Self-checking method and device for filter screen state of air purifier

Abstract

This specification discloses a self-testing method and apparatus for the status of an air purifier filter. The scheme includes: real-time acquisition of air velocity data at the inlet and outlet of the air purifier filter and pressure difference data across the filter; calculation of real-time filter resistance value based on the air velocity data and the pressure difference data; inputting a data sequence containing relevant data within a historical time window into a trained multi-task deep learning prediction model to simultaneously output a predicted filter resistance value for future times and a predicted index of the filter's remaining effective lifespan; performing fusion decision analysis based on the predicted filter resistance value, the predicted index of the filter's remaining effective lifespan, and a constructed multi-dimensional filter health status vector to determine whether a maintenance prompt should be triggered; and dynamically adjusting the air purifier's fan power using an adaptive algorithm including nonlinear compensation based on the predicted filter resistance value, the predicted index of the filter's remaining effective lifespan, and real-time environmental parameters.

Description

Technical Field

[0001] This application relates to the field of smart home technology, and in particular to a self-testing method and device for the status of an air purifier filter. Background Technology

[0002] In the field of smart home and air purification equipment, manufacturers or service providers sometimes equip air purifiers with filter status monitoring and maintenance reminder functions to enhance product intelligence, increase user stickiness, or build a post-service ecosystem. Currently, common technical solutions mainly rely on two methods: one is to provide countdown reminders based on a preset fixed usage time for the filter; the other is to indirectly infer filter clogging by monitoring airflow attenuation using a simple airflow sensor. However, these methods have some limitations in practical applications. These solutions often fail to accurately reflect the true degree of filter clogging and performance degradation, especially under the influence of different usage environments, different air pollution loads, and differences in actual user habits. This can easily lead to inaccurate reminders, premature replacement prompts resulting in filter waste, or delayed warnings causing a decrease in purification efficiency or even affecting the lifespan of the fan. Summary of the Invention

[0003] This specification provides a self-testing method and apparatus for the status of an air purifier filter to solve at least one of the technical problems mentioned above.

[0004] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows: According to a first aspect of the embodiments of this specification, a self-test method for the condition of an air purifier filter is provided, comprising: Real-time data collection of airflow velocity at the inlet and outlet of the air purifier filter, as well as pressure difference data across the filter; Based on the air velocity data and the pressure difference data, the real-time filter resistance value is calculated; The data sequence containing relevant data within the historical time window is input into a trained multi-task deep learning prediction model to simultaneously output the predicted values ​​of filter resistance and the remaining effective lifespan of the filter at future moments. Based on the predicted value of filter resistance, the predicted index of the remaining effective life of the filter, and the constructed multi-dimensional filter health status vector, a fusion decision analysis is performed to determine whether to trigger a maintenance prompt. Based on the predicted filter resistance, the predicted remaining effective lifespan of the filter, and real-time environmental parameters, the fan power of the air purifier is dynamically adjusted using an adaptive algorithm that includes nonlinear compensation.

[0005] In some optional implementations, the step of calculating the real-time filter resistance value based on the airflow velocity data and the pressure difference data uses the following formula: Among them, symbols Indicates air density, symbol This represents the pressure difference across the filter at the current sampling time, with the symbol... Indicates the air velocity at the current sampling time, symbol This indicates the real-time filter resistance value.

[0006] In some alternative implementations, the multi-task deep learning prediction model employs a shared encoder network and at least two independent decoder networks. The shared encoder is used to extract shared temporal features from the input data sequence; the first decoder is used to output the predicted filter resistance value. The second decoder is used to output the predicted value of the remaining effective lifespan of the filter. ; The total loss function of the multi-task deep learning prediction model The result is obtained by weighted summation of the losses from each task: in, This is the mean square error loss for the prediction of filter resistance. Loss predicted for remaining lifetime. and These are learnable task weight parameters.

[0007] In some optional implementations, the step of dynamically adjusting the fan power of the air purifier based on the predicted filter resistance value, the predicted index of the remaining effective lifespan of the filter, and real-time environmental parameters using an adaptive algorithm including nonlinear compensation includes: Based on the predicted filter resistance value With the preset resistance threshold The ratio is used to calculate the basic adjustment amount of the fan power; Based on the current ambient humidity Cumulative running time of the filter Calculate the nonlinear compensation amount for the wind turbine power; The adjusted wind turbine power setting value is determined based on the sum of the basic adjustment amount and the nonlinear compensation amount.

[0008] In some optional implementations, the step of performing a fusion decision analysis based on the predicted filter resistance value, the predicted index of the remaining effective lifespan of the filter, and the constructed multidimensional filter health status vector to determine whether to trigger a maintenance prompt includes: If the predicted value of the filter resistance Meet the conditions If so, a maintenance prompt will be triggered; and / or, If the conditions are met This will trigger a maintenance prompt, where the symbol... This represents the preset threshold for continuous increase in resistance, symbol... Indicates and The predicted value of the filter resistance at the adjacent previous prediction time; And / or, if the predicted value of the remaining effective lifespan of the filter Meet the conditions If this occurs, a maintenance prompt will be triggered; where the symbol... This indicates the critical value of the remaining effective lifespan of the filter.

[0009] In some alternative implementations, the statement based on the current ambient humidity... Cumulative running time of the filter The steps for calculating the nonlinear compensation amount of the wind turbine power are as follows: Among them, symbols Represents the nonlinear compensation quantity, symbol Indicates reference humidity, symbol Represents the time reference constant, symbol Indicates the calibrated compensation coefficient, symbol Represents the hyperbolic tangent function, symbol This represents an exponential function.

[0010] In some optional implementations, the adjusted fan power setpoint Calculated using the following formula: Among them, symbols Indicates the reference fan power setting value, symbol This represents the power regulation coefficient.

[0011] In some optional implementations, the method further includes the steps of performing closed-loop performance monitoring and online adaptive fine-tuning on the multi-task deep learning prediction model; The closed-loop performance monitoring includes: calculating the model performance drift index based on the historical sequence of the real-time filter resistance value and the corresponding predicted filter resistance value. The calculation formula is as follows: Among them, symbols Indicates the length of the evaluation window; When the performance drift index When the threshold is continuously exceeded, the online adaptive fine-tuning is triggered.

[0012] In some optional implementations, the constructed multidimensional filter health status vector includes at least the real-time filter resistance value, its historical rate of change, cumulative operating time, and a predictive index of the remaining effective lifespan of the filter.

[0013] According to a second aspect of the embodiments of this specification, a self-testing device for the status of an air purifier filter is provided, comprising: The sensor data acquisition module is used to collect real-time airflow velocity data at the inlet and outlet of the air purifier filter and pressure difference data on both sides of the filter. The filter resistance calculation module is used to calculate the real-time filter resistance value based on the air velocity data and the pressure difference data. The multi-task deep learning prediction module is used to input a data sequence containing relevant data within a historical time window into a trained multi-task deep learning prediction model, so as to simultaneously output the predicted values ​​of filter resistance and the remaining effective lifespan of the filter at future moments. The fusion decision and maintenance prompt module is used to perform fusion decision analysis based on the predicted value of filter resistance, the predicted index of the remaining effective life of the filter, and the constructed multi-dimensional filter health status vector, in order to determine whether to trigger a maintenance prompt. The fan power control module is used to dynamically adjust the fan power of the air purifier based on the predicted value of the filter resistance, the predicted index of the remaining effective life of the filter, and real-time environmental parameters, using an adaptive algorithm that includes nonlinear compensation.

[0014] One embodiment of this specification can achieve at least the following beneficial effects: In this technical solution, real-time airflow velocity and pressure difference data are collected and real-time resistance values ​​are calculated. Then, a multi-task deep learning model is used to simultaneously predict resistance trends and remaining lifespan. This allows for early maintenance alerts based on the fusion analysis of predicted values ​​and multi-dimensional health status vectors before the filter becomes actually clogged, thus preventing a decrease in purification efficiency. Furthermore, based on the predicted filter resistance, remaining lifespan indicators, and real-time environmental parameters, this technical solution dynamically adjusts the fan output using an adaptive algorithm that includes nonlinear compensation. This automatically maintains optimal purification efficiency and energy consumption balance when filter conditions change or environmental conditions fluctuate. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a self-test method for the status of an air purifier filter provided in the embodiments of this specification; Figure 2 For corresponding Figure 1 A schematic diagram of the structure of a self-testing device for the filter status of an air purifier. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of one or more embodiments of this specification.

[0018] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another.

[0019] Figure 1 This is a flowchart of a self-test method for the status of an air purifier filter provided in the embodiments of this specification; Figure 2 For corresponding Figure 1 A schematic diagram of the structure of a self-testing device for the filter status of an air purifier.

[0020] This application provides a self-testing method for the filter status of an air purifier, applied to a cloud server, such as... Figure 1 As shown, the method may include: Step 102: Collect real-time airflow velocity data at the inlet and outlet of the air purifier filter and pressure difference data on both sides of the filter.

[0021] In the embodiments described in this specification, specifically, air velocity data can be acquired in real time through sensing units deployed at the inlet and outlet of the air purifier's filter structure. Simultaneously, differential pressure sensing units installed across the filter can acquire real-time differential pressure data across the filter. The air velocity data characterizes the airflow through the filter, while the differential pressure data directly reflects the resistance encountered by the airflow through the filter. In practical applications, the acquisition of the above data can be achieved by an embedded system controlling the corresponding wind speed and differential pressure sensors to periodically or continuously measure and read the data.

[0022] Step 104: Calculate the real-time filter resistance value based on the air velocity data and the pressure difference data.

[0023] Step 106: Input the data sequence containing relevant data within the historical time window into the trained multi-task deep learning prediction model to simultaneously output the predicted values ​​of filter resistance and the remaining effective lifespan of the filter at future times.

[0024] In this embodiment of the specification, this step involves inputting time-series data—comprising multiple data points related to the filter's operating status collected within a historical time window—into a pre-trained multi-task deep learning prediction model. This model performs internal calculations and inferences, outputting two future prediction indicators simultaneously and in parallel: one is the predicted filter resistance value at a future point in time, and the other is the predicted remaining effective lifespan of the filter. In practical applications, from an execution perspective, this computational task can be undertaken by a processor unit with sufficient computing power locally within the air purifier or via a cloud computing platform connected to a network. This computing unit loads the pre-trained multi-task deep learning model, periodically receives historical time-series data packets from the data acquisition module, and uses these packets as input to the model for forward propagation calculations, thereby synchronously generating and outputting the two prediction results mentioned above.

[0025] Step 108: Based on the predicted value of filter resistance, the predicted index of the remaining effective life of the filter, and the constructed multi-dimensional filter health status vector, perform fusion decision analysis to determine whether to trigger a maintenance prompt.

[0026] In the embodiments of this specification, based on the predicted value of filter resistance and the predicted index of remaining effective life of the filter output by the prediction model, and combined with a pre-built health status vector that includes multi-dimensional parameters such as real-time resistance value, its historical rate of change, and cumulative running time, a comprehensive evaluation is performed through specific fusion analysis rules, thereby automatically making a decision on whether to issue a maintenance prompt to the user.

[0027] It should be noted that the technical solution of this application integrates multiple parameters from different dimensions reflecting diverse aspects of the filter's condition by setting a multi-dimensional filter health status vector, constructing a comprehensive and structured state representation. This provides a unified and information-rich input basis for subsequent fusion decision analysis. This vector can fuse multiple heterogeneous but related features, such as real-time measured filter resistance values, historical rate of change reflecting short-term dynamic changes in resistance, cumulative operating time reflecting the overall workload of the filter, and remaining effective life indicators output by the prediction model. This overcomes the potential bias and randomness that may exist when relying on a single indicator (such as instantaneous resistance or fixed operating time) for state assessment, enabling the judgment of filter health to be based on more comprehensive and coherent spatiotemporal information, thus improving the robustness and accuracy of maintenance decisions.

[0028] Step 110: Based on the predicted value of filter resistance, the predicted index of the remaining effective life of the filter, and real-time environmental parameters, dynamically adjust the fan power of the air purifier using an adaptive algorithm that includes nonlinear compensation.

[0029] In the embodiments described in this specification, the purpose of this step is to calculate and adjust the current optimal fan power setting based on the predicted filter resistance and remaining effective lifespan of the filter generated by the multi-task deep learning prediction model mentioned in the preceding steps, combined with environmental parameters collected in real time by environmental sensors. In practical applications, from an execution perspective, the algorithm can be implemented by a microprocessor or control unit integrated within or associated with the air purifier. This control unit periodically acquires the latest prediction indicators and sensor readings, runs the aforementioned adaptive algorithm, and then generates corresponding control signals to adjust the operating state of the fan drive circuit, thereby changing the actual speed and power of the fan.

[0030] Based on the technical solutions described above, this specification also provides some specific implementation schemes, which are described below.

[0031] In an optional embodiment, the step of calculating the real-time filter resistance value based on the air velocity data and the pressure difference data uses the following formula: Among them, symbols Indicates air density, symbol This represents the pressure difference across the filter at the current sampling time, with the symbol... Indicates the air velocity at the current sampling time, symbol This indicates the real-time filter resistance value.

[0032] In an optional embodiment, the multi-task deep learning prediction model employs a shared encoder network and at least two independent decoder networks. The shared encoder is used to extract shared temporal features from the input data sequence; the first decoder is used to output the predicted filter resistance value. The second decoder is used to output the predicted value of the remaining effective lifespan of the filter. ; The total loss function of the multi-task deep learning prediction model The result is obtained by weighted summation of the losses from each task: in, This is the mean square error loss for the prediction of filter resistance. Loss predicted for remaining lifetime. and These are learnable task weight parameters.

[0033] In this embodiment, the multi-task deep learning prediction model can be a neural network architecture that includes shared feature extraction and independent task inference. This architecture specifically consists of a shared encoder network and at least two independent decoder networks. The shared encoder network receives a data sequence composed of relevant parameters (such as air velocity and pressure difference) within a historical time window as input. Its function is to abstract and process the input sequence layer by layer through multiple neural network units, thereby automatically extracting the shared feature representations contained in the time-series data. These shared features capture the overall dynamic pattern and inherent correlation of the data's evolution over time, providing a general information foundation for subsequent different prediction tasks.

[0034] Subsequently, these extracted shared features are fed in parallel to two structurally independent decoder networks. The first decoder network, designed for filter resistance prediction, receives the shared features as input and learns the mapping from the shared features to future filter resistance values ​​through a series of network layers, ultimately outputting the predicted filter resistance value. The second decoder network is designed for predicting the remaining effective lifespan of the filter. Based on the same shared features, it learns and outputs a predictive index reflecting the duration of the filter's sustainable effective operation from the current moment, i.e., the predicted value of the filter's remaining effective lifespan. This design in the technical solution of this embodiment allows two related tasks to both utilize common timing patterns through a shared encoder and focus on their respective specific output targets through independent decoders.

[0035] Simultaneously, when training this multi-task deep learning prediction model, it is also necessary to define and optimize a comprehensive objective function to guide the learning of both tasks at the same time. This objective function is the total loss function. In the technical solution of this application, it is constructed as a weighted sum of the loss functions specific to each task, and the specific expression is as follows: In this formula, the symbol The loss in the task of predicting filter resistance can be calculated using the mean squared error loss function. Its purpose is to measure the resistance value sequence predicted by the model. The squared mean of the differences between the predicted and actual resistance values ​​is used to drive the predicted values ​​to approximate the actual values ​​infinitely on a numerical scale. (Symbol) This represents the loss in the task of predicting the remaining effective life of the filter. This loss function can also take the form of mean squared error or mean absolute error, and its purpose is to optimize the predicted remaining life value. To match the actual effective lifespan as accurately as possible. Weighting parameters and These are learnable model parameters. During training, these weights, like other weights in the network, are automatically adjusted and updated through backpropagation and gradient descent optimizers. This learnable weight mechanism in the technical solution of this embodiment allows the model to autonomously evaluate and balance the impact of the losses from two tasks on the overall optimization objective during training dynamics, thereby coordinating the learning pace of different tasks and preventing a single task from excessively dominating parameter updates and causing a decline in the performance of another task.

[0036] In an optional embodiment, the step of dynamically adjusting the air purifier's fan power based on the predicted filter resistance value, the predicted remaining effective lifespan of the filter, and real-time environmental parameters using an adaptive algorithm incorporating nonlinear compensation may include: Based on the predicted filter resistance value With the preset resistance threshold The ratio is used to calculate the basic adjustment amount of the fan power; Based on the current ambient humidity Cumulative running time of the filter Calculate the nonlinear compensation amount for the wind turbine power; The adjusted wind turbine power setting value is determined based on the sum of the basic adjustment amount and the nonlinear compensation amount.

[0037] In this embodiment, the process of dynamically adjusting the wind turbine power is a composite calculation process that integrates multi-source prediction information and environmental parameters. This process can specifically include three sequential calculation sub-steps. First, based on the predicted filter resistance value... With preset resistance threshold Based on the ratio relationship, a basic adjustment amount for the fan power is calculated. The purpose of this step is to ensure that the power adjustment responds proportionally to the current and predicted degree of filter clogging. Secondly, it is independently based on the real-time monitored ambient humidity. Cumulative running time of the filter A nonlinear compensation amount for the wind turbine power is calculated using a specific nonlinear functional relationship. Finally, the aforementioned basic adjustment amount and the nonlinear compensation amount are summed to determine the final power setpoint used to control the wind turbine operation.

[0038] Specifically, the basic adjustment in the first sub-step constructs the backbone of the power adjustment, and its response logic can be: when the predicted resistance... Below the threshold At the same time, the basic adjustment will guide the fan power to adjust towards the reference value or a lower level. When the predicted resistance approaches or exceeds the threshold, the basic adjustment will drive the fan power to increase accordingly to overcome the increased filter resistance and maintain the established air purification efficiency. The nonlinear compensation introduced in the second sub-step is a fine correction term superimposed on the above-mentioned main adjustment. It is used to handle the indirect effects of changes in ambient humidity on airflow characteristics and apparent filter resistance. It also considers the evolution of the physical properties of the filter material (such as its sensitivity to humidity) as it ages over time, allowing the power adjustment to adapt to different environmental conditions and equipment life cycle stages.

[0039] Ultimately, in this embodiment, by linearly summing the third sub-step, the adjustment of the main problem (filter resistance) can be organically combined with the compensation for secondary and complex disturbance factors (ambient humidity and time decay), thereby outputting a comprehensive, stable and adaptive fan power setting command.

[0040] In an optional embodiment, the step of basing the current ambient humidity... Cumulative running time of the filter The steps for calculating the nonlinear compensation amount of the wind turbine power can be performed using the following formula: Among them, symbols Represents the nonlinear compensation quantity, symbol Indicates reference humidity, symbol Represents the time reference constant, symbol Indicates the calibrated compensation coefficient, symbol Represents the hyperbolic tangent function, symbol This represents an exponential function.

[0041] In this embodiment, the formula used to calculate the nonlinear compensation amount of the wind turbine power is... It takes into account the influence of two physical factors: ambient humidity and equipment operating time. Among them, the hyperbolic tangent function... Acting on real-time ambient humidity value Compared with preset reference humidity The function's role in the difference is to generate a smooth and bounded compensation response to the humidity deviation. Its characteristics allow the output to change approximately linearly when the humidity deviation is small, and to saturate as the deviation increases. This effectively suppresses frequent fan power oscillations caused by instantaneous fluctuations in the humidity sensor or small fluctuations in ambient humidity, ensuring control stability. Simultaneously, the exponential function... Effect on the cumulative running time of the filter With time reference constant The ratio of these factors constitutes a decay factor. This part can be used to simulate the natural process of filter performance gradually aging over time. Its function is to make the compensation effect caused by the same humidity deviation decrease exponentially with the accumulation of equipment operating time.

[0042] Meanwhile, in this scheme, the scaling factor As a pre-factor, it can determine the maximum possible amplitude of the entire nonlinear compensation. Sensitivity coefficient Embedded in The input to the function can control the slope of the function curve near zero, that is, how sensitive the humidity deviation is to the effect of the compensation amount. The larger the value, the faster the compensation amount will approach its saturation limit for the same humidity deviation. Attenuation coefficient. It lies within the exponential term of the exponential function and can dominate the rate at which the compensation decreases with increasing cumulative running time. The higher the value, the faster the degradation occurs, indicating that the method considers the weakening effect of filter aging on humidity sensitivity earlier. (Reference humidity) This is a preset baseline value that can be set to the humidity of a typical comfortable environment or the nominal operating condition humidity of the equipment. Time reference constant. This can provide a normalized scale for runtime, making the decay rate coefficient... It has a clear physical meaning.

[0043] In practical engineering scenarios, through calibration , , This set of coefficients allows the compensation behavior to be accurately matched to the specific characteristics of the filter material, the performance of the fan, and the typical operating environment, so that the final calculated result is accurate. It is a dynamic, bounded compensation value that decays over time.

[0044] In an optional embodiment, the adjusted fan power setting value Calculated using the following formula: Among them, symbols Indicates the reference fan power setting value, symbol This represents the power regulation coefficient.

[0045] In this embodiment, the final wind turbine power setting value It is determined by adding a base adjustment amount based on filter resistance prediction and an independent nonlinear compensation amount. Wherein, This constitutes the basic framework for power adjustment. Among them, the reference turbine power setpoint... This represents a preset normal operating power level when the filter is in a healthy or initial state. The calculation process first evaluates the predicted filter resistance value. Relative to the preset resistance threshold The ratio, this ratio This allows for a direct and quantitative quantification of the degree to which the predicted resistance deviates from the critical level. Subtracting 1 from this ratio yields the relative deviation, which is then processed by a configurable power adjustment coefficient. By scaling the power and then multiplying and summing it with the base power, we can obtain the basic power adjustment result required to cope with changes in filter resistance. This part ensures that the fan power can be linked proportionally to changes in the predicted filter resistance.

[0046] Meanwhile, power regulation coefficient It can determine the effect of fan power on filter resistance changes. The response intensity or sensitivity, The magnitude of the value directly affects the range of power adjustment, i.e., when the predicted resistance... Below the threshold When the result within the parentheses is negative, the calculated base power will be lower than the baseline value, achieving energy-saving operation. When the predicted resistance reaches or exceeds the threshold, the result will be positive or zero, and the base power will be increased accordingly to overcome the increased resistance and maintain the preset air purification capacity. (The formula is added at the end.) The term represents an independent nonlinear compensation quantity, the purpose of which is to handle additional influences introduced by real-time parameters such as ambient humidity that are not included in the basic resistance-power relationship. Therefore, in this embodiment, the adjustment of the fan power can not only respond to the core resistance state of the filter, but also incorporate compensation for other environmental factors, thereby achieving a more refined and adaptive control objective.

[0047] In an optional embodiment, the step of performing a fusion decision analysis based on the predicted filter resistance value, the predicted index of the remaining effective lifespan of the filter, and the constructed multi-dimensional filter health status vector to determine whether to trigger a maintenance prompt may include: If the predicted value of the filter resistance Meet the conditions If so, a maintenance prompt will be triggered; and / or, If the conditions are met This will trigger a maintenance prompt, where the symbol... This represents the preset threshold for continuous increase in resistance, symbol... Indicates and The predicted value of the filter resistance at the adjacent previous prediction time; And / or, if the predicted value of the remaining effective lifespan of the filter Meet the conditions If this happens, a maintenance prompt will be triggered.

[0048] In this embodiment, the purpose of fusion decision analysis is to comprehensively utilize multiple predictive indicators to determine whether filter maintenance is necessary. This analysis is mainly based on three logical conditions. First, by comparing the predicted values ​​of filter resistance... With a preset fixed resistance threshold If the predicted resistance reaches or exceeds this threshold, it indicates that the filter is clogged to a critical level requiring intervention. Secondly, by examining the recent trend in filter resistance, i.e., calculating the current predicted value... The predicted value relative to its previous adjacent prediction time. The relative growth rate, and compare this growth rate with a preset continuous growth rate threshold. By comparison, if the increase exceeds this threshold, it means that the filter resistance is rising rapidly, which may indicate that serious clogging is imminent. Third, the predicted remaining effective lifespan of the filter is evaluated by directly outputting the multi-task deep learning prediction model. This is compared with a preset lifespan threshold, or in other words, the critical value of the remaining effective lifespan of the filter. By comparison, if the predicted remaining lifespan is lower than this threshold, it indicates that the filter's performance is nearing its end, based on the overall lifespan of the filter. Meeting any of the above conditions can independently trigger a maintenance reminder.

[0049] Specifically, in this embodiment, the first condition provides a static judgment benchmark based on absolute resistance value, ensuring that an alert is issued when filter performance deteriorates to a clear limit. The second condition focuses on dynamic trend monitoring, which can capture abnormally accelerated increases in resistance value, even before the absolute value reaches a fixed threshold. This occurs much earlier, allowing for earlier warnings. Among these, the parameters... A critical percentage for "continuous increase" is defined to filter out rapidly growing trends that require attention. The third condition introduces the comprehensive prognostic indicator of remaining effective lifespan, which predicts the duration of the filter's operation from its current state to complete failure, providing another dimension for judgment based on the end of its service life. This embodiment's technical solution, through this logical architecture combining "absolute threshold," "change trend," and "lifespan prognosis," enables cross-validation and fusion decision-making regarding the filter's health status from different perspectives and time scales, thereby improving the accuracy and reliability of maintenance prompts and avoiding false alarms or missed alarms that may result from a single judgment condition.

[0050] In optional embodiments, the technical solution may further include the steps of performing closed-loop performance monitoring and online adaptive fine-tuning on the multi-task deep learning prediction model; The closed-loop performance monitoring may include: calculating a model performance drift index based on the historical sequence of the real-time filter resistance value and the corresponding predicted filter resistance value. The calculation formula is as follows: Among them, symbols Indicates the length of the evaluation window; When the performance drift index When the threshold is continuously exceeded, the online adaptive fine-tuning is triggered.

[0051] In this embodiment, the purpose of closed-loop performance monitoring is to continuously evaluate the prediction accuracy of the multi-task deep learning prediction model in actual deployment and operation. This monitoring can be achieved through a quantified performance drift index. To achieve this, the calculation of this indicator is based on a historical data sequence over a period of time. Specifically, it selects historical data from the most recent M moments, and for each historical moment... Calculate the predicted value of the filter resistance at that moment. The actual value of the corresponding real-time filter resistance obtained afterward The relative error between them. Performance drift index. The calculation formula is: This formula calculates the root mean square value of the relative error of the model prediction within a specified evaluation window length M. Its magnitude can directly reflect the average magnitude and stability of the model's prediction results deviating from the actual measured values ​​within that time period.

[0052] In the technical solution of this embodiment, the performance drift index As a monitoring criterion, its continuous exceeding means that the model's predictive ability has declined statistically significantly and continuously, that is, "performance drift" has occurred. When the indicator continuously exceeds the preset threshold within a continuous monitoring period, it is determined that the current state of the model can no longer meet the prediction accuracy requirements, and then the online adaptive fine-tuning step is triggered. That is, using the latest accumulated actual operation data, some or all of the parameters of the multi-task deep learning prediction model are recalibrated or slightly updated, so that the internal mapping relationship of the model can adapt to the changes in data distribution caused by external factors such as slow changes in filter characteristics, changes in environmental conditions, or sensor feature shifts.

[0053] In an optional embodiment, the constructed multidimensional filter health status vector may include at least the real-time filter resistance value, its historical rate of change, cumulative operating time, and a prediction index of the remaining effective lifespan of the filter.

[0054] In this embodiment, the filter resistance value is calculated in real time. The instantaneous obstruction of airflow by the filter can be directly quantified. The historical rate of change of the real-time filter resistance value can be obtained by calculating the derivative, difference, or average slope of the recent resistance value series within a specific time window, which describes the dynamic rate and trend of resistance growth over time. The cumulative operating time of the filter... This allows for the accumulating time from the start of equipment operation or the last filter replacement, reflecting the total operating time and aging background of the filter. It is a predictive indicator of the remaining effective lifespan of the filter, output by a multi-task deep learning prediction model. This is an estimate of the filter's expected sustainable and effective working time from the current moment.

[0055] Specifically, real-time filter resistance value This constitutes the most direct, instantaneous observation point for health assessment, reflecting the filter's "current symptoms." Its historical rate of change provides a perspective on the speed of symptom development; a rapidly increasing rate of change may indicate that the filter is clogging faster, even if the current absolute resistance value is not yet high. Cumulative running time As an objective record of its "journey," it provides a temporal context for evaluation, as many performance degradations of filters are inherently correlated with runtime. Filter remaining effective life prediction metrics. This introduces a "prognostic" judgment based on model reasoning, which is a prediction of the future state of the filter from the perspective of the overall life cycle.

[0056] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps may be adjusted according to actual needs, or some steps may be omitted.

[0057] Based on the foregoing technical solutions, the present invention also provides a self-testing device for the status of an air purifier filter, such as... Figure 2 As shown, the device, from a macroscopic perspective, may include the following modules: The sensor data acquisition module 202 is used to collect air flow rate data and pressure difference data on both sides of the air purifier filter in real time at the filter inlet and outlet. The filter resistance calculation module 204 is used to calculate the real-time filter resistance value based on the air velocity data and the pressure difference data. The multi-task deep learning prediction module 206 is used to input a data sequence containing relevant data within a historical time window into a trained multi-task deep learning prediction model, so as to simultaneously output the predicted value of filter resistance and the predicted index of the remaining effective life of the filter at future times. The fusion decision and maintenance prompt module 208 is used to perform fusion decision analysis based on the predicted value of filter resistance, the predicted index of the remaining effective life of the filter, and the constructed multi-dimensional filter health status vector, in order to determine whether to trigger a maintenance prompt. The fan power control module 210 is used to dynamically adjust the fan power of the air purifier based on the predicted value of the filter resistance, the predicted index of the remaining effective life of the filter, and real-time environmental parameters, using an adaptive algorithm that includes nonlinear compensation.

[0058] Those skilled in the art will understand that the modules in the apparatus of the foregoing embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules, that is, the module division can be flexibly performed to implement the method embodiments described above.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A self-test method for the condition of an air purifier filter, characterized in that, Including the following steps: Real-time data collection of airflow velocity at the inlet and outlet of the air purifier filter, as well as pressure difference data across the filter; Based on the air velocity data and the pressure difference data, the real-time filter resistance value is calculated; The data sequence containing relevant data within the historical time window is input into a trained multi-task deep learning prediction model to simultaneously output the predicted values ​​of filter resistance and the remaining effective lifespan of the filter at future moments. Based on the predicted value of filter resistance, the predicted index of the remaining effective life of the filter, and the constructed multi-dimensional filter health status vector, a fusion decision analysis is performed to determine whether to trigger a maintenance prompt. Based on the predicted filter resistance, the predicted remaining effective lifespan of the filter, and real-time environmental parameters, the fan power of the air purifier is dynamically adjusted using an adaptive algorithm that includes nonlinear compensation.

2. The self-test method for the condition of an air purifier filter as described in claim 1, characterized in that, The step of calculating the real-time filter resistance value based on the air velocity data and the pressure difference data uses the following formula: Among them, symbols Indicates air density, symbol This represents the pressure difference across the filter at the current sampling time, with the symbol... Indicates the air velocity at the current sampling time, symbol This indicates the real-time filter resistance value.

3. The self-test method for the condition of an air purifier filter as described in claim 1, characterized in that, The multi-task deep learning prediction model employs a shared encoder network and at least two independent decoder networks. The shared encoder is used to extract shared temporal features from the input data sequence; The first decoder is used to output the predicted value of the filter resistance. ; The second decoder is used to output the predicted value of the remaining effective lifespan of the filter. ; The total loss function of the multi-task deep learning prediction model The result is obtained by weighted summation of the losses from each task: in, This is the mean square error loss for the prediction of filter resistance. Loss predicted for remaining lifetime. and These are learnable task weight parameters.

4. The self-test method for the condition of an air purifier filter as described in claim 1, characterized in that, The step of dynamically adjusting the fan power of the air purifier based on the predicted filter resistance value, the predicted index of the remaining effective lifespan of the filter, and real-time environmental parameters using an adaptive algorithm including nonlinear compensation includes: Based on the predicted filter resistance value With the preset resistance threshold The ratio is used to calculate the basic adjustment amount of the fan power; Based on the current ambient humidity Cumulative running time of the filter Calculate the nonlinear compensation amount for the wind turbine power; The adjusted wind turbine power setting value is determined based on the sum of the basic adjustment amount and the nonlinear compensation amount.

5. The self-test method for the condition of an air purifier filter as described in claim 4, characterized in that, The step of performing a fusion decision analysis based on the predicted filter resistance value, the predicted index of the remaining effective lifespan of the filter, and the constructed multi-dimensional filter health status vector to determine whether to trigger a maintenance prompt includes: If the predicted value of the filter resistance Meet the conditions If so, a maintenance prompt will be triggered; and / or, If the conditions are met This will trigger a maintenance prompt, where the symbol... This represents the preset threshold for continuous increase in resistance, symbol... Indicates and The predicted value of the filter resistance at the adjacent previous prediction time; And / or, if the predicted value of the remaining effective lifespan of the filter Meet the conditions If this occurs, a maintenance prompt will be triggered; where the symbol... This indicates the critical value of the remaining effective lifespan of the filter.

6. The self-test method for the condition of an air purifier filter as described in claim 4, characterized in that, Based on the current ambient humidity Cumulative running time of the filter The steps for calculating the nonlinear compensation amount of the wind turbine power are as follows: Among them, symbols Represents the nonlinear compensation quantity, symbol Indicates reference humidity, symbol Represents the time reference constant, symbol Indicates the calibrated compensation coefficient, symbol Represents the hyperbolic tangent function, symbol This represents an exponential function.

7. The self-test method for the condition of an air purifier filter as described in claim 6, characterized in that, The adjusted fan power setting value Calculated using the following formula: Among them, symbols Indicates the reference fan power setting value, symbol This represents the power regulation coefficient.

8. The self-test method for the condition of an air purifier filter as described in claim 1, characterized in that, It also includes the steps of performing closed-loop performance monitoring and online adaptive fine-tuning on the multi-task deep learning prediction model; The closed-loop performance monitoring includes: calculating the model performance drift index based on the historical sequence of the real-time filter resistance value and the corresponding predicted filter resistance value. The calculation formula is as follows: Among them, symbols Indicates the length of the evaluation window; When the performance drift index When the threshold is continuously exceeded, the online adaptive fine-tuning is triggered.

9. The self-test method for the condition of an air purifier filter as described in claim 1, characterized in that, The constructed multidimensional filter health status vector includes at least the real-time filter resistance value, its historical rate of change, cumulative operating time, and a predictive index of the remaining effective lifespan of the filter.

10. A self-testing device for the status of an air purifier filter, characterized in that, include: The sensor data acquisition module is used to collect real-time airflow velocity data at the inlet and outlet of the air purifier filter and pressure difference data on both sides of the filter. The filter resistance calculation module is used to calculate the real-time filter resistance value based on the air velocity data and the pressure difference data. The multi-task deep learning prediction module is used to input a data sequence containing relevant data within a historical time window into a trained multi-task deep learning prediction model, so as to simultaneously output the predicted values ​​of filter resistance and the remaining effective lifespan of the filter at future moments. The fusion decision and maintenance prompt module is used to perform fusion decision analysis based on the predicted value of filter resistance, the predicted index of the remaining effective life of the filter, and the constructed multi-dimensional filter health status vector, in order to determine whether to trigger a maintenance prompt. The fan power control module is used to dynamically adjust the fan power of the air purifier based on the predicted value of the filter resistance, the predicted index of the remaining effective life of the filter, and real-time environmental parameters, using an adaptive algorithm that includes nonlinear compensation.

Images