

Description

Technical Field

[0001] This invention relates to the field of vehicle thermal management technology, and in particular to a model-based dynamic frequency reduction temperature control system and method for safe operation that adapts to vehicle temperature thresholds. Background Technology

[0002] Currently, mass-produced smart vehicle systems are generally equipped with in-vehicle SoCs with computing power exceeding 100 TOPS. When continuously running large-scale environmental perception models and cockpit-driver fusion inference tasks, the power consumption of the core computing unit can instantly exceed 40W, and the chip temperature can rise to over 90℃ in a short period of time. In actual situations, temperature monitoring is necessary, and the frequency of the vehicle system should be reduced according to temperature changes.

[0003] Regarding this research, application CN202510128402.8 provides a method, system, storage medium, and electronic device for temperature control of a vehicle control unit. This technical solution includes: acquiring temperature and time parameters of the ECU components and motor; calculating the initial temperature at power-on; calculating the heat dissipation and heat gain of the ECU components and motor; estimating the temperature at the next moment based on the initial temperature; setting a temperature threshold; comparing the estimated temperature value with the temperature threshold to obtain a temperature level; and determining whether temperature degradation processing is needed. This technical solution can reduce heat accumulation, thereby preventing damage to the ECU components and motor due to high temperatures.

[0004] Another application, CN202510341044.9, provides a method, system, device, and medium for adjusting the frequency of a smart cockpit processor. This technical solution includes: acquiring multimodal sensor data from an in-vehicle smart cockpit, as well as the processor's current load data and current temperature data; inputting this data into a load prediction model to output the processor's load prediction result; classifying the sensor data into various priority queues; and if target sensor data exists and is classified into a target priority queue, acquiring the delay time of the target priority queue. This technical solution adjusts the processor's computing frequency to increase the computing frequency and maintain it for a target duration.

[0005] However, the above technical solutions rely solely on a single instantaneous temperature value as the basis for frequency reduction, making them susceptible to interference from instantaneous environmental temperature fluctuations and sensor sampling noise, which can indiscriminately trigger forced frequency reduction. This can easily lead to driving safety hazards such as insufficient perception frame rate and excessive response latency. Furthermore, the above technical solutions only respond to the current single temperature data and cannot accumulate historical characteristics of abnormal jumps in computing power. When similar high-load risk scenarios recur in the future, they cannot predict and avoid them in advance, resulting in the inability to independently iterate and improve the temperature control adaptation accuracy throughout the entire life cycle. Summary of the Invention

[0006] In view of the problems existing in the field of vehicle thermal management technology, the present invention is proposed.

[0007] Therefore, one of the objectives of this invention is to provide a model inference dynamic frequency reduction and temperature control safe operation system and method adapted to vehicle temperature thresholds. It uses a millisecond-level computing power-temperature linkage detection mechanism to trigger frequency reduction only when both the temperature change exceeds the standard and the computing power increase exceeds 10% at the same time. This not only prevents the waste of computing power due to erroneous frequency reduction, but also predicts the risk of temperature rise in advance, and automatically iterates and calibrates the threshold to ensure the computing power supply for high-safety intelligent driving inference tasks and meet the requirements of driving function safety.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] On one hand, the present invention provides a model inference dynamic frequency reduction temperature control safe operation system adapted to vehicle temperature thresholds, comprising:

[0010] The data acquisition module is used to acquire temperature, including the temperature of the vehicle's CPU, and to process the acquired temperature.

[0011] The data processing module includes a dynamic threshold calibration unit and an acquisition unit;

[0012] The dynamic threshold calibration unit responds to the data acquisition module and performs dynamic threshold calibration processing on the acquired temperature, including:

[0013] Obtain relevant data, including the general attributes of the CPU as specified at the factory, including the maximum computing load;

[0014] Based on the temperature, the change in the temperature of the vehicle's CPU is obtained, and the computing load of the CPU is obtained according to the change, including obtaining the corresponding computing load based on every 10% increase in the CPU temperature.

[0015] Based on the computing power load obtained for every 10% increase in temperature, obtain the increase in computing power load for every 10% increase in temperature;

[0016] Based on a 10% increase in temperature, the temperature corresponding to an increase in computing load exceeding 10% is obtained, and this temperature is considered as Temperature I. Temperatures less than Temperature I by 10% are considered as reference temperatures.

[0017] The time taken to raise the reference temperature to temperature I is obtained, and the time is divided into 10 time nodes, each of which is 2 milliseconds. The change in the increase of computing load in each of the time nodes is obtained.

[0018] The time taken is regarded as a reference time, and a temperature threshold related to temperature is preset based on the reference time;

[0019] Based on the temperature threshold, the CPU frequency is reduced. If, at a future time, the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, the CPU frequency is reduced.

[0020] In the future, if the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, but the increase in computing load does not exceed 10%, then the CPU will not be downclocked.

[0021] Based on the temperature threshold, if at a future time, within the same time frame as the reference time, the CPU experiences a temperature increase that is less than the temperature threshold, then the temperature threshold will be updated and calibrated based on the actual temperature increase.

[0022] The acquisition unit is used to perform relevant acquisition.

[0023] In a preferred embodiment of the present invention, the relevant acquisition includes:

[0024] Within the reference time, the time corresponding to each 0.5°C increase in temperature is obtained, and this time is regarded as the control time;

[0025] At a future time, if, within the same timeframe as the control, the time required to increase the temperature by 0.5°C is less than the control time, an early warning will be issued;

[0026] Once the warning is issued, if the CPU frequency is reduced if the time taken is continuously lower than the control time and the temperature increase exceeds 3%, the CPU will be reduced.

[0027] In a preferred embodiment of the present invention, the dynamic threshold calibration unit acquires the change in the increase rate of computing load at each time node, as shown below:

[0028] , =0, 1, 2, ..., 10;

[0029] In the formula, Indicates the first Each time point is acquired;

[0030] This indicates the start time corresponding to the reference temperature;

[0031] Indicates the sampling point number;

[0032] ;

[0033] In the formula, Indicates the relationship with the first CPU computing load at each time point;

[0034] Indicates the first The number of effective computing units that the CPU is running at each time point;

[0035] This indicates the total number of computing units at full load, as specified by the CPU manufacturer.

[0036] As a preferred embodiment of the present invention, it further includes the following:

[0037] ; =0, 1, 2, ..., 10;

[0038] In the formula, Indicates the first The month-on-month increase in computing load for each interval;

[0039] This indicates the computing load at the previous time point.

[0040] In a preferred embodiment of the present invention: based on the calculation results, the temperature corresponding to the increase in computing load exceeding 10% is obtained, and relevant positioning is performed at all time points. The relevant positioning includes locating the key time points corresponding to the abnormal jumps in the increase of computing load, as shown below:

[0041] ;

[0042] In the formula, This represents the set of moments corresponding to abnormal jumps in the increase of computing load;

[0043] This represents the average increase in computing load over 10 time points.

[0044] This represents the standard deviation of the increase in computing load over 10 time points.

[0045] In a preferred embodiment of the present invention, based on the calculation results, the increase corresponding to the abnormal jump in the increase of computing load is obtained at all time points, and the increase is regarded as the risk increase. If the increase in the computing load of the CPU is the same as the risk increase at a future time, the CPU frequency is reduced.

[0046] In a preferred embodiment of the present invention, the verification after downclocking the CPU includes:

[0047] After frequency reduction, a verification cycle is defined as at least six consecutive control periods.

[0048] If the temperature increases by 0.5°C during each comparison time in the verification period, but the computing load does not increase during the verification period, then the CPU frequency reduction will be cancelled.

[0049] In a preferred embodiment of the present invention: based on the set of identified times corresponding to abnormal jumps in the increase of computing power load, the time interval in which this set of times falls within the reference time interval is obtained, and the time interval is regarded as the reference time interval. In the future, within any time interval that is the same as the reference time interval, if the increase in computing power load does not change abnormally within the time interval that is the same as the reference time interval, it is determined that the increase in computing power load will not be abnormal within this time interval.

[0050] On the other hand, the present invention provides a method for a model inference dynamic frequency reduction temperature control safety operation system adapted to the vehicle temperature threshold as described above, comprising the following steps:

[0051] Temperature is acquired, including the temperature of the vehicle's CPU, and the acquired temperature is processed accordingly.

[0052] The acquired temperature is calibrated using a dynamic threshold, including:

[0053] Obtain relevant data, including the general attributes of the CPU as specified at the factory, including the maximum computing load;

[0054] Based on the temperature, the change in the temperature of the vehicle's CPU is obtained, and the computing load of the CPU is obtained according to the change, including obtaining the corresponding computing load based on every 10% increase in the CPU temperature.

[0055] Based on the computing power load obtained for every 10% increase in temperature, obtain the increase in computing power load for every 10% increase in temperature;

[0056] Based on a 10% increase in temperature, the temperature corresponding to an increase in computing load exceeding 10% is obtained, and this temperature is considered as Temperature I. Temperatures less than Temperature I by 10% are considered as reference temperatures.

[0057] The time taken to raise the reference temperature to temperature I is obtained, and the time is divided into 10 time nodes, each of which is 2 milliseconds. The change in the increase of computing load in each of the time nodes is obtained.

[0058] The time taken is regarded as a reference time, and a temperature threshold related to temperature is preset based on the reference time;

[0059] Based on the temperature threshold, the CPU frequency is reduced. If, at a future time, the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, the CPU frequency is reduced.

[0060] In the future, if the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, but the increase in computing load does not exceed 10%, then the CPU will not be downclocked.

[0061] Based on the temperature threshold, if at a future time, within the same time frame as the reference time, the CPU experiences a temperature increase that is less than the temperature threshold, then the temperature threshold will be updated and calibrated based on the actual temperature increase.

[0062] To obtain the relevant information.

[0063] Beneficial effects:

[0064] 1. Based on the linkage change law of temperature and computing load, this invention collects the details of computing load increase in 10 high-precision 2-millisecond time nodes, generates dynamic temperature thresholds adapted to the working conditions, and establishes a dual judgment rule: even if the temperature rise exceeds the preset threshold within the same time period in the future, as long as the increase in computing load does not exceed 10%, the frequency reduction operation will not be performed; frequency reduction is only triggered when the temperature rise exceeds the standard and the computing load rises abnormally at the same time. This mechanism avoids invalid frequency reduction caused by instantaneous temperature fluctuations in the environment and sensor sampling noise, and ensures the continuity of computing power supply for high-priority vehicle inference tasks to the greatest extent.

[0065] 2. By using the reference time corresponding to each 0.5℃ temperature rise as a benchmark, an early warning is immediately triggered when the time for the same temperature rise is detected to be lower than the reference value. When the temperature rise rate exceeds the standard for multiple consecutive periods and the total temperature increase exceeds 3%, the frequency reduction is started in advance. The temperature rise control is upgraded from "passive post-event cooling" to "active prediction and intervention", which greatly extends the time margin of the system temperature control response and completely avoids the risk of runaway due to sudden temperature surge under high load.

[0066] 3. By verifying the subsequent effects of each frequency reduction operation, if the temperature maintains a fixed increase but the computing power does not increase additionally within 6 consecutive control time periods after the frequency reduction, the frequency reduction can be automatically lifted and the full computing power can be restored to avoid unnecessary computing power loss. On the other hand, the original temperature threshold can be continuously iterated and calibrated based on the actual temperature rise data of future actual operation. After long-term operation, the temperature control adaptation accuracy will continue to improve.

[0067] 4. Furthermore, the present invention can memorize the key time nodes and corresponding reference time intervals of abnormal changes in computing power load. In future operation, as long as no matching abnormal increase is detected within the corresponding time window, it can be determined that the computing power risk throughout the entire cycle is under control, thus avoiding the driving safety hazards caused by indiscriminate forced frequency reduction, such as insufficient perceived frame rate and excessive response latency. Attached Figure Description

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the modular structure of the dynamic frequency reduction temperature control safety operation system adapted to the vehicle temperature threshold model in this embodiment of the invention. Figure 2 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the method flow for step S20 in an embodiment of the present invention;

[0069] The diagram is labeled as follows: 110 - Data acquisition module; 120 - Data processing module; 1201 - Dynamic threshold calibration unit; 1202 - Acquisition unit. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0071] Because existing technologies rely solely on a single instantaneous temperature value as the sole criterion for frequency reduction, they are susceptible to interference from ambient instantaneous temperature fluctuations and sensor sampling noise, which indiscriminately trigger forced frequency reduction, resulting in insufficient sensing frame rate. Furthermore, they cannot accumulate historical characteristics of abnormal jumps in computing power, thus preventing the temperature control adaptation accuracy from being improved through iterative iterations throughout the entire lifecycle.

[0072] Based on this, the present invention proposes a model inference dynamic frequency reduction and temperature control safe operation system and method adapted to vehicle temperature threshold. It uses a millisecond-level computing power-temperature linkage detection mechanism to trigger frequency reduction only when both the temperature change exceeds the standard and the computing power increase exceeds 10% at the same time. This not only prevents the waste of computing power due to false frequency reduction, but also predicts the risk of temperature rise in advance. Furthermore, it automatically iterates and calibrates the threshold to ensure the computing power supply for high-safety intelligent driving inference tasks and meet the requirements of driving function safety.

[0073] The present solution will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0074] Reference Figures 1 to 3 As one embodiment of the present invention, this embodiment provides a model inference dynamic frequency reduction temperature control safe operation system adapted to the vehicle's temperature threshold, comprising:

[0075] The data acquisition module 110 is used to acquire temperature, including the temperature of the CPU (GPU, NPU) of the vehicle system (when running), and to process the acquired temperature accordingly.

[0076] One automotive-grade NTC temperature sensor is embedded in the core of the vehicle's CPU, NPU, and GPU. The sensor collects real-time temperature data at a fixed frequency of 20Hz. An outlier value with a temperature rise exceeding 8℃ within one second is eliminated using a sliding window algorithm, and a reliable temperature sequence with an accuracy of ±0.3℃ is output.

[0077] It can also collect multi-dimensional data such as cooling fan speed, thermal grease aging resistance, and air outlet temperature of the cabin air conditioner to establish a temperature prediction model that integrates multi-source data, further improving the temperature sampling accuracy to ±0.15℃.

[0078] Data processing module 120, which includes dynamic threshold calibration unit 1201 and acquisition unit 1202;

[0079] The dynamic threshold calibration unit 1201 responds to the data acquisition module, which is used to perform dynamic threshold calibration processing on the acquired temperature, including:

[0080] Obtain relevant data, including the CPU's factory-calibrated general attributes, such as maximum computing load.

[0081] In summary, by retrieving common attributes such as the CPU's factory-calibrated maximum computing load and combining them with real-time temperature data to map the current computing load, a two-way correlation model of "temperature-computing load" is established.

[0082] This avoids the current technology's reliance on a single dimension of temperature for judgment, thus binding the hardware's factory baseline characteristics and real-time operating status, allowing threshold calibration to perfectly match the hardware's physical characteristics.

[0083] Based on temperature, the temperature change of the vehicle's CPU is obtained, and the computing load of the CPU is obtained according to the change, including obtaining the corresponding computing load based on every 10% increase in CPU temperature.

[0084] Based on the computing load obtained for every 10% increase in temperature, obtain the increase in computing load for every 10% increase in temperature;

[0085] Based on a 10% increase in temperature, obtain the temperature corresponding to an increase in computing load exceeding 10% (for example, if the temperature increases from 40℃ to 44℃, the temperature increases by 10%, but the computing load increases by 12%, therefore, 44℃ is the temperature corresponding to an increase in computing load exceeding 10%), and treat the temperature as temperature I, and treat the temperature 10% less than temperature I as the reference temperature.

[0086] Using a 10% increase in temperature as the observation window, the increase in computing power load within the corresponding interval is statistically analyzed, and the critical temperature at which the increase in computing power exceeds 10% is identified as Temperature I.

[0087] This can accurately pinpoint the risk threshold of "temperature rise accompanied by abnormal surge in computing power", transforming the vague temperature rise trend into a quantifiable judgment indicator.

[0088] The value shifted downwards by 10% from temperature I is taken as the reference temperature to define the initial boundary of the safe observation range.

[0089] Sufficient warning buffer zones can be reserved before the risk threshold to avoid frequent system jitter and state switching near the critical value, thereby improving operational stability.

[0090] The time taken to raise the reference temperature to temperature I is obtained, and the time is divided into 10 time nodes of 2 milliseconds each. The change in the increase of computing load at each time node is obtained (in order to achieve discretized observation in the time series dimension).

[0091] In one feasible implementation, the sampling accuracy can be further improved to 0.5ms, and the identification of slight fluctuations in computing power load can predict the abnormal rise in computing power 150ms in advance, significantly extending the early warning response time margin.

[0092] The time taken is regarded as a reference time, and a temperature threshold is preset based on the reference time (i.e., the temperature is increased by 10%).

[0093] The total time from the reference temperature to temperature I is evenly divided into 10 sampling nodes of 2ms each, and the dynamic changes in computing power increase are collected point by point.

[0094] This enables precise observation of the computing power-temperature rise process at the sub-millisecond level, capturing instantaneous computing power anomalies that are completely undetectable by traditional large time windows, and improving prediction accuracy by tens of times.

[0095] The CPU frequency is reduced based on a temperature threshold. If, at a future time, the CPU temperature rises beyond the temperature threshold within the same time frame as the reference time, the CPU frequency will be reduced.

[0096] In the future, if the CPU temperature rises beyond the temperature threshold within the same time frame as the reference time, but the increase in computing load does not exceed 10%, the CPU will not be downclocked.

[0097] Frequency reduction is only triggered when the temperature rises and the computing power increases by more than 10% simultaneously; frequency reduction is not performed if only the temperature exceeds the limit.

[0098] Eliminate invalid frequency reduction caused by instantaneous environmental fluctuations and sensor noise, and preserve the computing power supply for high-safety intelligent driving tasks to the greatest extent.

[0099] In terms of temperature control, reducing the frequency will reduce the temperature simultaneously.

[0100] In another feasible implementation, a vehicle-specific cooling fan with PWM speed control can be installed. It operates with low noise when the temperature is below 60°C and automatically increases its speed to enhance heat dissipation when the temperature exceeds 80°C.

[0101] The vehicle's cooling system and air conditioning duct can also be connected to automatically guide low-temperature air conditioning air to pre-cool the vehicle's area when needed, quickly removing accumulated heat.

[0102] It can also predict the CPU temperature rise trend in advance based on the computing power increase data of the 10 split 2ms time nodes. When the computing power load increase is about to exceed 10%, the computing power allocation can be slightly adjusted in advance to suppress the temperature rise in the early stage, avoid the temperature from rising and then being passively cooled down, and greatly improve the temperature control response efficiency.

[0103] Based on the temperature threshold, if the CPU temperature rises below the temperature threshold within the same time frame as the reference time at a future time, the temperature threshold will be updated and calibrated based on the actual temperature rise (after updating and calibrating the temperature threshold based on the actual temperature rise, the CPU frequency will be reduced based on the updated and calibrated temperature threshold).

[0104] The acquisition unit 1202 is used to perform relevant acquisition.

[0105] Related acquisitions include:

[0106] Within the reference time, the time corresponding to each 0.5℃ increase in temperature was obtained, and this time was regarded as the control time;

[0107] At a future time, if the time required to increase the temperature by 0.5°C is less than the control time within the same time frame, an early warning will be issued.

[0108] Once the warning is issued, if the CPU frequency is reduced if the time taken is continuously lower than the control time and the temperature rises by more than 3%, the CPU will be reduced.

[0109] Using the time taken for a 0.5℃ temperature rise as a baseline reference value, an early warning will be triggered if the time taken for the temperature rise is subsequently detected to be lower than the reference value.

[0110] This can upgrade temperature control intervention from "passive cooling after the fact" to "proactive prediction before the fact", significantly extending the system's response time margin and avoiding sudden temperature spikes that can lead to loss of control.

[0111] In the dynamic threshold calibration unit, the change in the increase rate of computing load at each time point is obtained, as shown below:

[0112] , =0, 1, 2, ..., 10;

[0113] In the formula, Indicates the first Each time point is acquired;

[0114] Indicates the start time corresponding to the reference temperature;

[0115] Indicates the sampling point number (from 0 to 10, a total of 11 nodes, covering the entire observation period);

[0116] ;

[0117] In the formula, Indicates the relationship with the first CPU computing load at each time point (expressed as a percentage).

[0118] Indicates the first The number of effective computing units that the CPU is running at each time point;

[0119] This indicates the total number of computing power units at full load (i.e., the maximum computing power load) that the CPU is factory-specified.

[0120] It also includes the following:

[0121] ; =0, 1, 2, ..., 10;

[0122] In the formula, Indicates the first The interval (the first interval) ~1 to the The month-on-month increase in computing load (between two consecutive 2ms time points);

[0123] This indicates the computing load at the previous time point.

[0124] Based on the calculation results, the temperature corresponding to the increase in computing load exceeding 10% was obtained, and relevant location was performed at all time points. This location included identifying the key time points corresponding to abnormal jumps in the increase of computing load, as shown below:

[0125] ;

[0126] In the formula, This represents the set of times (set of time points) that correspond to the abnormal jumps in the increase of computing load.

[0127] This represents the average increase in computing load over 10 time points.

[0128] The standard deviation of the increase in computing load over 10 time points (used to determine the threshold for abnormal jumps).

[0129] Based on the calculation results, the increase corresponding to the abnormal jump in the increase of computing load is obtained at all time points (for example, the computing load increases by 0.3% in the first 4 adjacent time points, and then suddenly increases by 1% in the last time point, which is an abnormal jump in the increase). The increase is regarded as the risk increase. If the increase in the CPU computing load is the same as the risk increase in the future, the CPU frequency will be reduced.

[0130] Verification is performed after the CPU is downclocked, including:

[0131] After frequency reduction, a verification cycle is defined as at least 6 consecutive control periods.

[0132] If the temperature increases by 0.5℃ in each comparison time during the verification period, but the computing load does not increase during the verification period, then the CPU frequency reduction will be cancelled.

[0133] After frequency reduction, the system continuously observes for 6 reference periods. Once it confirms that there is no abnormal increase in computing power, it automatically resumes full computing power operation.

[0134] To avoid unnecessary long-term frequency reduction and wasted computing power, computing resources are released as soon as the temperature control target is achieved, balancing safety and performance.

[0135] In one feasible implementation, a confidence verification step for model inference after frequency reduction can be added. If the confidence of the intelligent driving perception model does not decrease significantly after frequency reduction, the moderate frequency reduction state can be maintained to further reduce the overall power consumption without affecting functional safety.

[0136] Based on the set of identified moments corresponding to abnormal jumps in computing load, the time interval (i.e., time window) within the reference time period is obtained. Within the reference time period, it is determined which time nodes the abnormal jump in computing load occurred between, that is, from which 2 milliseconds did the abnormal jump in computing load begin. This time interval is considered the reference time interval. In the future, within any time period that is the same as the reference time period, if the increase in computing load does not experience an abnormal jump within the time interval (time window) that is the same as the reference time interval, then it is determined that the increase in computing load will not be abnormal within this time period.

[0137] The time window characteristic of abnormal fluctuations in memory computing power can be reused in similar scenarios to avoid risks in advance.

[0138] In one feasible implementation, a temperature control feature sharing network can be built at the vehicle edge, enabling synchronous iteration of temperature control data from multiple vehicles of the same model. New models can directly reuse mature temperature control feature libraries upon leaving the factory, significantly shortening the system adaptation cycle.

[0139] Based on the above, this application uses a millisecond-level computing power-temperature linkage detection mechanism to trigger frequency reduction only when both the temperature change exceeds the standard and the computing power increase exceeds 10% at the same time. This not only prevents the waste of computing power due to false frequency reduction, but also predicts the risk of temperature rise in advance, and automatically iterates and calibrates the threshold to ensure the computing power supply for high-safety intelligent driving inference tasks and meet the requirements of driving function safety.

[0140] This embodiment, based on the above-mentioned model reasoning for a dynamic frequency reduction temperature control safety operation system adapted to the vehicle's temperature threshold, also proposes a working method for this system, as follows:

[0141] S10: Acquire temperature, including the temperature of the vehicle's CPU, and perform relevant processing on the acquired temperature;

[0142] S20: Perform dynamic threshold calibration on the acquired temperature, including:

[0143] Obtain relevant data, including the CPU's factory-calibrated general attributes, such as maximum computing load.

[0144] Based on temperature, the temperature change of the vehicle's CPU is obtained, and the computing load of the CPU is obtained according to the change, including obtaining the corresponding computing load based on every 10% increase in CPU temperature.

[0145] Based on the computing load obtained for every 10% increase in temperature, obtain the increase in computing load for every 10% increase in temperature;

[0146] Based on a 10% increase in temperature, the temperature corresponding to an increase in computing load exceeding 10% is obtained and considered as Temperature I. Temperatures less than Temperature I by 10% are considered as reference temperatures.

[0147] The time taken to raise the reference temperature to temperature I is obtained, and the time is divided into 10 time nodes of 2 milliseconds each. The change in the increase of computing load at each time node is obtained.

[0148] The time taken is regarded as a reference time, and a temperature threshold is preset based on the reference time.

[0149] The CPU frequency is reduced based on a temperature threshold. If, at a future time, the CPU temperature rises beyond the temperature threshold within the same time frame as the reference time, the CPU frequency will be reduced.

[0150] In the future, if the CPU temperature rises beyond the temperature threshold within the same time frame as the reference time, but the increase in computing load does not exceed 10%, the CPU will not be downclocked.

[0151] Based on the temperature threshold, if the CPU temperature rises less than the temperature threshold within the same time frame as the reference time at a future time, the temperature threshold will be updated and calibrated based on the actual temperature rise.

[0152] S30: Perform relevant acquisition.

[0153] In summary, this invention relies on the CPU's factory-calibrated maximum computing load as a fundamental attribute. By capturing the correlation between CPU temperature changes and computing load increases, it breaks down the total time for the reference temperature to rise to temperature I into 10 high-precision 2-millisecond time nodes to complete the full-process feature acquisition. This not only constructs a dynamic self-calibrating temperature threshold system adapted to the current operating conditions but also introduces a differentiated frequency reduction judgment logic: "if the temperature rise exceeds the limit but the computing load increase does not exceed 10%, frequency reduction will not be triggered." Combined with a millisecond-level early warning mechanism for the reference time corresponding to every 0.5℃ temperature rise, it significantly avoids... It addresses the issues of sensor sampling noise and invalid frequency reduction caused by instantaneous temperature fluctuations in the environment. It also includes a verification mechanism that checks the frequency reduction for six consecutive comparison times. When it is confirmed that there is no abnormal increase in computing power, it can automatically cancel the frequency reduction and restore full computing power operation. It can also reuse risk memory throughout the entire life cycle by relying on the identified abnormal jump characteristics of computing power increase. It ensures the continuity of computing power supply for high-safety-level AI inference tasks such as intelligent driving throughout the entire process. While achieving controllable CPU temperature in the vehicle, it completely avoids the driving safety hazards caused by a one-size-fits-all forced frequency reduction and meets the requirements of automotive-grade functional safety management.

[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A model-based dynamic frequency reduction temperature control system for safe operation that adapts to vehicle temperature thresholds, characterized in that: include: The data acquisition module is used to acquire temperature, including the temperature of the vehicle's CPU, and to process the acquired temperature. The data processing module includes a dynamic threshold calibration unit and an acquisition unit; The dynamic threshold calibration unit responds to the data acquisition module and performs dynamic threshold calibration processing on the acquired temperature, including: Obtain relevant data, including the general attributes of the CPU as specified at the factory, including the maximum computing load; Based on the temperature, the change in the temperature of the vehicle's CPU is obtained, and the computing load of the CPU is obtained according to the change, including obtaining the corresponding computing load based on every 10% increase in the CPU temperature. Based on the computing power load obtained for every 10% increase in temperature, obtain the increase in computing power load for every 10% increase in temperature; Based on a 10% increase in temperature, the temperature corresponding to an increase in computing load exceeding 10% is obtained, and this temperature is considered as Temperature I. Temperatures less than Temperature I by 10% are considered as reference temperatures. The time taken to raise the reference temperature to temperature I is obtained, and the time is divided into 10 time nodes, each of which is 2 milliseconds. The change in the increase of computing load in each of the time nodes is obtained. The time taken is regarded as a reference time, and a temperature threshold related to temperature is preset based on the reference time; Based on the temperature threshold, the CPU frequency is reduced. If, at a future time, the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, the CPU frequency is reduced. In the future, if the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, but the increase in computing load does not exceed 10%, then the CPU will not be downclocked. Based on the temperature threshold, if at a future time, within the same time frame as the reference time, the CPU experiences a temperature increase that is less than the temperature threshold, then the temperature threshold will be updated and calibrated based on the actual temperature increase. The acquisition unit is used to perform relevant acquisition.

2. The model-based dynamic frequency reduction temperature control safety operation system for adapting to vehicle temperature thresholds as described in claim 1, characterized in that, The relevant acquisition includes: Within the reference time, the time corresponding to each 0.5°C increase in temperature is obtained, and this time is regarded as the control time; At a future time, if, within the same timeframe as the control, the time required to increase the temperature by 0.5°C is less than the control time, an early warning will be issued; Once the warning is issued, if the CPU frequency is reduced if the time taken is continuously lower than the control time and the temperature increase exceeds 3%, the CPU will be reduced.

3. The model-based dynamic frequency reduction temperature control safety operation system for adapting to vehicle temperature thresholds as described in claim 1, characterized in that, In the dynamic threshold calibration unit, the change in the increase rate of computing load at each time node is obtained, as shown below: ， =0，1，2，...，10； In the formula, Indicates the first Each time point is acquired; This indicates the start time corresponding to the reference temperature; Indicates the sampling point number; ； In the formula, Indicates the relationship with the first CPU computing load at each time point; Indicates the first The number of effective computing units that the CPU is running at each time point; This indicates the total number of computing units at full load, as specified by the CPU manufacturer.

4. The model-based dynamic frequency reduction temperature control safety operation system for adapting to vehicle temperature thresholds as described in claim 3, characterized in that, It also includes the following: ； =0，1，2，...，10； In the formula, Indicates the first The month-on-month increase in computing load for each interval; This indicates the computing load at the previous time point.

5. The model-based dynamic frequency reduction temperature control safety operation system for adapting to vehicle temperature thresholds as described in any one of claims 3 to 4, characterized in that, Based on the calculation results, the temperature corresponding to the increase in computing load exceeding 10% is obtained, and relevant positioning is performed at all time points. The relevant positioning includes locating the key time points corresponding to abnormal jumps in the increase of computing load, as shown below: ； In the formula, This represents the set of moments corresponding to abnormal jumps in the increase of computing load; This represents the average increase in computing load over 10 time points. This represents the standard deviation of the increase in computing load over 10 time points.

6. The model-based dynamic frequency reduction temperature control safety operation system for adapting to vehicle temperature thresholds as described in claim 5, characterized in that, Based on the calculation results, the increase corresponding to the abnormal jump in computing load is obtained at all time points. The increase is regarded as the risk increase. If the increase in CPU computing load is the same as the risk increase in the future, the CPU frequency is reduced.

7. The model-based dynamic frequency reduction temperature control safety operation system for adapting to vehicle temperature thresholds as described in claim 2, characterized in that, Verification is performed after the CPU is downclocked, including: After frequency reduction, a verification cycle is defined as at least six consecutive control periods. If the temperature increases by 0.5°C during each comparison time in the verification period, but the computing load does not increase during the verification period, then the CPU frequency reduction will be cancelled.

8. The model-based dynamic frequency reduction temperature control safety operation system for adapting to vehicle temperature thresholds as described in claim 6, characterized in that, Based on the set of identified moments corresponding to abnormal jumps in computing load, the time interval within which this set of moments falls within the reference time interval is obtained. This time interval is regarded as the reference time interval. In the future, within any time interval that is the same as the reference time interval, if the increase in computing load does not experience abnormal jumps within the time interval that is the same as the reference time interval, it is determined that the increase in computing load will not be abnormal within this time interval.

9. A method for applying the model inference dynamic frequency reduction temperature control safety operation system adapted to the vehicle's temperature threshold as described in claim 1, characterized in that, Includes the following steps: Temperature is acquired, including the temperature of the vehicle's CPU, and the acquired temperature is processed accordingly. The acquired temperature is calibrated using a dynamic threshold, including: Obtain relevant data, including the general attributes of the CPU as specified at the factory, including the maximum computing load; Based on the temperature, the change in the temperature of the vehicle's CPU is obtained, and the computing load of the CPU is obtained according to the change, including obtaining the corresponding computing load based on every 10% increase in the CPU temperature. Based on the computing power load obtained for every 10% increase in temperature, obtain the increase in computing power load for every 10% increase in temperature; Based on a 10% increase in temperature, the temperature corresponding to an increase in computing load exceeding 10% is obtained, and this temperature is considered as Temperature I. Temperatures less than Temperature I by 10% are considered as reference temperatures. The time taken to raise the reference temperature to temperature I is obtained, and the time is divided into 10 time nodes, each of which is 2 milliseconds. The change in the increase of computing load in each of the time nodes is obtained. The time taken is regarded as a reference time, and a temperature threshold related to temperature is preset based on the reference time; Based on the temperature threshold, the CPU frequency is reduced. If, at a future time, the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, the CPU frequency is reduced. In the future, if the CPU temperature increases beyond the temperature threshold within the same time frame as the reference time, but the increase in computing load does not exceed 10%, then the CPU will not be downclocked. Based on the temperature threshold, if at a future time, within the same time frame as the reference time, the CPU experiences a temperature increase that is less than the temperature threshold, then the temperature threshold will be updated and calibrated based on the actual temperature increase. To obtain the relevant information.

Images