An electrostatic dust removal dynamic control method based on multi-source data fusion

Abstract

The application discloses a kind of electrostatic precipitation dynamic control methods based on multi-source data fusion, belong to electrostatic precipitation technical field, technical scheme is, obtain particulate matter parameter, environmental parameter, operating parameter;Particulate matter parameter, environmental parameter and operating parameter are handled, and based on the data after processing constructs the feature quantity set for representing indoor dust load state, environmental influence state, device dust state and discharge risk state;Multi-source data fusion is carried out to the feature quantity set, and fusion state quantity is obtained;Based on fusion state quantity, indoor scene recognition is carried out, and scene recognition result is output;According to fusion state quantity and scene recognition result, according to preset control criterion, the control quantity of electrostatic precipitation device is calculated;Control quantity is output to electrostatic precipitation device to adjust electrostatic field working state, and in subsequent control cycle, repeat execution updates control quantity.The beneficial effects of the application are: provide a kind of electrostatic precipitation dynamic control method based on multi-source data fusion.

Description

Technical Field

[0001] This invention belongs to the field of electrostatic dust removal technology, and specifically relates to a dynamic control method for electrostatic dust removal based on multi-source data fusion. Background Technology

[0002] Indoor air commonly contains pollutants such as suspended particulate matter (e.g., PM2.5, PM10), cooking fumes aerosols, dust generated from cleaning and human / pet activities, and volatile organic compounds (VOCs). To improve indoor air quality, air purification equipment is widely used. Among these, electrostatic precipitators, due to their high capture efficiency for fine particulate matter, low pressure drop, and continuous operation, are used in home indoor air purification, fume purification, and air circulation.

[0003] Existing electrostatic precipitators typically include: a fan module (such as a centrifugal or axial fan) for forming airflow channels and providing circulating air volume; an electrode structure (such as discharge electrodes and dust collection plates / screens) for ionizing and capturing dust-laden gas; a high-voltage power supply module for providing the high-voltage electric field required for ionization and capture; and control circuits for equipment control and protection. The high-voltage power supply module can generally output DC high voltage or pulsed high voltage, and the ionization intensity and electric field distribution can be changed by adjusting parameters such as output voltage / current, pulse frequency, duty cycle, and rise slope. Some devices also have polarity switching or bipolar operation modes to reduce dust accumulation on the plates and improve capture stability. The fan module can control the airflow by adjusting its speed, thereby affecting the airflow residence time, capture efficiency, and equipment noise and energy consumption.

[0004] In residential indoor applications, the operating status of electrostatic precipitators is significantly affected by environmental conditions. For example, temperature and humidity affect the dielectric properties of air, the initiation conditions of corona discharge, the formation of water films on the electrode surface, and the insulation level; increased humidity makes it easier to increase leakage current, micro-discharge, and even the risk of breakdown. Oil fume aerosols and VOCs can, in some cases, lead to increased pollution on the electrode surface, forming conductive films or adhesion layers, increasing the probability of leakage current and discharge events; at the same time, dust accumulation on the electrode plates can change the electric field distribution and lead to a decrease in dust removal efficiency. To ensure operational safety and reliability, existing technologies typically monitor operating parameters such as leakage current and discharge events on the high-voltage circuit or electrode structure, and trigger protection in case of abnormalities, such as reducing the high-voltage setpoint, reducing the duty cycle, reducing the rise slope, switching the operating mode, or reducing the fan speed.

[0005] However, indoor pollution sources and environmental conditions in homes exhibit significant variability and scenario-specificity: for example, sleepers are more sensitive to noise at night, cooking fumes and VOCs increase dramatically in a short period, PM10 levels rise and fluctuate significantly during cleaning, pet activity generates intermittent dust, and the risk of electrical discharge increases significantly in high-humidity environments. Existing electrostatic precipitators often employ fixed-level control or simple closed-loop control (e.g., adjusting high voltage and fan based on a single particulate matter concentration or simple threshold rules), which often struggles to simultaneously address dust removal efficiency, energy consumption, noise constraints, and electrical discharge risk constraints. Furthermore, due to the diverse and rapidly changing sources of sensor data, the lack of multi-source data fusion and scenario recognition mechanisms can easily lead to problems such as mismatched control strategies, frequent switching, high energy consumption, excessive noise, or increased electrical discharge risk. Therefore, there is an urgent need for an electrostatic precipitator control method capable of operating across multiple scenarios within a home, integrating multi-source data, and implementing dynamic closed-loop adjustments to improve its adaptability, stability, and safety in real-world home environments. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic control method for electrostatic dust removal based on multi-source data fusion.

[0007] This invention is achieved through the following measures: a dynamic control method for electrostatic precipitators based on multi-source data fusion, characterized in that, S1. Acquire particulate matter parameters and environmental parameters of indoor air, and simultaneously acquire the operating parameters of the electrostatic precipitator; S2. Process the particulate matter parameters, environmental parameters, and operating parameters, and construct a set of feature quantities based on the processed data to characterize the indoor dust load state, environmental impact state, device dust accumulation state, and discharge risk state. S3. Perform multi-source data fusion on the set of features to obtain fused state quantities; wherein the fused state quantities include particle concentration state, environmental impact factor, device state factor and discharge risk factor.

[0008] S4. Based on the fused state variables, perform indoor scene recognition and output scene recognition results; wherein the scene recognition results include sleep silent scene, cooking fume scene, cleaning dust scene, pet activity dust scene, and high humidity environment scene. S5. Based on the fusion state quantity and the scene recognition result, calculate the control quantity of the electrostatic dust removal device according to the preset control criterion; wherein the control quantity includes high voltage drive parameters and fan operating parameters, and the preset control criterion is used to weigh dust removal efficiency, energy consumption, noise constraints and discharge risk constraints, and to match the control quantity with the scene recognition result. S6. Output the control quantity to the electrostatic dust removal device to adjust the working state of the electrostatic field, and repeat S1 to S5 in subsequent control cycles to update the control quantity, thereby realizing dynamic closed-loop control of indoor electrostatic dust removal.

[0009] Furthermore, the particulate matter parameters include PM2.5 and PM10; the environmental parameters include temperature, humidity, oil fume concentration, volatile organic compound (VOC) concentration, noise intensity, and wind speed; and the operating parameters include high-voltage output voltage or high-voltage output current, leakage current, and discharge event information.

[0010] Furthermore, time-domain filtering is performed on the particulate matter parameters and the environmental parameters to obtain smoothed data; noise reduction processing is performed on the leakage current signal in the operating parameters to obtain the leakage current baseline component and the peak component; based on the smoothed data, the leakage current baseline component, and the peak component, a set of feature quantities is constructed.

[0011] Furthermore, the set of features includes: Characteristic quantities used to characterize indoor dust load status: PM2.5 concentration, PM10 concentration, PM2.5 concentration change rate, PM10 concentration change rate, and particulate matter concentration fluctuation intensity; Characteristic quantities used to characterize the state of environmental impact: temperature, humidity, oil fume concentration, VOCs concentration, wind speed, and the rate of change of the above environmental parameters; Characteristic quantities used to characterize the dust accumulation state of the device: wind speed attenuation, fan load change, and leakage current baseline drift. Characteristic quantities used to characterize discharge risk states: leakage current spike amplitude, leakage current spike count, leakage current spike energy, and discharge event count. Further, S3 includes: The set of features is normalized, and corresponding confidence weights are generated based on the fluctuation intensity and saturation state of each feature. The fusion state is obtained based on the normalized feature quantities and their confidence weights, where the fusion state is updated by the fusion state quantity of the previous time step and the feature quantity of the current time step. Furthermore, the fusion state quantities include: Particulate concentration status: a particulate load status quantity obtained from PM2.5 concentration, PM10 concentration and their rate of change; Environmental impact factors: environmental impact state quantities obtained from temperature, humidity, oil fume concentration, VOCs concentration, noise intensity, and wind speed; Device state factor: A device dust accumulation state quantity obtained from wind speed attenuation, fan load change, and leakage current baseline drift; Discharge risk factor: a discharge risk state quantity obtained from leakage current spike amplitude, leakage current spike count, leakage current spike energy, and discharge event count.

[0012] Further, S4 includes: The particle concentration state, environmental impact factor, device state factor, and discharge risk factor in the fusion state quantity are constructed as a scene discrimination input vector. The confidence level of each scene is obtained based on the scene input vector; Based on the confidence level of each scenario, a scenario candidate set is selected: the scenario with the highest confidence level is selected as the scenario recognition result; when the maximum confidence level is lower than the preset confidence level threshold, the output is a mixed scenario, and the top two scenarios with the highest confidence level in the mixed scenario are recorded as concurrent scenario recognition results. A hold time and a switching hysteresis threshold are set for the scene recognition result. The current scene recognition result is maintained within the hold time, and the scene recognition result is updated only when the confidence of a new scene continuously exceeds the switching hysteresis threshold.

[0013] Furthermore, the rule-based discrimination model determines the scene candidate set based on the following conditions: When the noise intensity is lower than the preset noise threshold, the wind speed is lower than the preset wind speed threshold, and the rate of change of particle concentration is lower than the preset rate of change threshold, the sleep silent scene is added to the scene candidate set. When the concentration of cooking fumes and VOCs are both higher than the corresponding preset concentration thresholds, and the particle concentration shows an upward trend within a preset time, the cooking fume scene is added to the scene candidate set. When the proportion of PM10 relative to PM2.5 is higher than the preset proportion threshold, and the fluctuation intensity of the particulate concentration is higher than the preset fluctuation threshold, the clean dust scene will be added to the scene candidate set. When the particulate concentration fluctuates intermittently and the proportion of PM10 is higher than the preset proportion threshold, and the wind speed is in the preset low to medium wind speed range, the pet activity dust scene will be added to the scene candidate set. When the humidity is higher than the preset humidity threshold, and the discharge risk factor increases or the leakage current baseline drift corresponding to the device status factor is greater than the preset drift threshold, the high humidity environment scenario is added to the scenario candidate set.

[0014] Further, S5 includes: A control objective function is constructed based on the fused state variables, and target weights and constraint thresholds are set according to the scene recognition results. The control objective function includes a particle concentration reduction term to characterize the dust removal effect, a power consumption term to characterize energy consumption, a noise penalty term to characterize noise, and a risk penalty term to characterize discharge risk. Constraints are set such that the noise intensity does not exceed a preset noise limit, the discharge risk factor does not exceed a preset risk limit, and the high voltage limit is reduced when the humidity is higher than a preset humidity threshold.

[0015] A set of candidate control quantities is generated within a preset control cycle. The set of candidate control quantities includes several combinations of high-voltage drive parameters and fan operating parameters. The high-voltage drive parameters include the set value of high-voltage output voltage or high-voltage output current, pulse frequency, duty cycle, rise edge slope, and polarity switching strategy. The fan operating parameters include the set value of fan speed.

[0016] For each set of candidate control variables, the particle concentration change, power consumption, noise intensity, and discharge risk change in the next control cycle are predicted based on the fused state variables, and the corresponding comprehensive cost value is calculated; wherein the comprehensive cost value is calculated by the control objective function, and an infeasibility mark or additional penalty term is applied to candidate control variables that violate the constraints.

[0017] The group with the lowest comprehensive cost value is selected from the candidate control quantities that have never been marked as infeasible and output to the electrostatic dust removal device so that the control quantity matches the scene recognition result.

[0018] When the scene identification result is a sleep / silent scene, the noise penalty weight is increased and the fan speed setting value is decreased. Simultaneously, the high-voltage output setting value is decreased, and the dust removal effect is maintained by increasing the pulse frequency and adjusting the duty cycle. When the scene identification result is a cooking fume scene, the dust removal effect weight is increased and the fan speed setting value is increased. Simultaneously, the high-voltage output setting value is increased. Further, S6 includes: At the beginning of each control cycle, the control quantity is decomposed into a high-voltage drive command and a fan drive command, and output to the high-voltage power supply module and the fan module respectively; wherein the high-voltage drive command is used to set the high-voltage output voltage or high-voltage output current, pulse frequency, duty cycle, rise edge slope and polarity switching strategy, and the fan drive command is used to set the fan speed.

[0019] When updating the high-voltage drive command, a soft-start process is performed, which includes progressively increasing the high-voltage output setpoint within a preset time. When a leakage current exceeding a preset leakage threshold or a discharge event characterized by a discharge event occurs, a protection strategy is triggered. The protection strategy includes one or more of the following: reducing the high-voltage output setpoint, reducing the duty cycle, reducing the rise edge slope, switching the polarity switching strategy, and reducing the fan speed.

[0020] During the control cycle, updated particulate matter parameters, environmental parameters, and operating parameters are collected, and dust removal effect evaluation quantities are calculated. The dust removal effect evaluation quantities include the particulate matter concentration reduction rate and the dust removal amount per unit energy consumption. Based on the dust removal effect evaluation quantities, the candidate control quantity generation range or control objective function weight for the next control cycle is corrected.

[0021] In subsequent control cycles, steps S1 to S5 are repeated to update the control variables based on the latest fused state variables and scene recognition results, so as to achieve dynamic closed-loop control of indoor electrostatic dust removal.

[0022] When the change in the control quantity in an adjacent control cycle exceeds a preset change threshold, the control quantity is subjected to amplitude limiting or gradual change processing to suppress sudden changes in the electrostatic field working state caused by scene switching or transient fluctuations.

[0023] The beneficial effects of the technical solution provided by this invention are as follows: More accurate state representation driven by multi-source data: By simultaneously utilizing environmental parameters such as particulate matter, temperature and humidity / oil fumes / VOCs / noise / wind speed, and operational parameters such as high-voltage output, leakage current, and discharge events, multi-dimensional features such as dust load, environmental impact, device dust accumulation, and discharge risk are constructed and fused to obtain a fused state quantity. This makes the device's representation of the coupled state of "pollution-environment-device" more comprehensive, reducing misjudgments and miscontrols caused by single sensors or single indicators. Through indoor scene recognition based on the fused state quantity, control weights and constraint strategies can be dynamically selected in different scenarios such as quiet sleep, cooking fumes, cleaning dust, pet dust, and high humidity environments, ensuring that high-voltage drive parameters and fan operating parameters match the scenario requirements. Achieving a balance between efficiency, energy consumption, noise, and safety: By evaluating and selecting the best candidate control variables through preset control criteria (objective function + constraints), it can suppress power consumption and noise while ensuring a reduction in particulate matter concentration, and constrain discharge risks. This achieves multi-objective comprehensive optimal or near-optimal control at the system level, overcoming the limitations of traditional fixed-level or single-objective control. Utilizing risk characterization quantities such as leakage current baseline and peak components, and discharge event counts, combined with soft-start, limiting / gradual, and protective load reduction strategies, it can reduce the risk of transient discharge and breakdown caused by humidity changes, plate contamination, or sudden scene changes, improving the stability of the electrostatic field and extending the reliable operating time of the device. By collecting updated parameters and calculating dust removal effect evaluation quantities (such as concentration reduction rate and dust removal capacity per unit energy consumption) in each control cycle, the candidate control range or target weights for subsequent cycles are corrected, achieving dynamic closed-loop control and performance self-optimization, thus better adapting to long-term changes and individual differences in real-world home environments. Attached Figure Description

[0024] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings listed below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a dynamic control method for electrostatic dust removal based on multi-source data fusion in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0027] See Figure 1A dynamic control method for electrostatic precipitators based on multi-source data fusion, characterized in that... S1. Acquire particulate matter parameters and environmental parameters of indoor air, and simultaneously acquire the operating parameters of the electrostatic precipitator; Let the sampling sequence number k represent the kth sampling period, and the corresponding sampling time be denoted as k. The parameters are then denoted as: PM2.5 concentration is recorded as PM10 concentration is recorded as Temperature is recorded as Humidity is recorded as The concentration of cooking fumes is recorded as follows: Volatile organic compounds are denoted as VOCs concentration. Noise intensity is denoted as Wind speed is recorded as Operating parameters include the high-voltage output voltage, denoted as... Or the high voltage output current is denoted as Leakage current is denoted as Discharge event information is recorded as .

[0028] When used to identify cooking fumes, oily particles tend to adhere to the surface of the dust collection plate / electrode, leading to efficiency degradation and changes in the leakage current baseline.

[0029] temperature ,humidity Humidity affects ionization efficiency, dielectric breakdown threshold, formation of water film on electrode surface, and leakage current. At high speeds, it is necessary to reduce the field strength / limit the rising edge to suppress the risks of discharge and ozone byproducts. temperature Sensor drift is affected by air viscosity.

[0030] VOCs concentration Together with cooking fumes, it characterizes cooking / cleaning agents / volatile sources, assisting in scene recognition and control weight adjustment.

[0031] Noise intensity Used for sleep mute constraint input (upper limit constraint / penalty term) to suppress noise from fans and high-pressure regulators.

[0032] wind speed It affects the residence time of particles in the electric field and the capture probability.

[0033] High voltage output voltage or high voltage output current It determines the ionization intensity, trapping field strength, and energy consumption; it is also an important basis for judging whether the output reaches the set value or whether there is an abnormality.

[0034] Leakage current An increase in leakage current often indicates humidity, board contamination, or decreased insulation, and is a precursor signal of discharge / breakdown risk.

[0035] Discharge event information It reflects the breach of the electric field operating boundary and can be used for safety protection and parameter adaptive correction.

[0036] S2. Process the particulate matter parameters, environmental parameters, and operating parameters, and construct a set of feature quantities based on the processed data to characterize the indoor dust load state, environmental impact state, device dust accumulation state, and discharge risk state. For any sequence of the particulate matter parameters and the environmental parameters that needs filtering:

[0037] Time-domain filtering is performed to obtain smoothed data.

[0038] but ,in This is the smoothed value after time-domain filtering. For time-domain filtering operators, This represents the original sampled value for the k-th period. The time-domain filtering includes one or more of the following: moving average filtering, median filtering, and exponentially weighted moving average filtering.

[0039] The leakage current signal in the operating parameters is denoised to obtain the baseline component and spike component of the leakage current. (Leakage current baseline component) With peak component Leakage current The result of the decomposition is: . Based on the smoothed data, the baseline component of the leakage current, and the peak component, a set of feature quantities is constructed. The set of feature quantities includes: Characteristic quantities used to characterize indoor dust load status: , , , as well as ;in This is the smoothed PM2.5 concentration. This is the smoothed PM10 concentration. The rate of change in PM2.5 concentration. The rate of change of PM10 concentration. This represents the intensity of particulate matter concentration fluctuations.

[0040] Characteristic quantities used to characterize the state of environmental impact: , , , , and the rate of change of the above environmental parameters ;in, The temperature is after smoothing. The humidity value is after smoothing. This is the smoothed oil fume concentration value. VOCs concentration value, This is the smoothed wind speed value. These represent the rates of change of the corresponding environmental parameters.

[0041] The rates of change for PM2.5 concentration, PM10 concentration, and various environmental parameters are all calculated as the ratio of the smoothed difference between adjacent sampling periods to the sampling period. Taking PM2.5 concentration as an example, the PM2.5 concentration change rate... Specifically:

[0042] In the formula, To represent the smoothed PM2.5 concentration value after the kth sampling period, This represents the smoothed PM2.5 concentration value from the previous sampling period. This indicates the time interval between adjacent sampling periods.

[0043] Similarly, taking temperature as an example, the rate of temperature change Specifically:

[0044] In the formula, To represent the smoothed temperature value after the kth sampling period, This represents the smoothed temperature value from the previous sampling period. This indicates the time interval between adjacent sampling periods.

[0045] In addition, the intensity of particulate matter concentration fluctuations for:

[0046] in, The window length (number of sample points) used to calculate the fluctuation intensity. When the k-th sampling period is taken as the "current point", the most recent The average level of smoothed PM2.5 concentration within each sampling point; For the first Smoothed PM2.5 concentration for each sampling period, The summation symbol indicates that the summation is performed by adding the summation symbol to the summation symbol. All of these items added together.

[0047] Characteristic quantities used to characterize the dust accumulation state of the device: wind speed attenuation, fan load change, and leakage current baseline drift. Wind speed attenuation is When dust accumulation / clogging causes resistance to increase, the actual wind speed decreases and the attenuation increases.

[0048] In the formula, This is the wind speed attenuation. For reference wind speed (take the "calibrated wind speed under the same fan setting").

[0049] The change in fan load is When the resistance increases, the fan current rises, and the load change can help determine the dust accumulation trend.

[0050] In the formula, This represents the change in fan load. This is the smoothed fan current for the current sampling period. The smoothed fan current, This represents the smoothed current of the fan after the previous sampling period.

[0051] Leakage current baseline drift is Panel contamination / water film / insulation degradation can cause a slow rise in baseline, and the drift can be used for health assessment and risk constraints.

[0052] In the formula, This represents the baseline drift of the leakage current. This is the baseline component of the leakage current in the current sampling period. The baseline component of leakage current in the previous sampling period.

[0053] Characteristic quantities used to characterize discharge risk states include: leakage current spike amplitude, leakage current spike count, leakage current spike energy, and discharge event count. Let the risk calculation window length be... When performing risk calculation in the k-th sampling period, the sampling period corresponding to the samples within the window is: Equivalently using indexes This represents each sample point within the window.

[0054] Leakage current spike amplitude (for the most recent The maximum absolute value of the leakage current spike component within each sampling point is specifically expressed as:

[0055] In the formula, This represents the peak amplitude of the leakage current within the k-th period window. This indicates taking the maximum value within the range of window index i. For the first Periodic leakage current spike components.

[0056] Leakage current spike count (for the most recent) Within each sampling point, satisfying The number of sampling points is specifically expressed as:

[0057] In the formula, For the peak count within the k-th period window, The number of elements in the set (counting operator). The threshold for determining the peak value. This indicates that the sampling point was judged as a spike.

[0058] Leakage current spike energy (for the most recent Within each sampling point, the summation of the squares of the peak components multiplied by the sampling period. The obtained energy-type index is specifically expressed as follows:

[0059] The peak energy within the periodic window, To the window Summing of each sample, squared peak component , The sampling period is the peak component of the leakage current.

[0060] Discharge event count (in the most recent) Discharge event flag at each sampling point The accumulated value), specifically represented as:

[0061] In the formula, This represents the number of discharge events within the k-th period window.

[0062] S3. Perform multi-source data fusion on the set of features to obtain fused state quantities; wherein the fused state quantities include particle concentration state, environmental impact factor, device state factor and discharge risk factor.

[0063] S3 includes: normalizing the set of features and generating corresponding confidence weights based on the fluctuation intensity and saturation state of each feature. Specifically, the features obtained from S2 are concatenated into a feature vector: ; In the formula, This represents the eigenvector of the k-th period. Let M represent the i-th feature component, and M be the total feature dimension.

[0064] Normalization is performed on each feature, specifically as follows:

[0065] In the formula, This represents the i-th feature after normalization. This represents the statistical mean of the i-th feature. This represents the statistical standard deviation of the i-th feature. To prevent tiny constants with a denominator of zero, Greater than zero.

[0066] The fluctuation intensity is defined as follows for the i-th feature:

[0067] Let be the window variance (a measure of fluctuation intensity) of the i-th normalized feature. The window length is used to calculate the fluctuation intensity (variance), j represents the offset index within the window, and j represents the number of steps to backtrack from the current period k, where k is the index of the sampling period. For the k-th period, with the window length The window mean of the i-th normalized feature is calculated. This represents the period index corresponding to the j-th backtracking point within the window.

[0068] Define the saturation indicator (whether the sensor's upper / lower limits have been reached): , This indicates that the sensor corresponding to the i-th feature is in a saturated state; This indicates that the product is not saturated.

[0069] The credibility weight can be set as follows (the greater the fluctuation or the closer to saturation, the lower the weight; this is used to suppress the impact of unstable features on the fusion result):

[0070] In the formula, This represents the confidence weight of the i-th feature. This is the fluctuation suppression coefficient (value greater than 0). This is the saturation penalty coefficient.

[0071] The fusion state is obtained by normalizing the feature quantities and their confidence weights, and the fusion state is updated by the fusion state quantity of the previous time step and the feature quantity of the current time step.

[0072] The fusion state variable is defined as:

[0073] In the formula, This refers to the particle concentration state. As an environmental impact factor, For device state factors, Let T be the discharge risk factor, and T be the transpose operator.

[0074] The weighted aggregate values ​​for the four types of features are constructed as follows (taking particle concentration state as an example, the rest are similar):

[0075] Weighted observation of granular features; : A set of feature indexes belonging to the particle load class (e.g. (Corresponding index); the numerator is the weighted sum, and the denominator is the weight normalization factor.

[0076] Recursive update (updating the fused state from the previous time step with the feature values ​​from the current time step together) yields the particle concentration state: ; In the formula, Particle state fusion gain, This represents the particle state of the previous cycle.

[0077] Similarly:

[0078]

[0079]

[0080] In the formula, Corresponding state fusion gain (for environment, device, and discharge, respectively). The weighted aggregate values ​​of the corresponding category features (for environment, device, and discharge, respectively).

[0081] The fusion state variables include: Particulate concentration status: a particulate load status quantity obtained from PM2.5 concentration, PM10 concentration and their rate of change; Environmental impact factors: environmental impact state quantities obtained from temperature, humidity, oil fume concentration, VOCs concentration, noise intensity, and wind speed; Device state factor: A device dust accumulation state quantity obtained from wind speed attenuation, fan load change, and leakage current baseline drift; Discharge risk factor: a discharge risk state quantity obtained from leakage current spike amplitude, leakage current spike count, leakage current spike energy, and discharge event count.

[0082] S4. Based on the fused state variables, perform indoor scene recognition and output scene recognition results; wherein the scene recognition results include sleep silent scene, cooking fume scene, cleaning dust scene, pet activity dust scene, and high humidity environment scene. S4 includes: constructing the particle concentration state, environmental impact factor, device state factor and discharge risk factor in the fusion state quantity into a scene discrimination input vector;

[0083] in, The scene discrimination input vector is defined, and the parameters are consistent with the previous description.

[0084] The confidence scores for each scene are obtained based on the scene input vector, specifically as follows: Let the scene set S={1,2,3,4,5} correspond to: sleep silence, cooking fumes scene, cleaning dust scene, pet activity dust scene, and high humidity environment scene, respectively; For each scenario, a machine learning classification model (which can be generated by logistic regression / MLP / tree models, etc.) is used based on... Calculate scene score The original confidence level is obtained by softmax:

[0085] In the formula, The original confidence score of scene s output by the classification model satisfies and , It is an exponential function. Score the scene. , Let be the weight vector corresponding to scene s. The bias corresponding to scene s The summation process involves calculating the scene score corresponding to the scene indicated by the scene index r, where r iterates through all scenes in the scene set S. When r = s, and This indicates the score for the same scenario.

[0086] To avoid the difficulty of calculating the candidate set using natural language, a candidate indicator quantity is defined for each scene. , This indicates that scene s has been added to the scene candidate set. If the statement indicates that the member was not included, then the portfolio confidence level is: .

[0087] Threshold and interval symbols appearing in the definition rules: Noise threshold: Wind speed threshold: Rate of change threshold: Oil fume threshold: VOCs threshold: Percentage threshold: Fluctuation threshold: Humidity threshold: Drift threshold: ; Low to medium wind speed range: .

[0088] When the noise intensity is lower than the preset noise threshold And the wind speed is lower than the preset wind speed threshold. Furthermore, the rate of change in particle concentration is lower than a preset rate of change threshold. When necessary, add the sleep silence scene to the scene candidate set; When the concentration of cooking fumes and VOCs are both higher than the corresponding preset concentration threshold ( and When the particle concentration shows an upward trend within a preset time, the cooking fume scene will be added to the scene candidate set. When the proportion of PM10 relative to PM2.5 is higher than the preset proportion threshold Furthermore, the intensity of the fluctuation in particle concentration is higher than the preset fluctuation threshold. At that time, the scene of cleaning up dust was added to the scene candidate set; the ratio of PM10 to PM2.5 was: , To prevent tiny constants with a denominator of zero.

[0089] When the particulate concentration fluctuates intermittently and the proportion of PM10 is higher than the preset proportion threshold. And the wind speed is within the preset low to medium wind speed range. At that time, add scenes of pet activities causing dust to the scene candidate set; When the humidity is higher than the preset humidity threshold Furthermore, the discharge risk factor increases or the baseline drift of the leakage current corresponding to the device state factor exceeds the preset drift threshold. At that time, high humidity environment scenes are added to the scene candidate set.

[0090] Based on the confidence level of each scenario, a scenario candidate set is selected: the scenario with the highest confidence level is selected as the scenario recognition result; when the maximum confidence level is lower than the preset confidence level threshold, the output is a mixed scenario, and the top two scenarios with the highest confidence level in the mixed scenario are recorded as concurrent scenario recognition results.

[0091] The formulas for maximum confidence and mixed-scenario criteria are as follows:

[0092] To identify the scene number, To obtain the index that maximizes the objective, For maximum confidence, if Then the output will be a mixed scene. Confidence threshold; for concurrent scenarios, the top two confidence levels are used. .

[0093] S5. Based on the fusion state quantity and the scene recognition result, calculate the control quantity of the electrostatic dust removal device according to the preset control criterion; wherein the control quantity includes high voltage drive parameters and fan operating parameters, and the preset control criterion is used to weigh dust removal efficiency, energy consumption, noise constraints and discharge risk constraints, and to match the control quantity with the scene recognition result. S5 includes: constructing a control objective function based on the fused state variables, and setting target weights and constraint thresholds according to the scene recognition results; wherein the control objective function simultaneously includes a particle concentration reduction term for characterizing dust removal effect, a power consumption term for characterizing energy consumption, a noise penalty term for characterizing noise, and a risk penalty term for characterizing discharge risk; and setting constraint conditions such as noise intensity not exceeding a preset noise upper limit, discharge risk factor not exceeding a preset risk upper limit, and high voltage upper limit reduction when humidity is higher than a preset humidity threshold.

[0094] Define the control variable (decision variable) vector: , In the formula, Set the voltage for high voltage. Set the current for high voltage. The pulse frequency, Duty cycle, The rising edge slope parameter (which can be characterized by unit time increment). Polarity switching strategy encoding (e.g., 0 = no switching, 1 = alternating polarity, 2 = short-term reverse bias, etc.). Fan speed setting value.

[0095] Define electrostatic high voltage power consumption prediction:

[0096] For high voltage power consumption, This is the actual high-voltage output voltage. This is the actual high-voltage output current.

[0097] The power consumption of the fan can be expressed as the fan power or an approximation: , For wind turbine power consumption, The power consumption coefficient of the fan (related to the fan model specification). This refers to the fan speed.

[0098] The term "decrease in particle concentration" is: ; The total decrease in concentration, This is a weighting factor for the decrease in PM10 relative to the decrease in PM2.5 (used to combine the changes in the two particle sizes into a single comprehensive indicator). This represents the predicted / estimated value of PM2.5 concentration in the (k+1)th control period. This represents the predicted / estimated value of PM10 concentration for the (k+1)th control period.

[0099] Noise penalty items:

[0100] The candidate noise prediction value can be mapped from the fan speed. ), The upper limit of noise, For truncation operators.

[0101] The same applies to risk penalty items:

[0102] Risk upper limit threshold Predicted risk value for candidates.

[0103] The formula for reducing the high-pressure upper limit constraint triggered by humidity is:

[0104] The upper limit of permissible high pressure under humidity conditions. The upper limit of the benchmark high voltage, Humidity derating factor, Humidity threshold.

[0105] The control objective function is then:

[0106] For the target weight, The cost of the objective function.

[0107] The candidate control quantity set includes several combinations of high-voltage drive parameters and fan operating parameters; wherein the high-voltage drive parameters include set values ​​of high-voltage output voltage or high-voltage output current, pulse frequency, duty cycle, rise edge slope, and polarity switching strategy, and the fan operating parameters include fan speed set values. A candidate control quantity set is generated within a preset control cycle, and the candidate control quantity set is represented as follows:

[0108] In the formula, Candidate control variable set Number of candidate groups Let m be the vector of the candidate control quantity.

[0109] For each set of candidate control variables, the particle concentration change, power consumption, noise intensity, and discharge risk change in the next control cycle are predicted based on the fused state variables, and the corresponding comprehensive cost value is calculated; wherein the comprehensive cost value is calculated by the control objective function, and an infeasibility mark or additional penalty term is applied to candidate control variables that violate the constraints.

[0110] If candidate control quantity If any hard constraint is violated, then the condition is marked as infeasible.

[0111] Feasibility flag (1 for feasible, 0 for infeasible), ∧ for logical AND, and other symbols as defined above.

[0112] The group with the lowest comprehensive cost value is selected from the candidate control quantities that have never been marked as infeasible and output to the electrostatic dust removal device so that the control quantity matches the scene recognition result.

[0113] When the scene recognition result is a sleep / silent scene, the noise penalty weight is increased and the fan speed setting value is reduced. At the same time, the high voltage output setting value is reduced, and the dust removal effect is maintained by increasing the pulse frequency and adjusting the duty cycle. When the scene recognition result is a cooking fume scene, the dust removal effect weight is increased and the fan speed setting value is increased. At the same time, the high voltage output setting value is increased.

[0114] S6. Output the control quantity to the electrostatic dust removal device to adjust the working state of the electrostatic field, and repeat S1 to S5 in subsequent control cycles to update the control quantity, thereby realizing dynamic closed-loop control of indoor electrostatic dust removal. S6 includes: at the beginning of each control cycle, decomposing the control quantity into a high-voltage drive command and a fan drive command, and outputting them to the high-voltage power supply module and the fan module respectively; wherein the high-voltage drive command is used to set the high-voltage output voltage or high-voltage output current, pulse frequency, duty cycle, rise edge slope and polarity switching strategy, and the fan drive command is used to set the fan speed.

[0115] Mapping representation from control variables to instructions:

[0116]

[0117] This is the high-voltage drive command vector. This is a wind turbine drive command.

[0118] When updating the high-voltage drive command, a soft-start process is performed, which includes progressively increasing the high-voltage output setpoint within a preset time. When a leakage current exceeding a preset leakage threshold or a discharge event characterized by a discharge event occurs, a protection strategy is triggered. The protection strategy includes one or more of the following: reducing the high-voltage output setpoint, reducing the duty cycle, reducing the rise edge slope, switching the polarity switching strategy, and reducing the fan speed.

[0119] Let the target value be set as The maximum climb step size is ,but:

[0120] This is the set voltage for the output in this cycle. The voltage was set in the previous cycle. The maximum climb step size, i.e., the maximum allowable increment in the cycle. Use the smaller value operator. Implement tiered boosting to avoid corona instability and discharge caused by instantaneous overshoot.

[0121] The leakage current threshold is denoted as Protection trigger conditions:

[0122] Whether protection is triggered, ∨ logical OR, leakage current threshold, Discharge event flag.

[0123] Example of actions after triggering (simultaneously reducing field strength, pulse energy, and airflow disturbance, quickly moving away from the discharge boundary):

[0124] ←: Assignment update; : Load reduction factor.

[0125] During the control cycle, updated particulate matter parameters, environmental parameters, and operating parameters are collected, and dust removal effect evaluation quantities are calculated. The dust removal effect evaluation quantities include the particulate matter concentration reduction rate and the dust removal amount per unit energy consumption. Based on the dust removal effect evaluation quantities, the candidate control quantity generation range or control objective function weight for the next control cycle is corrected.

[0126] Particulate matter concentration reduction rate (using PM2.5 as an example):

[0127] PM2.5 reduction rate.

[0128] Dust removal capacity per unit of energy consumption (combined PM2.5 + PM10):

[0129] The unit energy consumption for dust removal (the larger the value, the more energy-efficient), ε is used to prevent division by zero, the numerator is the total dust removal capacity, and the denominator is the periodic energy consumption.

[0130] Example of adaptive weighting (increases energy consumption weight if energy efficiency is low):

[0131] Limiting function, Weight adjustment step size Target energy efficiency reference value, Energy consumption weighting upper and lower limits.

[0132] In subsequent control cycles, steps S1 to S5 are repeated; the control variables are updated based on the latest fused state variables and scene recognition results to achieve dynamic closed-loop control of indoor electrostatic dust removal.

[0133] The closed-loop iteration is represented as:

[0134] Control policy mapping, Merge and update mappings, Fusion state Feature vector Scene recognition results.

[0135] When the change in the control quantity in an adjacent control cycle exceeds a preset change threshold, the control quantity is limited / gradualized to suppress sudden changes in the electrostatic field operating state caused by scene switching or transient fluctuations.

[0136] For any control component ,For example Let the maximum change step size be... ,but: .

[0137] This is the current (to be output / updated) value of the j-th control component in the k-th control cycle. The control component can be any control component such as high voltage setpoint, pulse frequency, duty cycle, rise edge slope, polarity switching related parameters, or fan speed. A saturation limiting function is used to restrict the change δ within an interval. Inside; : Original change; : Vector of the maximum allowable change amplitude for each control component.

[0138] The original change Limit the amplitude so that the change in control quantity between adjacent cycles does not exceed This enables the limiting / gradual output of control quantities, suppressing control abrupt changes caused by scene switching or transient fluctuations.

[0139] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic control method for electrostatic precipitators based on multi-source data fusion, characterized in that, include: S1. Acquire particulate matter parameters and environmental parameters of indoor air, and simultaneously acquire the operating parameters of the electrostatic precipitator; S2. Process the particulate matter parameters, environmental parameters, and operating parameters, and construct a set of feature quantities based on the processed data to characterize the indoor dust load state, environmental impact state, device dust accumulation state, and discharge risk state. S3. Perform multi-source data fusion on the set of features to obtain fused state quantities; wherein the fused state quantities include particle concentration state, environmental impact factor, device state factor, and discharge risk factor; S4. Based on the fused state variables, perform indoor scene recognition and output the scene recognition result; S5. Based on the fusion state quantity and the scene recognition result, calculate the control quantity of the electrostatic dust removal device according to the preset control criteria; S6. Output the control quantity to the electrostatic dust removal device to adjust the working state of the electrostatic field, and repeat S1 to S5 in subsequent control cycles to update the control quantity, thereby realizing dynamic closed-loop control of indoor electrostatic dust removal. Time-domain filtering is performed on the particulate matter parameters and the environmental parameters to obtain smooth data; noise reduction processing is performed on the leakage current signal in the operating parameters to obtain the baseline component and peak component of the leakage current. Based on the smoothed data, the baseline component of the leakage current, and the spike component, a set of feature quantities is constructed; The set of features includes: Characteristic quantities used to characterize indoor dust load status: PM2.5 concentration, PM10 concentration, PM2.5 concentration change rate, PM10 concentration change rate, and particulate matter concentration fluctuation intensity; Characteristic quantities used to characterize the state of environmental impact: temperature, humidity, oil fume concentration, VOCs concentration, wind speed, and the rate of change of the above environmental parameters; Characteristic quantities used to characterize the dust accumulation state of the device: wind speed attenuation, fan load change, and leakage current baseline drift. Characteristic quantities used to characterize discharge risk states: leakage current spike amplitude, leakage current spike count, leakage current spike energy, and discharge event count. S3 includes: The set of features is normalized, and corresponding confidence weights are generated based on the fluctuation intensity and saturation state of each feature. The fusion state is obtained based on the normalized feature quantities and their confidence weights, where the fusion state is updated by the fusion state quantity of the previous time step and the feature quantity of the current time step. S4 includes: The particle concentration state, environmental impact factor, device state factor, and discharge risk factor in the fusion state quantity are constructed as a scene discrimination input vector. The confidence level of each scene is obtained based on the scene input vector; Based on the confidence level of each scenario, a scenario candidate set is selected: the scenario with the highest confidence level is selected as the scenario recognition result; when the maximum confidence level is lower than the preset confidence level threshold, the output is a mixed scenario, and the top two scenarios with the highest confidence level in the mixed scenario are recorded as concurrent scenario recognition results. A hold time and a switching hysteresis threshold are set for the scene recognition result. The current scene recognition result is maintained within the hold time, and the scene recognition result is updated only when the confidence of a new scene continuously exceeds the switching hysteresis threshold. The candidate set of scenes is determined based on the following conditions, including: When the noise intensity is lower than the preset noise threshold, the wind speed is lower than the preset wind speed threshold, and the rate of change of particle concentration is lower than the preset rate of change threshold, the sleep silent scene is added to the scene candidate set. When the concentration of cooking fumes and VOCs are both higher than the corresponding preset concentration thresholds, and the particle concentration shows an upward trend within a preset time, the cooking fume scene is added to the scene candidate set. When the proportion of PM10 relative to PM2.5 is higher than the preset proportion threshold, and the fluctuation intensity of the particulate concentration is higher than the preset fluctuation threshold, the clean dust scene will be added to the scene candidate set. When the particulate concentration fluctuates intermittently and the proportion of PM10 is higher than the preset proportion threshold, and the wind speed is in the preset low to medium wind speed range, the pet activity dust scene will be added to the scene candidate set. When the humidity is higher than the preset humidity threshold, and the discharge risk factor increases or the leakage current baseline drift corresponding to the device state factor is greater than the preset drift threshold, the high humidity environment scenario is added to the scenario candidate set. S5 includes: A control objective function is constructed based on the fused state variables, and the objective weights and constraint thresholds are set according to the scene recognition results. Generate a set of candidate control variables within a preset control period; For each set of candidate control variables, the particle concentration change, power consumption, noise intensity and discharge risk change in the next control cycle are predicted based on the fused state variables, and the corresponding comprehensive cost value is calculated. The group with the lowest comprehensive cost value is selected from the candidate control quantities that have never been marked as infeasible and output to the electrostatic dust removal device so that the control quantity matches the scene recognition result.

2. The dynamic control method for electrostatic dust removal based on multi-source data fusion according to claim 1, characterized in that, The particulate matter parameters include PM2.5 and PM10; the environmental parameters include temperature, humidity, oil fume concentration, volatile organic compound (VOC) concentration, noise intensity, and wind speed; the operating parameters include high voltage output voltage or high voltage output current, leakage current, and discharge event information.

3. The dynamic control method for electrostatic dust removal based on multi-source data fusion according to claim 2, characterized in that, S6 includes: At the beginning of each control cycle, the control quantity is decomposed into high-voltage drive command and fan drive command, and output to the high-voltage power supply module and the fan module respectively. When updating the high-voltage drive command, a soft-start process is performed, which includes progressively increasing the high-voltage output setpoint within a preset time. When a leakage current exceeding a preset leakage threshold or a discharge event characterized by a discharge event occurs, a protection strategy is triggered. The protection strategy includes one or more of the following: reducing the high-voltage output setpoint, reducing the duty cycle, reducing the rise edge slope, switching the polarity switching strategy, and reducing the fan speed. During the control cycle, updated particulate matter parameters, environmental parameters, and operating parameters are collected, and the dust removal effect evaluation quantity is calculated. In subsequent control cycles, steps S1 to S5 are repeated to update the control variables based on the latest fused state variables and scene recognition results, so as to achieve dynamic closed-loop control of indoor electrostatic dust removal. When the change in the control quantity in an adjacent control cycle exceeds a preset change threshold, the control quantity is subjected to amplitude limiting or gradual change processing to suppress sudden changes in the electrostatic field working state caused by scene switching or transient fluctuations.

Images