Microcontroller-based automatic speed regulation system for electric fan

Abstract

This invention relates to the field of intelligent home appliance control and embedded system technology, specifically a microcontroller-based automatic speed control system for electric fans. The system includes a microcontroller, a sensor array, a motor drive module, and an electric fan motor. The sensor array collects multi-dimensional environmental time-series characteristic data such as temperature, humidity, and airflow resistance, and inputs this data into a preset edge computing prediction model to calculate the target speed control command. During model operation, underlying task scheduling data is extracted, quantified to generate a calculated oscillation entropy value, and compared with a preset computing power triggering threshold. When the entropy value is not higher than the threshold, the target speed control command is output; when the entropy value is higher than the threshold, the prediction model is suspended, and a degradation control strategy is invoked to generate a suboptimal speed control command. Finally, the output command adjusts the fan speed via the motor drive module, and continuous feedback from the sensors forms a closed-loop control, thus balancing environmental regulation effectiveness with microcontroller operational stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent home appliance control and embedded system technology, specifically to an automatic speed control system for electric fans based on a microcontroller. Background Technology

[0002] As a heat dissipation control device in edge computing node environments, the automatic speed regulation system of microcontroller-based electric fans is crucial for ensuring the continuous operation of equipment. In scenarios such as enclosed data cabins and biochemical isolation chambers, the timeliness and stability of its speed regulation response directly affect the thermal safety and airflow organization within the controlled space. Therefore, achieving reliable automatic speed regulation is the key to ensuring the continuous operation of the equipment.

[0003] Existing speed control methods have many problems. For example, they often rely on a single temperature threshold or fixed gear mapping for control, making it difficult to comprehensively reflect multi-dimensional environmental changes such as temperature, humidity, and airflow resistance. When there is external electromagnetic interference, sensor data jitter, or airflow obstruction, misjudgment and speed control lag are likely to occur. Especially under the condition of limited microcontroller computing power, if complex prediction, data cleaning, and interruption processing occupy resources at the same time, there is often a lack of monitoring, circuit breaking, and degradation recovery mechanisms for internal task congestion, which cannot effectively avoid fan control link delay, loss of synchronization, or even heat dissipation interruption. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides a microcontroller-based automatic speed control system for electric fans. Specifically, the technical solution of the present invention includes:

[0005] The system includes a microcontroller, as well as a sensor array, a motor drive module, and a fan motor connected to the microcontroller. The microcontroller runs a real-time operating system and is configured to perform the following steps:

[0006] Multidimensional environmental time-series characteristic data are collected through a sensor array. The multidimensional environmental time-series characteristic data includes temperature fluctuation sequence, humidity fluctuation sequence and airflow resistance fluctuation sequence.

[0007] Multidimensional environmental time-series characteristic data is input into a preset edge computing prediction model to calculate the target speed regulation command. During the operation of the edge computing prediction model, the underlying task scheduling data of the microcontroller is extracted, and a calculated oscillation entropy value representing the disorder of the system task scheduling state is generated based on the underlying task scheduling data. The relationship between the calculated oscillation entropy value and the preset computing power circuit breaker threshold is judged. If the calculated oscillation entropy value is less than or equal to the computing power circuit breaker threshold, the target speed regulation command is used as the final output command. If the calculated oscillation entropy value is greater than the computing power circuit breaker threshold, the edge computing prediction model is suspended, and a preset degradation control strategy is called to generate a suboptimal speed regulation command, which is used as the final output command. The final output command is sent to the motor drive module, which adjusts the speed of the electric fan motor. The sensor array continuously collects the adjusted multidimensional environmental time-series characteristic data to form an environmental feedback control closed loop.

[0008] Optionally, the underlying task scheduling data includes memory reallocation frequency, interrupt nesting depth, and model weight update latency. The step of quantizing and generating the oscillation entropy value based on the underlying task scheduling data includes: extracting memory reallocation frequency, interrupt nesting depth, and model weight update latency; inputting memory reallocation frequency, interrupt nesting depth, and model weight update latency into a preset on-chip monitoring model to calculate the congestion index of the current instruction cycle; and using the congestion index of the current instruction cycle as the oscillation entropy value.

[0009] Optionally, the step of generating suboptimal speed control instructions by invoking a preset degradation control strategy includes: sending a thread suspension signal to the real-time operating system running in the microcontroller; freezing the random access memory space occupied by the edge computing prediction model and releasing computing resources based on the thread suspension signal; extracting the absolute temperature value from the multidimensional environmental time-series feature data as the basic physical scalar; inputting the basic physical scalar into a preset static mapping table for matching, and outputting the suboptimal speed control instructions.

[0010] Optionally, the step of acquiring multidimensional environmental time-series characteristic data through a sensor array includes: acquiring an initial environmental signal; performing analog-to-digital conversion on the initial environmental signal to generate multidimensional environmental time-series characteristic data; wherein the multidimensional environmental time-series characteristic data includes temperature fluctuation sequence, humidity fluctuation sequence, and airflow resistance fluctuation sequence.

[0011] Optionally, before inputting the multi-dimensional environmental time-series feature data into a preset edge computing prediction model to calculate the target speed control command, the method further includes: performing noise filtering on the multi-dimensional environmental time-series feature data to generate cleaned feature data; counting the number of clock cycles consumed by the noise filtering operation per unit time; and inputting the multi-dimensional environmental time-series feature data into a preset edge computing prediction model to calculate the target speed control command, specifically including: inputting the cleaned feature data into the edge computing prediction model to calculate the target speed control command.

[0012] Optionally, the underlying task scheduling data also includes the number of execution clock cycles. The step of inputting the memory reallocation frequency, interrupt nesting depth, and model weight update delay into a preset on-chip monitoring model to calculate the congestion index of the current instruction cycle includes: after dimensionless processing of the memory reallocation frequency, interrupt nesting depth, model weight update delay, and execution clock cycles, performing a weighted summation calculation according to dynamically allocated weight coefficients. The weight coefficients are allocated according to preset rules corresponding to each feature parameter, wherein the weight coefficients corresponding to interrupt nesting depth and execution clock cycles are greater than the weight coefficients corresponding to memory reallocation frequency and model weight update delay, generating a comprehensive load feature; and outputting the congestion index of the current instruction cycle through the on-chip monitoring model based on the comprehensive load feature.

[0013] Optionally, before determining the relationship between the calculated oscillation entropy value and the preset computing power circuit breaker threshold, the method further includes: obtaining the maximum tolerable instruction delay time of the microcontroller; calculating the critical computing power load value by combining the global clock frequency of the microcontroller with the maximum tolerable instruction delay time; and using the critical computing power load value as the preset computing power circuit breaker threshold.

[0014] Optionally, if the calculated oscillation entropy value is greater than the computing power circuit breaker threshold, the edge computing prediction model is suspended, and a suboptimal speed regulation instruction is generated by calling a preset degradation control strategy. After the step of using the suboptimal speed regulation instruction as the final output instruction, the method further includes: continuously monitoring the calculated oscillation entropy value of the microcontroller to obtain the current calculated oscillation entropy value; determining the relationship between the current calculated oscillation entropy value and the preset recovery threshold; if the current calculated oscillation entropy value is less than or equal to the recovery threshold, sending a thread wake-up instruction to the microcontroller; resuming the edge computing prediction model based on the thread wake-up instruction; and maintaining the execution of the degradation control strategy if the current calculated oscillation entropy value is greater than the recovery threshold.

[0015] Optionally, the microcontroller-based automatic fan speed control system is deployed in an edge computing node environment; wherein, the edge computing node environment is a sealed data container or a biochemical isolation chamber; the microcontroller is also configured to: identify non-target environmental noise caused by external electromagnetic interference in multi-dimensional environmental time-series characteristic data, the identification conditions for non-target environmental noise include: the change amplitude of environmental channels exceeds a preset change boundary threshold, and the synchronous peaks or valleys of multiple channels do not match the external reference information; after identifying non-target environmental noise, perform corresponding noise suppression operations.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. This system collects multi-dimensional environmental time-series feature data, including temperature fluctuation sequences, humidity fluctuation sequences, and airflow resistance fluctuation sequences, through a sensor array, and inputs it into a preset edge computing prediction model to calculate the target speed regulation command. This mechanism effectively overcomes the problem of incomplete environmental change reflection caused by relying solely on a single temperature threshold or fixed gear mapping for control in existing technologies. It can comprehensively identify dynamic changes in multi-dimensional environments, avoid misjudgments and speed regulation lags that are prone to occur when the system faces complex environmental fluctuations, and significantly improve the timeliness and accuracy of the automatic speed regulation system response.

[0018] 2. This system extracts the underlying task scheduling data of the microcontroller, quantizes it to generate the calculated oscillation entropy value, and compares it with the computing power circuit breaker threshold calculated from the maximum tolerable instruction delay time and the global clock frequency. When computing power congestion occurs, the system sends a thread suspension signal to the real-time operating system, freezes the random access memory space occupied by the edge computing prediction model and releases computing resources, and extracts the absolute temperature value into the static mapping table to generate suboptimal speed regulation instructions. This mechanism effectively solves the internal task congestion problem caused by the concurrency of complex tasks under the condition of limited computing power of the microcontroller. Through timely degradation control strategy, it avoids the delay, loss of synchronization or heat dissipation interruption of the fan control link.

[0019] 3. This system performs noise filtering on multi-dimensional environmental time-series characteristic data for edge computing node environments such as sealed data cabins or biochemical isolation chambers. It also identifies non-target environmental noise caused by external electromagnetic interference and performs corresponding noise suppression operations, effectively avoiding control misjudgments caused by sensor data jitter. At the same time, after invoking the degradation control strategy, the system continuously monitors the current calculated oscillation entropy value of the microcontroller. Once it is less than or equal to the preset recovery threshold, it sends a thread wake-up command to the microcontroller to resume running the edge computing prediction model. This mechanism not only improves the system's noise resistance under complex interference, but also makes up for the lack of degradation recovery mechanism in existing technologies, ensuring the long-term stable operation of the environmental feedback control closed loop. Attached Figure Description

[0020] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0021] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0023] like Figure 1 As shown, a microcontroller-based automatic fan speed control system includes a microcontroller, a sensor array, a motor drive module, and a fan motor connected to the microcontroller. The microcontroller runs a real-time operating system and is configured to perform the following steps:

[0024] Multidimensional environmental time-series characteristic data are collected through a sensor array. The multidimensional environmental time-series characteristic data includes temperature fluctuation sequence, humidity fluctuation sequence and airflow resistance fluctuation sequence.

[0025] Multidimensional environmental time-series characteristic data is input into a preset edge computing prediction model to calculate the target speed regulation command. During the operation of the edge computing prediction model, the underlying task scheduling data of the microcontroller is extracted, and a calculated oscillation entropy value representing the disorder of the system task scheduling state is generated based on the underlying task scheduling data. The relationship between the calculated oscillation entropy value and the preset computing power circuit breaker threshold is judged. If the calculated oscillation entropy value is less than or equal to the computing power circuit breaker threshold, the target speed regulation command is used as the final output command. If the calculated oscillation entropy value is greater than the computing power circuit breaker threshold, the edge computing prediction model is suspended, and a preset degradation control strategy is called to generate a suboptimal speed regulation command, which is used as the final output command. The final output command is sent to the motor drive module, which adjusts the speed of the electric fan motor. The sensor array continuously collects the adjusted multidimensional environmental time-series characteristic data to form an environmental feedback control closed loop.

[0026] This embodiment provides a microcontroller-based automatic fan speed control mechanism. Specifically, this embodiment deploys the system in an edge heat dissipation terminal inside a sealed data cabin. The cabin is equipped with high-density computing devices. The heat release inside the cabin is continuous and the fluctuation range exceeds the preset normal environmental fluctuation threshold. Once local heat dissipation fails, the server intake air temperature will rise at a rate of increase exceeding the preset temperature rise rate threshold. Therefore, fan speed control cannot rely solely on a single temperature threshold, but needs to pay attention to multiple environmental characteristics such as temperature change trends, humidity changes, and whether the airflow channel is blocked.

[0027] Specifically, the sensor array periodically collects environmental data inside the cabin, and the microcontroller generates a continuous time-series data stream. The multidimensional environmental time-series feature data here is not an isolated sample value, but a set of continuously changing information that can reflect the heat exchange state inside the cabin. For example, a continuous rise in temperature often indicates an increase in heat source release or insufficient air supply; a sudden change in humidity may indicate a change in the external air exchange state; an abnormal increase in airflow resistance is commonly seen in dust accumulation on the filter, cable obstruction, or compression of local air ducts. After receiving these time-series features, the edge computing prediction model outputs a target speed adjustment command to correct the fan speed in advance, avoiding the system being able to only passively remedy the situation when the temperature exceeds the preset safety threshold.

[0028] In this embodiment, the edge computing prediction model is specifically a basic time-series prediction model based on a long short-term memory network or a lightweight neural network based on a gated recurrent unit. The initial weights of the network are obtained through offline training using the historical environmental temperature and humidity fluctuation characteristics of the edge nodes and the corresponding optimal fan speed dataset.

[0029] Furthermore, while running the predictive model, the microcontroller continuously reads the underlying task scheduling status to generate a calculation oscillation entropy value. This metric is not used to evaluate whether the environment itself has deteriorated, but rather to characterize whether resource contention has occurred within the microcontroller due to data cleaning, model inference, interrupt handling, and task switching. When this metric remains within a safe range, it indicates that the microcontroller can still stably complete environmental prediction and motor control, and the target speed regulation command is adopted at this time. When this metric exceeds the computing power circuit breaker threshold, it indicates that the system is in a state where the complex control algorithm is still trying to maintain the preset adjustment accuracy, but the underlying hardware resources are close to their carrying limit. At this time, the optimal solution is no longer continuously executed, but the predictive model is actively suspended and a degraded control strategy is adopted, outputting a suboptimal speed regulation command to prioritize ensuring that the fan control link is not interrupted.

[0030] In the actual control link, the final output command is converted into a corresponding drive signal by the motor drive module, which drives the electric fan motor to change its speed. After the fan speed changes, the cabin temperature, humidity and airflow resistance will change accordingly. The sensor array continues to sample these changes and send them back to the microcontroller, thus forming a closed loop. This closed loop is not a simple temperature measurement-speed regulation loop, but a composite closed loop of environmental change-model decision-computing power status monitoring-control output-environmental change again.

[0031] As a backup safety strategy, if some sensors temporarily lose connection during a certain sampling period, the microcontroller can use the corresponding channel data from the most recent valid period to participate in the model input and mark the channel as a degraded confidence state. If multiple critical channels fail simultaneously, or if the underlying task scheduling data cannot be obtained, the prediction model is skipped directly, and the degraded control strategy is entered to prevent the fan control sequence from being occupied due to waiting for invalid data. Furthermore, if the motor drive module feedback is abnormal, such as the drive signal has been sent but the motor speed does not respond, the system can maintain high-priority safe airflow output and report fault codes to avoid local overheating of the cabin.

[0032] For example, during normal operation of the sealed data container, the temperature in the air intake area of ​​the cabinet rises at a rate lower than the preset rate threshold, the humidity change remains within the preset allowable error range, and the airflow resistance remains within the preset resistance reference range. Based on this, the prediction model judges that the temperature rise is expected due to the increase in load and outputs a moderate speed-up command. When strong external electromagnetic interference occurs, the fluctuation frequency of the environmental data reported by the sensor exceeds the preset interference frequency limit. The microcontroller frequently switches tasks to process abnormal data, and the calculated oscillation entropy value shows a significant upward trend. When the system recognizes that the indicator has reached the danger zone, it immediately stops the complex prediction process and executes a simplified suboptimal speed regulation strategy to prioritize the continuous and stable air delivery of the fan, so that the heat dissipation of the container is not interrupted due to excessive algorithm load.

[0033] Under the conditions of limited edge computing capabilities and complex environmental noise, the environmental regulation effect and the microcontroller's operational stability are simultaneously incorporated into the control framework, so as to achieve an automatic speed adjustment effect that can maintain the stability of the heat dissipation link without relying on continuous high computing power.

[0034] In this embodiment, the underlying task scheduling data includes memory reallocation frequency, interrupt nesting depth, and model weight update delay time. The step of quantizing and calculating the oscillation entropy value based on the underlying task scheduling data includes: extracting memory reallocation frequency, interrupt nesting depth, and model weight update delay time; inputting memory reallocation frequency, interrupt nesting depth, and model weight update delay time into a preset on-chip monitoring model to calculate the congestion index of the current instruction cycle; and using the congestion index of the current instruction cycle as the oscillation entropy value.

[0035] The congestion index characterizes the degree of scheduling disorder and time slice fragmentation of the underlying hardware resources of a microcontroller among multiple concurrent tasks. Since its value is positively correlated with the disorder and unpredictability of the system task scheduling state, it is quantitatively defined in a physical sense as the calculation of the oscillation entropy value.

[0036] This embodiment provides an on-chip monitoring mechanism for quantifying the internal computing power risk of a microcontroller. Specifically, based on the previous embodiment, relying solely on whether fan commands are output in time to determine if the system is overloaded often results in time lag and fails to meet the preset real-time control requirements. Since fan control out-of-sync typically occurs after underlying resource conflicts accumulate beyond a preset conflict count value, this embodiment further selects memory reallocation frequency, interrupt nesting depth, and model weight update delay time as internal load indicators to construct a congestion index.

[0037] Specifically, memory reallocation frequency reflects whether the microcontroller continuously requests, releases, or adjusts buffers within a short period of time. If the environmental data is stable, the memory layout is usually relatively stable. If the dispersion of the input data exceeds the preset data distribution threshold, actions such as data cleaning, cache overwriting, and feature splicing will cause more frequent memory adjustments, indicating that the system has begun to consume additional computing resources in order to process and filter out abnormal data. The interrupt nesting depth reflects whether the underlying task is continuously interrupted by external sampling interrupts, timing interrupts, or communication interrupts. The deeper the nesting level, the more likely it is that the main control thread is difficult to execute completely, and the higher the risk of the fan drive timing being compromised. The model weight update delay time reflects the processing lag within the inference task. If the model-related update actions cannot be completed as expected for a long time, it often means that the computing power has been fully occupied by front-end data preprocessing or interrupted services.

[0038] The on-chip monitoring model can be implemented using a lightweight rule model or a lightweight state evaluation model pre-installed in the microcontroller. Its function is not to reproduce the complete environment prediction, but to integrate multiple underlying characteristics into an indicator that can characterize the congestion level of the current instruction cycle. This indicator is directly used to calculate the oscillation entropy value, so that the system can identify the fluctuations in the internal processing link that have exceeded the preset range and are approaching the preset congestion judgment threshold before the fan control actually loses synchronization.

[0039] Furthermore, the model weight update delay time here is not limited to online updates in the sense of neural network training in engineering implementation. It can also be the delay time of parameter block refresh, quantization coefficient reloading, calibration coefficient rewriting, lookup table replacement, or equivalent model state update actions performed by the prediction model during operation. In other words, as long as the action should have been completed in the current or several adjacent control cycles, but was delayed due to processor congestion, it can be included in this delay. This can cover both prediction models with online adaptive capabilities and edge models that use fixed weights but have runtime parameter refresh mechanisms, avoiding the misunderstanding that this indicator must be trained in real time on the microcontroller.

[0040] Furthermore, to facilitate the practical implementation of the on-chip monitoring model, the three types of input symptoms in the current instruction cycle can be converted into a common risk quantity, i.e., the larger the value, the higher the congestion risk. Among them, the memory reallocation frequency can be directly counted according to the number of reallocations that occur in the current cycle, the interrupt nesting depth can be taken as the maximum nesting level observed in the cycle, and the model weight update delay time can be taken as the time difference between the planned trigger time and the actual completion time of a certain update action. If there are multiple update actions in the current cycle, the maximum value or average value is taken for monitoring. The on-chip monitoring model outputs a single congestion index based on this, so that the underlying symptoms from different sources can be compared uniformly under the same criterion.

[0041] As a backup safety strategy, if a certain underlying symptom is not available in the current cycle, such as if the model update action is not triggered in this cycle, the default safety value or the value of the previous valid cycle can be used for monitoring; if all three types of symptoms are abnormally missing at the same time, the system will no longer output a low congestion judgment, but will directly regard the current state as high risk and trigger conservative control; this can avoid misjudging the system as healthy due to missing monitoring data.

[0042] For example, when the data cabin is running under continuous high load, the rate of change of ambient temperature is lower than the preset surge threshold. However, due to periodic electromagnetic disturbances outside the cabin, the temperature and humidity sampling values ​​repeatedly jump. In order to continuously reorganize the cache and respond to sampling interruptions, the microcontroller increases the frequency of memory reallocation, deepens the interrupt nesting level, and the number of times or the total duration of the update action in the prediction thread is postponed exceeds the preset postponement limit. The update action here can be either the adaptive parameter refresh inside the model or the periodic replacement of the coefficient block required for quantization inference. Based on this, the on-chip monitoring model determines that there is obvious congestion in the current cycle. Even if the temperature inside the cabin has not yet reached the dangerous upper limit, it will output a higher calculated oscillation entropy value in advance to prompt the system to take protective actions as early as possible.

[0043] This extends the traditional control that only focuses on ambient temperature to control that also focuses on the internal scheduling state of the chip, thereby enabling early identification of computing power congestion and reducing fan command delays caused by delayed identification.

[0044] In this embodiment, the step of generating a suboptimal speed regulation instruction by invoking a preset degradation control strategy includes: sending a thread suspension signal to the real-time operating system running in the microcontroller; freezing the random access memory space occupied by the edge computing prediction model and releasing computing resources based on the thread suspension signal; extracting the absolute temperature value from the multidimensional environmental time-series feature data as the basic physical scalar; inputting the basic physical scalar into a preset static mapping table for matching, and outputting the suboptimal speed regulation instruction.

[0045] This embodiment provides a degradation control mechanism after computing power circuit breaker is triggered. Specifically, although the above implementation can identify when it is not appropriate to continue running complex prediction models, if the system remains in a state of waiting for recovery after the circuit breaker is triggered, the fan control will experience a control interruption period with no instruction output. Therefore, this embodiment further stipulates that after the circuit breaker is triggered, the real-time operating system immediately suspends the prediction thread, freezes the random access memory space it occupies, and switches to static mapping control based on absolute temperature.

[0046] Specifically, thread suspension essentially cuts off high-computing-power prediction tasks from competing for processor time at the task scheduling level; freezing random access memory space is to prevent model-related buffers from being repeatedly rewritten by other tasks during suspension, avoiding data corruption during recovery, and releasing the computing power available for control threads; the basis for choosing absolute temperature as the basic physical scalar after degradation is that, in the heat dissipation problem of a closed cabin, absolute temperature most directly corresponds to the risk of thermal damage to the equipment, and even if complex trend prediction is lost, the most basic thermal safety can be maintained based solely on absolute temperature;

[0047] The static mapping table can be pre-set according to the device's allowable inlet air temperature, fan performance curve, and cabin layout. For example, different temperature ranges can be corresponding to low speed, medium speed, high speed, or full speed air volume. This mapping table does not rely on complex inference and does not require continuous updating of the model state, so it is very suitable as an emergency strategy during periods of computing power shortage.

[0048] Furthermore, freezing the random access memory space occupied by the edge computing prediction model and releasing computing resources in this embodiment means: performing write-prohibition, pausing refresh, or maintaining the address mapping unchanged on the parameter area, cache area, and intermediate result area related to the prediction model context so that the original context can continue to be used when it is restored later; at the same time, stopping the processor scheduling corresponding to the prediction thread, pausing its direct memory access and transfer requests, and terminating the continuation of feature splicing and inference calculations, thereby releasing processor time slices, bus access opportunities, and interrupt response margins; that is to say, what is frozen is mainly the consistency of the model context, while what is released is mainly the computing execution resources and scheduling resources. The two have different objects of action and there is no logical conflict between them.

[0049] Furthermore, the matching process of the static mapping table can be implemented by direct interval lookup. For example, the absolute temperature is first divided into safe zone, warning zone, risk zone and emergency zone, and then corresponding to low air volume, medium air volume, high air volume and maximum safe air volume, respectively. When the sampled value is near the boundary of two temperature zones, the previous cycle speed can be maintained or the output can be set to a higher speed to avoid frequent speed increases and decreases of the fan caused by boundary jitter. Thus, even without running a complex predictive model, the degradation control can still provide repeatable, verifiable and responsive suboptimal speed control commands.

[0050] As a backup safety strategy, if the temperature sensor happens to be abnormally distorted when the circuit breaker is triggered, the most recent reliable temperature value can be temporarily used and a higher fan speed can be output with a safety margin. If a reliable temperature cannot be obtained for several consecutive cycles, the system will directly enter the highest safe fan speed mode until the critical sensor recovers. If the thread suspension instruction fails to execute successfully, the real-time operating system can again increase the priority of the control thread and limit the time slice of the prediction thread to ensure that the degraded control is executed first.

[0051] For example, if the load of a server group in the data cabin increases beyond a preset surge threshold within a preset time window, and external electromagnetic interference causes drastic fluctuations in sensor data, and the calculated oscillation entropy value exceeds the circuit breaker threshold, the microcontroller immediately sends a suspension command to the real-time operating system; the prediction thread stops occupying the processor, the memory occupied by the model is fixed in a read-only hold or paused write state, and the control thread directly reads the current absolute temperature; for example, if the sampled value shows that the air intake area is close to the device's allowable limit, the static mapping table directly gives a high airflow setting, causing the fan to quickly increase its speed, ensuring continuous heat dissipation of the cabin even if the power consumption corresponding to this instruction is not the lowest.

[0052] When complex predictions can no longer run reliably, a degradation path that is almost independent of computing power but can immediately restore thermal safety control is provided, thereby achieving a bottom-line strategy that prioritizes maintaining basic physical heat dissipation and abandons high-precision predictions.

[0053] In this embodiment, the step of collecting multi-dimensional environmental time-series feature data through a sensor array includes: acquiring an initial environmental signal; performing analog-to-digital conversion on the initial environmental signal to generate multi-dimensional environmental time-series feature data; wherein, the multi-dimensional environmental time-series feature data includes a temperature fluctuation sequence, a humidity fluctuation sequence, and an airflow resistance fluctuation sequence.

[0054] This embodiment provides a mechanism for collecting and forming multidimensional environmental time-series characteristic data. Specifically, using only absolute temperature as the degradation basis can ensure the bottom line safety, but if only temperature is collected during normal operation, it is difficult to identify problems such as air duct blockage, external humid air intrusion, or local vortex formation in advance. Therefore, this embodiment further converts the initial environmental signal into temperature fluctuation sequence, humidity fluctuation sequence, and airflow resistance fluctuation sequence after analog-to-digital conversion.

[0055] Specifically, temperature fluctuation sequences are used to characterize the rate of heat accumulation and diffusion, rather than just instantaneous hot and cold; humidity fluctuation sequences can help identify changes in air exchange in a closed cabin, because the heat exchange characteristics of the radiator surface and the insulation environment of electronic components will be affected after humid air enters; airflow resistance fluctuation sequences can be indirectly constructed by duct differential pressure sensors, wind speed sensors or fan load feedback, and are used to reflect whether the air path is obstructed; for cabin heat dissipation, an increase in temperature does not necessarily indicate an increase in heat source power, but may also be due to an increase in duct resistance causing the cooling airflow to not reach the critical area. Therefore, introducing airflow resistance sequences can help the system distinguish between heat source problems and ventilation problems.

[0056] The role of analog-to-digital conversion is to convert the continuous physical quantity output by the sensor into a discrete digital quantity that can be processed by the microcontroller while maintaining comparability in terms of period. After multiple sampling periods are connected in series, time-series characteristic data that can reflect the changing trend are formed, and the edge prediction model makes speed adjustment judgments based on this environmental state that evolves over time.

[0057] As a backup safety strategy, if a certain environmental signal becomes saturated, drifts, or exceeds its range before analog-to-digital conversion, the microcontroller can mark that data channel as an abnormal channel and reduce its participation weight in the current cycle; if the airflow resistance channel is temporarily unavailable, the system can still use the temperature and humidity sequence to maintain operation, but the conservative airflow will be increased; if only one of the three channels is effective, the safety control function corresponding to the absolute temperature will be retained first.

[0058] For example, after the data cabin has been in operation for a long time, dust accumulates on the filter components. At this time, the temperature rise does not necessarily show a sudden change with a slope exceeding the preset step threshold, but the airflow resistance fluctuation sequence will reach an abnormal state before the temperature. If, at the same time, external humid air enters briefly due to cabin door maintenance, the humidity fluctuation sequence will also change significantly. After the multi-dimensional sequence after analog-to-digital conversion is sent to the microcontroller, the system can identify that this is not simply an increase in server load, but a change in ventilation conditions, and thus arrange the fan speed more effectively.

[0059] This provides multi-source environmental data with practical engineering significance for subsequent prediction and degradation, thereby enabling earlier and more stable heat dissipation regulation than single-variable temperature control.

[0060] In this embodiment, before the step of inputting multi-dimensional environmental time-series feature data into a preset edge computing prediction model to calculate the target speed regulation command, the method further includes: performing a noise filtering operation on the multi-dimensional environmental time-series feature data to generate cleaned feature data; counting the number of clock cycles consumed by the noise filtering operation per unit time; and specifically, the step of inputting multi-dimensional environmental time-series feature data into a preset edge computing prediction model to calculate the target speed regulation command includes: inputting the cleaned feature data into the edge computing prediction model to calculate the target speed regulation command.

[0061] This embodiment provides a preprocessing mechanism for environmental noise. Specifically, in the aforementioned scheme, if the sensor sampled values ​​are directly entered into the edge prediction model, external electromagnetic interference, sampling transient interference pulses, or instantaneous communication jitter will be mistaken by the model as drastic changes in the real environment, resulting in repeated model recalculation and unnecessary computational power consumption. Therefore, this embodiment adds a noise filtering operation before prediction and counts the number of clock cycles consumed by the filtering process per unit time.

[0062] Specifically, the goal of noise filtering is not to smooth all fluctuations, but to distinguish between environmental changes with physical continuity and abnormal fluctuations without physical basis. For example, the temperature in a data cabin cannot fluctuate repeatedly with an amplitude exceeding a preset jump threshold within a very short period without a heat source; humidity should not exhibit an alternating waveform with a slope exceeding a preset waveform threshold instantaneously in a confined space; airflow resistance generally will not reverse at high frequencies if it is not accompanied by changes in the duct state. Therefore, the microcontroller can perform window smoothing, outlier removal, mutation consistency verification, or adjacent channel cross-validation on the time series data to form cleaned feature data, which is then input into the prediction model.

[0063] Furthermore, the number of clock cycles consumed in the execution of statistical noise filtering is of great significance. In extreme interference environments, the system bottleneck is not only due to environmental noise interference, but also due to the computing resources consumed in performing noise filtering. If the filtering overhead continues to increase, it indicates that the quality of the input data is deteriorating and the microcontroller has spent a lot of time on data cleaning. At this point, even if the prediction model itself is not yet congested, the overall control link is already close to the edge of risk.

[0064] As a backup safety strategy, if there is too little effective data within a certain time window after noise filtering, the system can abandon the fine prediction of that window and use the stable characteristics of the most recent effective window to continue control; if the overhead of the filtering module itself increases abnormally and exceeds the preset time budget, the deep cleaning can be terminated in advance, and only the basic filtering can be retained to prevent the preprocessing consumption from exceeding the preset allocated upper limit of clock cycles and blocking the main control link; if the original data is consistently unreliable, the system will directly switch to the degraded control strategy.

[0065] For example, when the data container encounters strong external electromagnetic disturbances, the number of transient interference pulses at the temperature sampling points exceeds the preset density threshold, and the humidity curve also shows obvious jumps. The microcontroller first filters these sequences for noise, removes isolated spikes that lack physical continuity, and then sends the cleaned stable sequences into the prediction model. At the same time, the system finds that the clock cycles occupied by the filtering module exceed the preset time allocation limit, indicating that although the environmental data can still be partially cleaned, its computing resource consumption is increasing. This information will be retained for subsequent computing power risk assessment.

[0066] While ensuring the availability of environmental information, the chip cost of cleaning up dirty data is made explicit, thereby achieving synchronous management of input noise and computing power consumption.

[0067] In this embodiment, the underlying task scheduling data also includes the number of execution clock cycles. The step of inputting the memory reallocation frequency, interrupt nesting depth, and model weight update delay time into a preset on-chip monitoring model to calculate the congestion index of the current instruction cycle includes: performing dimensionless processing on the memory reallocation frequency, interrupt nesting depth, model weight update delay time, and execution clock cycles, and then performing weighted summation calculation according to dynamically allocated weight coefficients. The weight coefficients are allocated according to preset rules corresponding to each feature parameter, wherein the weight coefficients corresponding to interrupt nesting depth and execution clock cycles are greater than the weight coefficients corresponding to memory reallocation frequency and model weight update delay time, generating a comprehensive load feature; and outputting the congestion index of the current instruction cycle through the on-chip monitoring model based on the comprehensive load feature.

[0068] This embodiment provides a comprehensive load feature construction mechanism. Specifically, while observing only memory reallocation frequency, interrupt nesting depth, and model update latency can reflect signs of congestion within the microcontroller, it still has a drawback: it cannot accurately reflect the additional pressure on the system caused by abnormal interference data cleaning. Since the prediction model itself does not deteriorate significantly in some scenarios, the core factor leading to increased system processing latency lies in the front-end filtering action itself. Therefore, this embodiment incorporates the number of clock cycles consumed into the on-chip monitoring model.

[0069] Specifically, these four types of indicators come from different levels: memory reallocation frequency reflects the frequency of storage resource reallocation operations, interrupt nesting depth reflects the depth to which the processor is forcibly suspended by external requests, model weight update latency reflects the delay in the execution cycle of inference tasks, and the number of clock cycles consumed reflects the degree to which preprocessing steps such as data cleaning encroach on real-time performance. Since they have different dimensions and ranges of change, direct comparison is prone to distortion. Therefore, they are first processed to be dimensionless, so that each indicator is converted into load indicators that can be evaluated in parallel. Then, they are combined into comprehensive load characteristics according to preset weights, and the on-chip monitoring model outputs the congestion index of the current instruction cycle.

[0070] The weighting here prioritizes the real-time requirements of fan control. For heat dissipation scenarios like enclosed data cabins, interrupt nesting and preprocessing clock consumption typically have a more direct impact on pulse width modulation control timing, thus allowing for larger dynamic weighting coefficients. The preset weighting coefficients are dynamically allocated based on the severity of each characteristic parameter's encroachment on the underlying motor control timing. Specifically, the weighting coefficients corresponding to interrupt nesting depth and the number of clock cycles consumed during execution are greater than those corresponding to memory reallocation frequency and model weight update latency. If the cabin environment is stable over a long period but model updates are frequent, model latency can be given higher priority. The focus of this mechanism is not on the individual intensity of a single symptom, but on whether multiple indicators collectively point to a near-blockage of the control link.

[0071] As a backup safety strategy, if one symptom is temporarily high while the others are normal, the system does not need to immediately trip; it can first enter an early warning state and continue to observe. If multiple symptoms rise simultaneously, even if some individual values ​​have not yet reached the individual warning level, it can be determined that overall congestion has occurred. If the reference boundary required for dimensionless measurement has not been fully calibrated in the initial stage of equipment operation, the load range recommended by the chip manufacturer can be used as the initial benchmark and gradually corrected in subsequent operation.

[0072] For example, when the data container encounters continuous interference, the system finds that although the temperature and humidity data can still be predicted after cleaning, the clock cycles occupied by the cleaning action have increased significantly, and the nesting of interrupts has also begun to deepen. If only the model update delay is considered, it may not have reached a dangerous state yet. However, after incorporating the preprocessing overhead and other symptoms into the comprehensive load characteristics, the on-chip monitoring model can identify the congestion index rising earlier, thereby limiting high load prediction behavior in advance.

[0073] Furthermore, to avoid ambiguity in statistical interpretation between the number of execution clock cycles consumed per unit time and the congestion index of the current instruction cycle, all four types of symptoms are collected using the same monitoring window in this embodiment. The monitoring window is preferably the current instruction cycle. If a fixed-length sliding time window is used in the system engineering implementation to statistically filter noise overhead, the number of execution clock cycles consumed within the sliding time window is converted into the equivalent occupancy of the corresponding current instruction cycle before being input into the on-chip monitoring model along with the other three types of symptoms. In other words, the memory reallocation frequency, interrupt nesting depth, model weight update delay time, and number of execution clock cycles consumed are all compared under the same time base before participating in dimensionless conversion, thereby ensuring that the output congestion index has a unified periodic meaning.

[0074] Furthermore, dimensionless processing can be achieved by interval normalization or hierarchical mapping based on the safety reference boundaries corresponding to each symptom. The purpose is to maintain the unidirectional relationship that larger values ​​indicate higher congestion risk. Specifically, the memory reallocation frequency can be mapped relative to a preset safe reallocation upper limit, the interrupt nesting depth can be mapped relative to the maximum acceptable nesting level, the model weight update delay time can be mapped relative to the planned time limit of the update action, and the number of clock cycles consumed can be mapped relative to the time budget allocated to the noise filtering module in the current instruction cycle. After dimensionless processing, the four types of symptoms enter the same weighted summation link, avoiding any one item from unreasonably dominating the overall load characteristics simply because its original dimension is large.

[0075] Furthermore, before the comprehensive load characteristics are input into the on-chip monitoring model, a boundary pruning or saturation limit process can be performed first. That is, when a certain symptom exceeds its maximum risk boundary, it is used in the calculation according to the full risk value corresponding to that boundary, instead of being infinitely amplified. The purpose of doing this is to make the on-chip monitoring model focus on reflecting whether the system has entered a high congestion zone, rather than allowing extreme outliers to destroy the stability of the overall criteria.

[0076] The microcontroller is shifted from focusing solely on whether the model can complete inference calculations on time to simultaneously monitoring whether there is a risk of computing power overload in the entire data processing chain, thereby achieving a computing power assessment that is closer to actual operational risks.

[0077] In this embodiment, before determining the relationship between the calculated oscillation entropy value and the preset computing power circuit breaker threshold, the method further includes: obtaining the maximum tolerable instruction delay time of the microcontroller; calculating the critical computing power load value by combining the global clock frequency of the microcontroller with the maximum tolerable instruction delay time; and using the critical computing power load value as the preset computing power circuit breaker threshold.

[0078] This embodiment provides a mechanism for determining the computing power circuit breaker threshold. Specifically, in the aforementioned implementation, if the circuit breaker threshold adopts an empirically fixed value, two types of problems are likely to occur: one is that the threshold is too conservative, causing the prediction model to be suspended frequently and the system to remain in suboptimal control for a long time; the other is that the threshold is too high, causing the circuit breaker action to be triggered too late, and the fan command has already experienced an unacceptable delay. Therefore, this embodiment determines the critical computing power load value based on the microcontroller's maximum tolerable instruction delay time and the global clock frequency, and uses it as the computing power circuit breaker threshold.

[0079] Specifically, the maximum tolerable instruction delay time can be determined by the thermal inertia of the heat dissipation object and the fan control requirements. In a closed data cabin, once the server air intake exceeds the allowable control hysteresis, the local temperature may enter an unsafe zone in a short time, so the delay tolerance is limited. The global clock frequency represents the processing cycle that the microcontroller can control per unit time. When the two are combined, what is actually determined is: the comprehensive load limit that the microcontroller can support under the premise that the fan control commands can still be issued on time. Once the calculated oscillation entropy value approaches this boundary, it means that continuing to insist on complex prediction will crowd out the underlying drive timing.

[0080] The key to this threshold determination method is that it does not start from control precision, but from the bottom line of real-time survival; that is, the system prioritizes ensuring that the control command is issued no later than the thermal safety allowable range of the equipment, and pursues prediction quality within the remaining computing power.

[0081] Furthermore, to avoid the threshold determination process remaining at an abstract level, this embodiment can calculate the critical computing load value as follows: first, based on the global clock frequency ( ) and maximum tolerable instruction delay time ( ), thus obtaining the maximum number of instruction cycles that can be controlled within this delay boundary ( ),satisfy( After that, deduct the basic reserved number of clock cycles required for motor drive, sensor sampling, communication maintenance, and real-time operating system scheduling. This yields the critical computing load value that can be used for predicting, cleaning, and monitoring link occupancy. ),satisfy( ); where global clock frequency The unit is Hertz, and the maximum tolerable instruction delay time is... The unit is seconds; symbol The symbol - indicates multiplication, and the symbol - indicates subtraction; when the calculated oscillation entropy value output by the on-chip monitoring model has been mapped to the equivalent load within the same period, it can be directly converted ( This is compared with the computing power circuit breaker threshold;

[0082] Furthermore, if the oscillation entropy value is calculated using a dimensionless congestion index instead of a direct clock cycle number, a one-to-one calibration relationship can be established during the equipment commissioning phase. Several typical congestion index samples can be mapped to their actual consumption cycle intervals, thereby converting the result to (…). The corresponding circuit breaker threshold; this does not change the technical path of using the critical computing power load value as the preset computing power circuit breaker threshold in the embodiment, and also makes the threshold comparable to the aforementioned congestion index in terms of dimensions, avoiding the problem that one uses an index and the other uses a clock cycle and cannot be directly determined.

[0083] As a backup safety strategy, if the device has not yet obtained a stable maximum tolerable instruction delay time in the early stage, a conservative initial value can be set according to the highest safe air intake requirements of the heat dissipation object; if the microcontroller is running in a state of dynamic frequency reduction or power supply fluctuation, the threshold corresponding to the global clock frequency can be updated in real time to avoid misjudgment caused by using the static threshold; if the accurate clock state cannot be obtained, the system should adopt a more conservative fuse boundary.

[0084] For example, in the data cabin, the microcontroller where the fan controller is located is set to complete the speed adjustment command output before the thermal risk spreads; the system establishes a delay boundary based on this, and then combines it with the current processor clock capability to obtain the critical computing load that the machine can withstand; normally, if the calculated oscillation entropy value is close to but does not exceed the boundary, the system still allows the prediction model to work; however, when the interference increases or the internal task scheduling is significantly congested, it will immediately shut down once the boundary is exceeded, instead of continuing to maintain complex predictions after the real-time boundary is exceeded.

[0085] Furthermore, in this embodiment, the maximum tolerable instruction delay time preferably corresponds to the longest allowed time from the end of the current monitoring window to the final output instruction being issued, and its statistical caliber is consistent with the congestion index of the aforementioned current instruction cycle; that is, the monitoring window used to generate the calculation of the oscillation entropy value, the time boundary used to calculate the critical computing power load value, and the judgment cycle used to actually compare whether the circuit breaker is triggered are preferably defined by the same control cycle or by equivalent cycle definitions that can be converted to each other, so as to avoid the problem that although the threshold has been calculated, the comparison objects are not at the same time scale;

[0086] Furthermore, the base retained beat count ( This is not a fixed constant, but rather a minimum budget for incompressible tasks in the control link. It covers at least sensor sampling, analog-to-digital conversion, motor drive output, real-time operating system scheduling and maintenance, and necessary communication and fault reporting. If equipment operating mode switching causes an increase or decrease in the above basic tasks, the budget will be adjusted accordingly. This ensures that the computing power circuit breaker threshold is always anchored on the principle of first guaranteeing the control bottom line and then allocating the remaining computing power;

[0087] Furthermore, in the actual calibration process, several representative operating states can be selected as samples, such as normal steady state, mild noise disturbance, severe noise disturbance, and frequent interruption state, and the correspondence between their congestion index and actual clock consumption can be recorded respectively. The mapping obtained in this way does not require complex fitting formulas to be used for engineering implementation. It is only necessary to ensure that the higher the congestion index, the greater the equivalent load, and to find the relationship between ( Intersecting circuit breakers; this maintains the consistency of terminology and criteria in the instruction manual, and also gives the comparison between the calculated oscillation entropy value and the computing power circuit breaker threshold a clear engineering meaning;

[0088] This allows the circuit breaker decision to be directly anchored to the actually acceptable control hysteresis boundary, thereby enabling the transferable safety threshold setting between different chip platforms and different heat dissipation objects.

[0089] In this embodiment, after the step of suspending the edge computing prediction model and calling a preset degradation control strategy to generate a suboptimal speed adjustment instruction, and using the suboptimal speed adjustment instruction as the final output instruction, the method further includes: continuously monitoring the calculated oscillation entropy value of the microcontroller to obtain the current calculated oscillation entropy value; determining the relationship between the current calculated oscillation entropy value and a preset recovery threshold; if the current calculated oscillation entropy value is less than or equal to the recovery threshold, sending a thread wake-up instruction to the microcontroller; resuming the edge computing prediction model based on the thread wake-up instruction; and maintaining the execution of the degradation control strategy if the current calculated oscillation entropy value is greater than the recovery threshold.

[0090] This embodiment provides a recovery control mechanism after a circuit breaker is triggered. Specifically, the aforementioned degradation scheme can ensure that the system does not crash during high-risk periods. However, if there is no clear recovery path after a circuit breaker is triggered, the system may remain in static mapping control for a long time, resulting in problems such as increased system power consumption and speed regulation accuracy lower than the preset smoothing standard. Conversely, if the prediction model is restored immediately after the risk has just decreased, it may switch frequently near the boundary, forming new scheduling jitter. Therefore, this embodiment increases the recovery threshold and continuously monitors the current calculated oscillation entropy value after degradation.

[0091] Specifically, the recovery threshold and the circuit breaker threshold can be set to different levels to create hysteresis. The engineering significance of this is that the system only wakes up the prediction thread after the congestion inside the microcontroller has been significantly relieved. This avoids repeated jitter of circuit breaker failure followed by recovery and recovery followed by circuit breaker failure, which is especially suitable for scenarios in data cabins where external interference occurs in intermittent pulses. When the thread is woken up, the basic operating context required by the model can be restored first, and then the target speed control output can be gradually taken over again. During the switching period, the static mapping control and prediction output can also be checked in parallel for a short time. The model can be completely exited from the degraded mode only after it is confirmed that it has returned to normal.

[0092] As a backup safety strategy, if the calculated oscillation entropy value drops briefly but there are still many abnormal sensor inputs, the system can delay waking up the prediction thread to avoid sending unstable data into the complex model again; if the circuit breaker is triggered again in a short time after recovery, the system can extend the next observation window to reduce frequent switching; if the thread wake-up fails, the system will maintain degraded control and continue monitoring until the recovery conditions are met and the system scheduler allows reconnection.

[0093] For example, after the data cabin experiences a round of strong electromagnetic interference, the microcontroller has entered a degraded speed regulation based on absolute temperature. As the interference weakens, the noise filtering burden decreases, the interrupt nesting gradually becomes shallower, and the current calculated oscillation entropy value is continuously below the recovery threshold. Based on this, the system sends a thread wake-up command to the real-time operating system to resume the operation of the edge prediction model. If it is found that the input data has deteriorated again after recovery, the complex prediction will not be forcibly maintained, but the degraded operation will be performed again to maintain the stability and continuity of the cabin's heat dissipation control.

[0094] While ensuring the system's survival baseline, the control strategy is endowed with self-recovery capabilities, thereby achieving closed-loop management from complex prediction and degraded operation to recovery prediction.

[0095] In this embodiment, the microcontroller-based automatic fan speed control system is deployed in an edge computing node environment; wherein, the edge computing node environment is a sealed data container or a biochemical isolation chamber; the microcontroller is also configured to: identify non-target environmental noise caused by external electromagnetic interference in multi-dimensional environmental time-series feature data, the identification conditions for non-target environmental noise include: the change amplitude of environmental channels exceeds a preset change boundary threshold, and the synchronous peaks or valleys of multiple channels do not match the external reference information; after identifying non-target environmental noise, perform the corresponding noise suppression operation.

[0096] This embodiment provides a noise identification and suppression mechanism for special operating conditions of edge computing nodes. Specifically, the aforementioned implementation methods can already perform circuit breaking and degradation when computing power is strained. However, in scenarios such as enclosed data cabins or biochemical isolation chambers, external electromagnetic interference often does not directly damage the fan body, but first contaminates the sensor input, inducing the microcontroller to perform a large number of meaningless calculations. Therefore, this embodiment further identifies and suppresses non-target environmental noise caused by external electromagnetic interference.

[0097] Specifically, non-target environmental noise refers to pseudo-signals that do not correspond to actual heat exchange processes, changes in air humidity, or changes in duct resistance. For example, under electromagnetic shock, the temperature channel may exhibit a step change that does not conform to thermal inertia and exceeds a preset physical correlation threshold; the humidity channel may exhibit instantaneous anomalies that cannot be explained by the actual air exchange in the sealed chamber; and multiple channels may also exhibit synchronous spikes simultaneously. However, these changes have no reasonable causal relationship with changes in fan speed or equipment power. The microcontroller can identify such noise through time continuity verification, inter-channel consistency verification, and correlation verification with fan execution feedback, and perform corresponding suppression operations, such as shielding isolated spikes, extending the confirmation window, reducing the participation of abnormal channels, or enabling anti-interference sampling mode.

[0098] This mechanism is particularly important in the context of biochemical isolation chambers. This is because not only is a suitable temperature required in the isolation chamber, but also stable airflow organization is required to avoid local stagnation and pollution diffusion. If electromagnetic noise causes the control system to misjudge sudden environmental changes and irrationally increase or decrease the wind speed, it may disrupt the originally designed airflow directionality. Therefore, noise suppression is not only a necessary measure to improve data quality, but also to maintain the stability of the flow field in the controlled space.

[0099] As a backup safety strategy, if the system cannot clearly distinguish between a sudden change in the real environment and electromagnetic noise, a strategy of prioritizing safety and then seeking refinement can be adopted. That is, first output a conservative airflow of medium to high level and enter an enhanced monitoring mode. If the critical channel remains unavailable after noise suppression, the participation of the prediction model can be limited, and the system can instead rely on absolute temperature and safe airflow strategies. If the electromagnetic interference lasts for too long, the system can record the interference event and isolate the data of this stage from the model adaptive update link to avoid contaminating the basis of subsequent decisions.

[0100] For example, during the power supply switchover of the sealed data container, the operation of nearby high-power equipment caused significant electromagnetic disturbances. The temperature and humidity sequences reported by the sensor array simultaneously showed high-frequency abnormal pulses, but the cabinet power, fan feedback, and container door status did not change accordingly. Based on this, the microcontroller identified that the data was mainly due to external electromagnetic noise rather than actual environmental deterioration. Therefore, it suppressed the anomalies and reduced the proportion of them entering the prediction model. If the interference continued to increase, the system combined the calculation of the oscillation entropy value to determine whether to shut down the prediction thread and switch to degraded control.

[0101] Furthermore, to maintain consistency in terminology throughout the text, the noise suppression operation in this embodiment can be understood as a targeted or parameterized implementation of the noise filtering operation in the embodiment under external electromagnetic interference scenarios; that is, the two are not two independent conflicting mechanisms, but rather have different levels of action: the former emphasizes identifying and taking targeted measures against non-target environmental noise caused by external electromagnetic interference, while the latter emphasizes preprocessing and cleaning multi-dimensional environmental time-series feature data before inputting the prediction model; in engineering implementation, the microcontroller can first complete the electromagnetic interference identification, and then call the corresponding noise filtering rules, confirmation windows, or channel weight adjustment strategies to form a unified data preprocessing link;

[0102] Furthermore, when identifying non-target environmental noise caused by external electromagnetic interference, the following characteristics can be checked first: First, the change amplitude of the environmental channel exceeds the reasonable change boundary that the corresponding physical object can achieve within the current sampling period; Second, multiple channels show highly synchronized peaks or valleys, but lack causal support in terms of heat exchange, humidity exchange, or air duct changes; Third, abnormal changes in sensors are inconsistent with external reference information such as fan execution feedback, equipment power changes, and door opening and closing status. When the above characteristics meet the preset combination conditions, the system will mark the current abnormal data as non-target environmental noise and reduce its priority in entering the prediction model and adaptive update link.

[0103] Furthermore, to avoid introducing new control jitter through noise identification itself, the noise suppression operation preferably does not directly change the current output speed of the fan, but instead prioritizes its effect on the input data side. This includes shielding isolated spikes, extending the confirmation window for short-term abnormal periods, reducing the weight of contaminated channels, or temporarily freezing their participation. Only when noise contamination has caused a significant decrease in the overall credibility of the input data will the system further activate the aforementioned computing power circuit breaker and degradation control mechanism. As a result, data governance, computing power protection, and fan speed regulation output can maintain consistent processing logic under external electromagnetic interference scenarios.

[0104] This enables the system to distinguish between two different scenarios: degradation of real-world environmental characteristics and data distortion caused by electromagnetic interference, thereby achieving stable heat dissipation control for edge nodes in complex electromagnetic environments.

[0105] 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.

Claims

1. A microcontroller-based automatic speed control system for electric fans, characterized in that, The system includes a microcontroller, as well as a sensor array, a motor drive module, and a fan motor connected to the microcontroller. The microcontroller runs a real-time operating system and is configured to perform the following steps: Multidimensional environmental time-series characteristic data are collected through a sensor array. The multidimensional environmental time-series characteristic data includes temperature fluctuation sequence, humidity fluctuation sequence and airflow resistance fluctuation sequence. Multidimensional environmental time-series feature data are input into a preset edge computing prediction model to calculate the target speed regulation command; During the operation of the edge computing prediction model, the underlying task scheduling data of the microcontroller is extracted, and a computational oscillation entropy value representing the disorder of the system task scheduling state is generated based on the underlying task scheduling data. The relationship between the computational oscillation entropy value and the preset computing power circuit breaker threshold is determined. If the computational oscillation entropy value is less than or equal to the computing power circuit breaker threshold, the target speed adjustment command is used as the final output command. If the computational oscillation entropy value is greater than the computing power circuit breaker threshold, the edge computing prediction model is suspended, and a preset degradation control strategy is called to generate a suboptimal speed adjustment command, which is used as the final output command. The final output command is sent to the motor drive module, which adjusts the speed of the electric fan motor. The sensor array continuously collects multi-dimensional environmental time-series characteristic data after adjustment to form an environmental feedback control closed loop.

2. The microcontroller-based automatic speed control system for electric fans according to claim 1, characterized in that, The underlying task scheduling data includes memory reallocation frequency, interrupt nesting depth, and model weight update latency. The steps for generating and calculating the oscillation entropy value based on the underlying task scheduling data quantization include: extracting memory reallocation frequency, interrupt nesting depth and model weight update delay time. The memory reallocation frequency, interrupt nesting depth, and model weight update delay are input into the preset on-chip monitoring model to calculate the congestion index of the current instruction cycle; the congestion index of the current instruction cycle is used as the oscillation entropy value for calculation.

3. The microcontroller-based automatic speed control system for electric fans according to claim 1, characterized in that, The steps for generating suboptimal speed control instructions by invoking a preset degradation control strategy include: sending a thread suspension signal to the real-time operating system running in the microcontroller; freezing the random access memory space occupied by the edge computing prediction model and releasing computing resources based on the thread suspension signal; extracting the absolute temperature value from the multidimensional environmental time-series feature data as the basic physical scalar; inputting the basic physical scalar into a preset static mapping table for matching, and outputting the suboptimal speed control instructions.

4. The microcontroller-based automatic speed control system for electric fans according to claim 3, characterized in that, The steps of acquiring multidimensional environmental time-series characteristic data through a sensor array include: acquiring initial environmental signals; performing analog-to-digital conversion on the initial environmental signals to generate multidimensional environmental time-series characteristic data; wherein, the multidimensional environmental time-series characteristic data includes temperature fluctuation sequences, humidity fluctuation sequences, and airflow resistance fluctuation sequences.

5. The microcontroller-based automatic speed control system for electric fans according to claim 1, characterized in that, Before the step of inputting multi-dimensional environmental time-series feature data into a preset edge computing prediction model to calculate the target speed regulation command, the method further includes: performing noise filtering on the multi-dimensional environmental time-series feature data to generate cleaned feature data; counting the number of clock cycles consumed by the noise filtering operation per unit time; and inputting multi-dimensional environmental time-series feature data into a preset edge computing prediction model to calculate the target speed regulation command, specifically including: inputting the cleaned feature data into the edge computing prediction model to calculate the target speed regulation command.

6. The microcontroller-based automatic speed control system for electric fans according to claim 5, characterized in that, In conjunction with the features of claim 2, the underlying task scheduling data also includes the number of execution clock cycles; the step of inputting the memory reallocation frequency, interrupt nesting depth, and model weight update delay time into a preset on-chip monitoring model to calculate the congestion index of the current instruction cycle includes: after performing dimensionless processing on the memory reallocation frequency, interrupt nesting depth, model weight update delay time, and execution clock cycles, performing weighted summation calculation according to dynamically allocated weight coefficients, the weight coefficients are allocated according to preset rules corresponding to each feature parameter, wherein the weight coefficients corresponding to interrupt nesting depth and execution clock cycles are greater than the weight coefficients corresponding to memory reallocation frequency and model weight update delay time, generating a comprehensive load feature; based on the comprehensive load feature, outputting the congestion index of the current instruction cycle through the on-chip monitoring model.

7. The microcontroller-based automatic speed control system for electric fans according to claim 1, characterized in that, Before determining the relationship between the calculated oscillation entropy value and the preset computing power circuit breaker threshold, the method further includes: obtaining the maximum tolerable instruction delay time of the microcontroller; calculating the critical computing power load value by combining the global clock frequency of the microcontroller with the maximum tolerable instruction delay time; and using the critical computing power load value as the preset computing power circuit breaker threshold.

8. The microcontroller-based automatic speed control system for electric fans according to claim 1, characterized in that, If the calculated oscillation entropy value is greater than the computing power circuit breaker threshold, the edge computing prediction model is suspended, and a suboptimal speed adjustment instruction is generated by calling a preset degradation control strategy. After the suboptimal speed adjustment instruction is used as the final output instruction, the process also includes: continuously monitoring the calculated oscillation entropy value of the microcontroller to obtain the current calculated oscillation entropy value; determining the relationship between the current calculated oscillation entropy value and the preset recovery threshold; if the current calculated oscillation entropy value is less than or equal to the recovery threshold, sending a thread wake-up instruction to the microcontroller; resuming the edge computing prediction model based on the thread wake-up instruction; and maintaining the execution of the degradation control strategy if the current calculated oscillation entropy value is greater than the recovery threshold.

9. The microcontroller-based automatic speed control system for electric fans according to claim 1, characterized in that, The microcontroller-based automatic fan speed control system is deployed in an edge computing node environment, which is a sealed data container or a biochemical isolation chamber. The microcontroller is also configured to identify non-target environmental noise caused by external electromagnetic interference in multi-dimensional environmental time-series characteristic data. The identification conditions for non-target environmental noise include: the change amplitude of environmental channels exceeds the preset change boundary threshold, and the synchronous peaks or valleys of multiple channels do not match the external reference information. After identifying non-target environmental noise, the system performs corresponding noise suppression operations.

Images