An AIoT technology performs energy consumption dynamic adjustment method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI SHENGDI INFORMATION TECH CO LTD
- Filing Date
- 2026-05-01
- Publication Date
- 2026-06-09
AI Technical Summary
Existing AIoT technology cannot effectively coordinate smoke exhaust safety and energy consumption optimization in restaurant kitchens. Especially under the dynamic uncertainty of cooking conditions, there is a deviation between the fan control output and the real-time operating requirements, resulting in environmental safety risks or energy loss.
By performing independent regression prediction on multidimensional feature vectors, and combining the sensitivity to environmental parameter fluctuations and the rate of change over time, the weights and confidence levels of the predicted values are adaptively adjusted. Based on the prediction confidence level, the wind turbine frequency is dynamically corrected to generate precise control commands.
It achieves deep adaptation of fan frequency control to kitchen operating conditions, ensuring the coordinated implementation of environmental safety and energy consumption optimization, and improving the accuracy of energy consumption regulation and the stability of the control process.
Smart Images
Figure CN122170078A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of energy consumption management and wind turbine frequency conversion control technology, and in particular to a method for dynamic energy consumption adjustment using AIoT technology. Background Technology
[0002] Commercial kitchen fume purification fans are core energy-consuming equipment in catering establishments, and their operation and control performance directly determines the safety and compliance of the kitchen environment and the energy consumption costs of catering operations. Currently, the industry widely uses dynamic energy consumption adjustment solutions for fans based on AIoT technology. This type of solution collects kitchen environmental parameters, cooking condition data, and fan operating status information in real time through an AIoT sensing network. Relying on supporting control logic, it generates dynamic adjustment commands for the fan operating frequency, thereby matching and adapting the fan operating status to the kitchen's fume extraction needs. In stable, routine operating scenarios, it can achieve a certain degree of energy consumption optimization and is currently the mainstream technical approach for energy consumption management of catering kitchen fans.
[0003] However, in the actual operation of restaurant kitchens, the cooking conditions are highly dynamic and uncertain. The switching between multiple stoves, the frequent adjustment of cooking modes, and the sudden increase or decrease in the amount of instantaneous oil fumes occur frequently. When existing dynamic adjustment solutions are running continuously under such complex and fluctuating conditions, their control output is prone to deviating from the actual needs of the real-time conditions. When deviation occurs, either the instantaneous smoke exhaust demand cannot be matched in time, leading to the risk of excessive oil fume concentration and failure to meet environmental safety standards in the kitchen, or the fans are kept running at high load to avoid safety risks, resulting in continuous unnecessary energy consumption. It is impossible to achieve a coordinated and stable balance between smoke exhaust safety and precise energy consumption optimization. Summary of the Invention
[0004] This invention provides a method for dynamically adjusting energy consumption using AIoT technology to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides a method for dynamic energy consumption adjustment using AIoT technology, comprising: Perform independent regression predictions on multidimensional feature vectors to obtain a set of prediction values consisting of multiple independent prediction values; Simultaneously, the central tendency statistics and dispersion determination of multiple independent predicted values are performed to obtain the preliminary predicted value of wind turbine frequency and the dispersion measure, and the dispersion measure is mapped to the prediction confidence level. The correction range for the initial predicted value of the wind turbine frequency is determined based on the prediction confidence level. The initial predicted value of the wind turbine frequency is then corrected based on the correction range to obtain the wind turbine frequency control command.
[0006] Preferably, the step of performing independent regression predictions on the multidimensional feature vectors to obtain a set of predicted values consisting of multiple independent predicted values includes: Multiple regression predictions are performed on the multidimensional feature vector. Each regression prediction is applied to a different feature subset of the multidimensional feature vector, resulting in multiple independent predicted values, which constitute a set of predicted values. The feature dimensions selected for different regression predictions are not entirely the same.
[0007] Preferably, the step of performing multiple regression predictions on the multidimensional feature vector, with each regression prediction acting on a different feature subset of the multidimensional feature vector, includes: Divide multiple feature subsets in a multidimensional feature vector into a first feature subset and a second feature subset; Smoothing and denoising processing is performed on the feature values in the first feature subset to obtain the denoised first feature subset, and a regression prediction is performed based on the denoised first feature subset to obtain the first type of independent prediction value. Perform at least one regression prediction directly on the second feature subset to obtain the second type of independent predicted value; The predicted value set consists of the first type of independent predicted value and the second type of independent predicted value.
[0008] Preferably, dividing the multiple feature subsets in the multidimensional feature vector into a first feature subset and a second feature subset includes: Obtain the time-series fluctuation statistics of various environmental parameters in the kitchen environment within a preset historical observation period; Based on the time-series fluctuation statistics, determine the environmental parameter fluctuation sensitivity corresponding to each feature dimension in the multidimensional feature vector; Feature dimensions whose sensitivity to environmental parameter fluctuations is higher than a preset sensitivity threshold are classified into the first feature subset, and feature dimensions whose sensitivity to environmental parameter fluctuations is not higher than the preset sensitivity threshold are classified into the second feature subset.
[0009] Preferably, the step of performing central tendency statistics on multiple independent predicted values to obtain a preliminary predicted value for the wind turbine frequency includes: Obtain the temporal rate of change of key environmental parameters of the kitchen environment within the current time frame and adjacent historical window; When the time series change rate is lower than the preset stability threshold, the predicted values belonging to the first type of independent predicted values are assigned a higher weight than the predicted values belonging to the second type of independent predicted values, and a weighted average is performed to obtain the preliminary predicted value of the wind turbine frequency. When the time series change rate is not lower than the preset stability threshold, the predicted values belonging to the second type of independent predicted values are assigned a higher weight than the predicted values belonging to the first type of independent predicted values, and a weighted average is performed to obtain the preliminary predicted value of the wind turbine frequency.
[0010] Preferably, the step of determining the degree of dispersion of multiple independent predicted values to obtain a measure of dispersion includes: When the time series change rate is lower than the preset stability threshold, the dispersion measure of the predicted value set is calculated using the first statistical method. When the time series change rate is not lower than the preset stationarity threshold, the second statistical method is used to calculate the dispersion measure of the predicted value set; The first statistical method is less sensitive to extreme values than the second statistical method.
[0011] Preferably, mapping the degree of dispersion measure to the prediction confidence includes: An adaptive mapping coefficient is constructed based on the time series rate of change, wherein the adaptive mapping coefficient decreases monotonically as the time series rate of change increases; The prediction confidence is obtained by combining the discreteness measure with the adaptive mapping coefficients.
[0012] Preferably, determining the correction range for the initial predicted value of the wind turbine frequency based on the prediction confidence level includes: Obtain the frequency band identifier of the preliminary predicted wind turbine frequency value; The adjustment sensitivity coefficient is determined based on the frequency band identifier, wherein the adjustment sensitivity coefficient in the low frequency operating band is lower than that in the high frequency operating band. The basic correction magnitude is calculated based on the deviation between the predicted confidence level and the preset confidence threshold. The basic correction magnitude increases as the deviation increases. The correction magnitude is obtained by multiplying the basic correction magnitude with the adjustment sensitivity coefficient.
[0013] Preferably, before correcting the initial predicted value of the wind turbine frequency based on the correction magnitude, the method further includes: Obtain real-time values of key environmental parameters in the kitchen environment; When the real-time value of a key environmental parameter triggers a preset safety intervention condition, a mandatory safety correction magnitude is generated, and the correction magnitude determined based on the prediction confidence level is replaced by the mandatory safety correction magnitude.
[0014] Preferably, the step of correcting the initial predicted value of the wind turbine frequency based on the correction magnitude to obtain the wind turbine frequency control command includes: Apply the correction range or forced safety correction range to the initial predicted value of the wind turbine frequency to generate the frequency value to be output. Obtain historical frequency control commands issued in the previous control cycle; Calculate the frequency change between the output frequency value and the historical frequency control commands; When the frequency change exceeds the preset adjustment rate limit, the output frequency value is truncated to limit the frequency change within the preset adjustment rate limit and generate a fan frequency control command.
[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By mapping the dispersion measure of multiple sets of independent predicted values to the prediction confidence level, and by adaptively determining the correction range of the initial predicted value of the fan frequency based on the prediction confidence level to complete the directional correction, it is possible to achieve deep adaptation between the fan frequency control output and the dynamic changes of the kitchen operating conditions, accurately match the kitchen smoke exhaust needs under different operating conditions, and achieve refined management and control of fan operation energy consumption while ensuring the safety of the kitchen environment, thus achieving the synergistic implementation of environmental safety assurance and energy consumption optimization.
[0016] 2. By differentiating and predicting the feature subsets of the multidimensional feature vectors according to their fluctuation characteristics, and by adaptively adjusting the weighting of different predicted values in conjunction with the time-series change rate of kitchen environmental parameters, the bidirectional adaptability of the fan frequency prediction value to both stable and transient fluctuating conditions in the kitchen can be further enhanced. This can avoid unnecessary energy consumption losses caused by frequent frequency fluctuations under stable conditions, and ensure the smoke exhaust response speed under transient fluctuating conditions, thereby further improving the accuracy of fan energy consumption regulation and the stability of the control process. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating a method for dynamically adjusting energy consumption using AIoT technology, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the feature subset partitioning method in this invention. Detailed Implementation
[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0020] Reference Figures 1-2 As shown, a method for dynamic energy consumption adjustment using AIoT technology includes: Step 1. Perform independent regression predictions on the multidimensional feature vectors to obtain a set of prediction values consisting of multiple independent prediction values.
[0021] In this embodiment, the multidimensional feature vector covers all real-time collected parameters of the kitchen environment and equipment operation, and all parameters are synchronously collected and uploaded through AIoT sensing layer devices.
[0022] When classifying the feature dimensions, specifically, we first statistically analyze the temporal fluctuations of each environmental parameter within a preset historical observation period, and calculate the coefficient of variation of each parameter, which is the ratio of the temporal standard deviation to the temporal mean of the parameter in the historical period. This quantifies the sensitivity of the parameter to the fluctuations of cooking actions. The larger the ratio, the greater the fluctuation of the parameter due to the influence of cooking actions.
[0023] Then, feature dimensions with fluctuation sensitivity higher than the calibration threshold are assigned to the first feature subset, and those with sensitivity not higher than the calibration threshold are assigned to the second feature subset.
[0024] It should be noted that the above-mentioned sensitive thresholds were obtained by: collecting fluctuation data of various environmental parameters based on the need to distinguish between perturbation-type and basic-type parameters, combined with the actual cooking scenario in the kitchen, and completing on-site calibration through testing and screening.
[0025] The first feature subset consists of perturbation-type environmental parameters that frequently undergo transient pulse changes due to cooking actions. These parameters include real-time oil fume concentration, number of stoves lit, cooking level, and real-time PM2.5 value. These parameters undergo transient pulse changes due to cooking start-up and shutdown, stir-frying, and other actions, and are the core source of disturbance in the kitchen environment.
[0026] The second feature subset consists of basic environmental parameters that exhibit a continuous and gradual change over time. These parameters include kitchen temperature and humidity, average CO concentration, fan operating time, and outdoor wind pressure. These parameters exhibit a continuous and gradual change over a short period of time without drastic transient fluctuations, and are stable basic features of the kitchen environment.
[0027] In summary, this scheme achieves accurate feature subset division through fluctuation sensitivity, decouples transient perturbation features from slowly varying base features, avoids transient noise interfering with base trend judgment during subsequent prediction, and does not lose the responsiveness to transient perturbations due to excessive smoothing.
[0028] As a preferred implementation, each regression prediction is applied to different feature subsets of the multidimensional feature vector, and the feature dimensions selected in different predictions are not exactly the same, ensuring the independence of each prediction result.
[0029] In practice, the feature values of the first feature subset are first subjected to sliding median filtering for smoothing and noise reduction. During the smoothing process, the filtering operation needs to be carried out in combination with the preset sampling period and the preset window range. Finally, the noise-reduced feature values of the first feature subset are obtained through this smoothing process.
[0030] It should be noted that the above-mentioned preset sampling period, preset window range, and specific window values can be obtained as follows: the preset sampling period is obtained by calibrating the sensor through on-site working condition tests based on the transient disturbance response requirements of kitchen cooking and the sensor acquisition accuracy, combined with the sensor sampling performance; the preset window range is obtained by calibrating the sensor through multiple sets of window tests based on the requirements of noise filtering and parameter trend retention; and the specific preset window value is obtained by further testing and calibration, combining the above-mentioned sampling period and window range.
[0031] Then, based on the first feature subset after noise suppression, a lightweight gradient boosting tree is used to train the first regression prediction model, which outputs the first type of independent prediction value.
[0032] Furthermore, the processing of the second feature subset is directly based on the full features and partial features. For example, two independent regression prediction models are trained, using linear regression and LSTM time series prediction models respectively, which output two independent second-class prediction values. The linear regression model inputs the full features of the second feature subset, while the LSTM model only inputs the temperature, humidity, and outdoor wind pressure features from the second feature subset, ensuring that the input feature dimensions of the two predictions do not completely overlap, further enhancing the independence of the prediction results.
[0033] Finally, the predicted value set is composed of one first-class independent predicted value and two second-class independent predicted values.
[0034] In summary, this approach obtains multi-perspective prediction results through multiple independent regression predictions based on different feature subsets and model structures. Each prediction result is independent and there is no parameter sharing, which avoids prediction inaccuracies caused by overfitting of a single model or feature bias, and greatly improves the robustness of the prediction results. At the same time, the differentiated smoothing process takes into account both the noise resistance requirements of perturbation parameters and the trend preservation requirements of basis parameters.
[0035] Step 2: Perform central tendency statistics and determine dispersion for multiple independent predicted values to obtain preliminary predicted values of wind turbine frequency and dispersion measure, and map the dispersion measure to prediction confidence.
[0036] This step adaptively adjusts the statistical strategy based on the real-time changes in the kitchen environment, while quantifying the uncertainty of the prediction results to provide a quantitative basis for subsequent corrections.
[0037] Specifically, adaptive central tendency statistics are performed to obtain preliminary predicted values for wind turbine frequencies.
[0038] In practice, the concentration of cooking fumes is selected as a key environmental parameter of the kitchen environment. First, the temporal change rate of the concentration of cooking fumes in the adjacent preset historical window at the current moment is calculated, that is, the relative change of the concentration of cooking fumes at the current moment and in front of the preset historical window, so as to determine the stable state of the kitchen environment.
[0039] When the time series change rate is lower than the calibrated threshold, it is determined that the kitchen environment is in a stable state and there is no sudden cooking action. At this time, the first type of independent prediction value after smoothing and noise reduction has stronger anti-interference ability and higher reliability. Therefore, the first type of independent prediction value is given a higher weight, and the second type of independent prediction value is given a lower weight. The preliminary prediction value of the fan frequency is obtained by weighted averaging.
[0040] When the time series change rate is not lower than the calibrated threshold, it is determined that the kitchen environment is in a transient change state and there is a sudden cooking action. At this time, the parameters of the first feature subset fluctuate violently, and smoothing is prone to losing transient response information. The second type of independent prediction value is based on the slowly changing basis parameters, which can better reflect the overall change trend of the environment. Therefore, the second type of independent prediction value is given a higher weight, and the first type of independent prediction value is given a lower weight. The preliminary prediction value of the fan frequency is obtained by weighted averaging.
[0041] It should be noted that the stability threshold used to judge the environmental state can be obtained by collecting time-series data on the changes in oil fume concentration under different scenarios, based on the need to distinguish between stable and transient cooking scenarios and the actual cooking intensity in the kitchen, and then completing on-site calibration through testing and screening.
[0042] In summary, this solution adaptively adjusts the weight of different prediction values based on the real-time dynamic changes in the kitchen environment. Under stable conditions, prediction results with strong noise resistance are given priority to avoid energy waste caused by frequent fluctuations in fan frequency. Under transient changes, prediction results with strong trend responsiveness are given priority to ensure smoke extraction effect during sudden cooking.
[0043] Furthermore, the degree of dispersion of multiple independent predicted values is determined to obtain a measure of dispersion.
[0044] The degree of dispersion measure is used to quantify the degree of disagreement between multiple independent predictions. The greater the disagreement, the higher the uncertainty of the prediction result.
[0045] This embodiment can adaptively select the statistical method according to the environmental state. Specifically, when the time series change rate is lower than the calibration threshold in a stable state, the interquartile range can be used to calculate the degree of dispersion. This statistical method has extremely low sensitivity to extreme values and can filter out abnormal jumps in individual predicted values in a stable state, accurately reflecting the inherent dispersion of the prediction results.
[0046] When the rate of change of time series is not lower than the calibrated threshold, the standard deviation can be used to calculate the degree of dispersion. This statistical method is more sensitive to extreme values and can sensitively capture the divergence of predicted values under transient changes, and reflect the uncertainty changes of the prediction results in a timely manner.
[0047] In summary, the differentiated discreteness statistical method ensures both the stability of discreteness calculation under stationary conditions and avoids interference from extreme values, while also ensuring the sensitivity of discreteness calculation under transient conditions. It can quickly capture changes in prediction uncertainty and provide accurate input for subsequent confidence mapping.
[0048] Furthermore, the measure of dispersion is mapped to prediction confidence.
[0049] Understandably, prediction confidence can quantify the reliability of the prediction result; the higher the confidence level, the more reliable the prediction result.
[0050] In this embodiment, an adaptive mapping coefficient can be constructed first. This coefficient decreases monotonically as the rate of change of time increases. The more drastic the transient changes in the kitchen environment, the smaller the mapping coefficient becomes, thus avoiding excessive reduction of confidence due to large discrepancies in the predicted values.
[0051] Then, the dispersion measure and the adaptive mapping coefficient are combined to obtain the prediction confidence.
[0052] The prediction confidence level can be calculated using the following formula:
[0053] in, To predict confidence levels, the reliability of the initial wind turbine frequency prediction can be quantified, providing a core quantitative basis for subsequent correction calculations.
[0054] As an adaptive mapping coefficient, it can be constructed based on key parameters of the kitchen environment, such as the time-series change rate of oil fume concentration. The time-series change rate decreases monotonically, requiring no additional on-site calibration, and is dynamically and adaptively generated by the time-series change rate.
[0055] As a measure of dispersion, it can quantify the degree of disagreement among multiple independent predicted values in the predicted value set. The greater the disagreement, the higher the prediction uncertainty. It can be obtained through adaptive statistics based on the state of the kitchen environment.
[0056] The preset maximum value of dispersion can standardize the confidence calculation range, avoid extreme prediction discrepancies that could lead to distortion in the confidence calculation, and ensure that the confidence value is within a reasonable range of 0-1. Based on the prediction discrepancy coverage requirements and the confidence calculation accuracy requirements, combined with the range of wind turbine operating parameters, prediction discrepancy data under different operating conditions are collected and calibrated on-site through testing and debugging.
[0057] It is a natural exponential function that can realize the nonlinear mapping from the degree of dispersion to the confidence level, transforming the degree of disagreement into a confidence level in the 0-1 interval, and ensuring that the confidence level monotonically decreases as the degree of disagreement increases; it is derived from a conventional nonlinear mapping function in the field of mathematics, and is selected in combination with the confidence quantification requirements of this method.
[0058] The above formula clearly shows that the greater the degree of dispersion, the lower the prediction confidence; and for the same degree of dispersion, the more drastic the transient changes in the environment, the higher the prediction confidence.
[0059] Overall, this scheme deeply couples the dynamic characteristics of environmental changes with the prediction dispersion through adaptive mapping coefficients, solving the problem that fixed mapping rules cannot adapt to the changing working conditions in the kitchen. Under stable conditions, the prediction confidence is strictly quantified to avoid erroneous adjustments caused by low confidence predictions. Under transient conditions, the confidence decay range is appropriately relaxed to ensure the response speed under sudden working conditions and to prevent excessive suppression of adjustment actions due to prediction discrepancies.
[0060] Step 3: Determine the correction range for the initial predicted value of the wind turbine frequency based on the prediction confidence level, and correct the initial predicted value of the wind turbine frequency based on the correction range to obtain the wind turbine frequency control command.
[0061] This step achieves adaptive correction based on confidence level, while setting safety margins and smoothing limits to balance energy consumption optimization, control stability, and environmental safety.
[0062] Specifically, the correction range for the initial predicted value of the wind turbine frequency is first determined based on the prediction confidence level.
[0063] In this embodiment, the correction magnitude is negatively correlated with the prediction confidence level, and the sensitivity is adjusted according to the differences in the operating frequency range of the fan to avoid excessive correction in the low frequency range, which would lead to insufficient smoke extraction.
[0064] Specifically, the frequency range of the fans is first divided and the adjustment sensitivity coefficient is determined. In the low frequency range, the fans have weak smoke exhaust capacity, and reducing the adjustment sensitivity can avoid insufficient smoke exhaust caused by large corrections. In the high frequency range, the fans are under high load, and increasing the adjustment sensitivity can maximize energy consumption optimization.
[0065] Next, the basic correction magnitude is calculated. When the prediction confidence is not lower than the calibration threshold, the prediction result is reliable and the basic correction magnitude is 0, so there is no need to correct the initial prediction value. When the prediction confidence is lower than the calibration threshold, the larger the deviation between the confidence and the reliability threshold, the larger the basic correction magnitude.
[0066] Finally, the basic correction range is multiplied by the adjustment sensitivity coefficient to obtain the final correction range. The direction of the correction range is determined by the deviation between the current oil fume concentration and the target concentration: when the oil fume concentration is higher than the target value, the correction range is positive, and the fan frequency is increased; when the oil fume concentration is lower than the target value, the correction range is negative, and the fan frequency is decreased.
[0067] It should be noted that the above-mentioned adjustment sensitivity coefficient and confidence threshold can be obtained as follows: the adjustment sensitivity coefficient is obtained through on-site commissioning and calibration based on the smoke exhaust safety and energy consumption optimization requirements of each frequency band and the smoke exhaust capacity parameters of the fan; the confidence threshold is obtained through testing and calibration based on the prediction accuracy and control stability requirements and historical prediction error data.
[0068] In summary, this solution balances smoke exhaust safety in the low-frequency range and energy consumption optimization in the high-frequency range by setting differentiated sensitivity in different frequency bands. Based on the calculation of the correction magnitude of confidence level, it realizes the closed-loop linkage between prediction confidence and control action, avoiding the fan misadjustment caused by directly using low-confidence prediction results for control, and greatly improving the stability of control.
[0069] Furthermore, as a preferred implementation of this solution, before correcting the initial predicted value of the fan frequency based on the correction range, a pre-approval safety intervention can be performed to prioritize the safety of the kitchen environment.
[0070] Specifically, the system acquires real-time values of oil fume concentration and CO concentration in the kitchen environment. As soon as any safety condition is triggered, a mandatory safety correction range is immediately generated and replaced with the correction range calculated based on confidence level, prioritizing the safety of the kitchen environment.
[0071] It should be noted that the safety thresholds corresponding to the above safety conditions can be obtained by collecting concentration data under safe boundary conditions based on industry safety standards and the performance of purification equipment, combined with the size of the kitchen space, and calibrating through on-site safety testing.
[0072] Overall, this implementation method resolves the core conflict between energy consumption optimization and environmental safety by setting up hard safety safeguards. When the concentration of kitchen fumes or harmful gases exceeds the standard, mandatory intervention is immediately triggered to avoid insufficient smoke exhaust due to prediction correction, thus ensuring the health of kitchen operators and environmental compliance.
[0073] Finally, the initial predicted value of the wind turbine frequency is corrected based on the correction range to obtain the wind turbine frequency control command.
[0074] Specifically, the correction range or forced safety correction range is applied to the preliminary predicted value of the wind turbine frequency to generate the output frequency value; then the historical frequency control command issued in the previous control cycle is obtained; and the frequency change between the output frequency value and the historical command is calculated.
[0075] When the frequency change exceeds the preset limit, the output frequency value is truncated to limit the frequency change within the calibration range and generate the final fan frequency control command; when the frequency change does not exceed the limit, the output frequency value is directly used as the final control command.
[0076] It should be noted that the above-mentioned frequency change rate adjustment limit can be obtained through on-site calibration based on the equipment operation stability and control response speed requirements, combined with the hardware parameters of the fan and frequency converter, through multiple rounds of equipment operation tests.
[0077] In summary, this solution achieves smooth adjustment of the fan frequency by regulating the rate limit, avoiding equipment damage and noise pollution caused by large frequency jumps, extending the service life of the fan, and ensuring the stability of the control process, avoiding additional energy consumption caused by frequent start-stop and large fluctuations.
[0078] In the several embodiments provided by this invention, it should be understood that the disclosed method can be implemented in other ways.
[0079] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0080] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, and technology that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.
[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for dynamic energy consumption adjustment using AIoT technology, characterized in that, The method includes: Perform independent regression predictions on multidimensional feature vectors to obtain a set of prediction values consisting of multiple independent prediction values; Simultaneously, the central tendency statistics and dispersion determination of multiple independent predicted values are performed to obtain the preliminary predicted value of wind turbine frequency and the dispersion measure, and the dispersion measure is mapped to the prediction confidence level. The correction range for the initial predicted value of the wind turbine frequency is determined based on the prediction confidence level. The initial predicted value of the wind turbine frequency is then corrected based on the correction range to obtain the wind turbine frequency control command.
2. The method for dynamic energy consumption adjustment using AIoT technology as described in claim 1, characterized in that, The process of performing independent regression predictions on multidimensional feature vectors yields a set of predicted values consisting of multiple independent predicted values, including: Multiple regression predictions are performed on the multidimensional feature vector. Each regression prediction is applied to a different feature subset of the multidimensional feature vector, resulting in multiple independent predicted values, which constitute a set of predicted values. The feature dimensions selected for different regression predictions are not entirely the same.
3. The method for dynamic energy consumption adjustment using AIoT technology as described in claim 2, characterized in that, The process of performing multiple regression predictions on a multidimensional feature vector, where each regression prediction applies to a different subset of features of the multidimensional feature vector, includes: Divide multiple feature subsets in a multidimensional feature vector into a first feature subset and a second feature subset; Smoothing and denoising processing is performed on the feature values in the first feature subset to obtain the denoised first feature subset, and a regression prediction is performed based on the denoised first feature subset to obtain the first type of independent prediction value. Perform at least one regression prediction directly on the second feature subset to obtain the second type of independent predicted value; The predicted value set consists of the first type of independent predicted value and the second type of independent predicted value.
4. The method for dynamic energy consumption adjustment using AIoT technology as described in claim 3, characterized in that, The step of dividing multiple feature subsets in a multidimensional feature vector into a first feature subset and a second feature subset includes: Obtain the time-series fluctuation statistics of various environmental parameters in the kitchen environment within a preset historical observation period; Based on the time-series fluctuation statistics, determine the environmental parameter fluctuation sensitivity corresponding to each feature dimension in the multidimensional feature vector; Feature dimensions whose sensitivity to environmental parameter fluctuations is higher than a preset sensitivity threshold are classified into the first feature subset, and feature dimensions whose sensitivity to environmental parameter fluctuations is not higher than the preset sensitivity threshold are classified into the second feature subset.
5. The method for dynamic energy consumption adjustment using AIoT technology as described in claim 4, characterized in that, The process of performing central tendency statistics on multiple independent predicted values to obtain preliminary predicted values for wind turbine frequencies includes: Obtain the temporal rate of change of key environmental parameters of the kitchen environment within the current time frame and adjacent historical window; When the time series change rate is lower than the preset stability threshold, the predicted values belonging to the first type of independent predicted values are assigned a higher weight than the predicted values belonging to the second type of independent predicted values, and a weighted average is performed to obtain the preliminary predicted value of the wind turbine frequency. When the time series change rate is not lower than the preset stability threshold, the predicted values belonging to the second type of independent predicted values are assigned a higher weight than the predicted values belonging to the first type of independent predicted values, and a weighted average is performed to obtain the preliminary predicted value of the wind turbine frequency.
6. The method for dynamic energy consumption adjustment using AIoT technology as described in claim 5, characterized in that, The determination of the degree of dispersion of multiple independent predicted values to obtain a measure of dispersion includes: When the time series change rate is lower than the preset stability threshold, the dispersion measure of the predicted value set is calculated using the first statistical method. When the time series change rate is not lower than the preset stationarity threshold, the second statistical method is used to calculate the dispersion measure of the predicted value set; The first statistical method is less sensitive to extreme values than the second statistical method.
7. The method for dynamic energy consumption adjustment using AIoT technology as described in claim 6, characterized in that, The mapping of the degree of dispersion measure to the prediction confidence includes: An adaptive mapping coefficient is constructed based on the time series rate of change, wherein the adaptive mapping coefficient decreases monotonically as the time series rate of change increases; The prediction confidence is obtained by combining the discreteness measure with the adaptive mapping coefficients.
8. A method for dynamic energy consumption adjustment using AIoT technology as described in claim 7, characterized in that, The determination of the correction range for the initial predicted wind turbine frequency based on the prediction confidence level includes: Obtain the frequency band identifier of the preliminary predicted wind turbine frequency value; The adjustment sensitivity coefficient is determined based on the frequency band identifier, wherein the adjustment sensitivity coefficient in the low frequency operating band is lower than that in the high frequency operating band. The basic correction magnitude is calculated based on the deviation between the predicted confidence level and the preset confidence threshold. The basic correction magnitude increases as the deviation increases. The correction magnitude is obtained by multiplying the basic correction magnitude with the adjustment sensitivity coefficient.
9. A method for dynamic energy consumption adjustment using AIoT technology as described in claim 1, characterized in that, Before correcting the initial predicted value of the wind turbine frequency based on the correction magnitude, the method further includes: Obtain real-time values of key environmental parameters in the kitchen environment; When the real-time value of a key environmental parameter triggers a preset safety intervention condition, a mandatory safety correction magnitude is generated, and the correction magnitude determined based on the prediction confidence level is replaced by the mandatory safety correction magnitude.
10. A method for dynamic energy consumption adjustment using AIoT technology as described in claim 9, characterized in that, The step of correcting the initial predicted value of the wind turbine frequency based on the correction magnitude to obtain the wind turbine frequency control command includes: Apply the correction range or forced safety correction range to the initial predicted value of the wind turbine frequency to generate the frequency value to be output. Obtain historical frequency control commands issued in the previous control cycle; Calculate the frequency change between the output frequency value and the historical frequency control commands; When the frequency change exceeds the preset adjustment rate limit, the output frequency value is truncated to limit the frequency change within the preset adjustment rate limit and generate a fan frequency control command.