Adaptive learning-based personalized evaluation method for residential vibration comfort

Abstract

The application discloses a kind of self-adapting learning's residential vibration comfort degree personalized evaluation method, belong to building vibration monitoring and artificial intelligence evaluation field. Including: collection residential indoor vibration acceleration time history data;Obtain 20 1 / 3 octave band time history data by second-order Butterworth band-pass filtering;Based on edge computing, extract Z vibration level maximum, weighted root mean square value, human vibration perception ratio, vertical quartic vibration dose value and annoyance rate multidimensional index;With user subjective comfort feedback as supervision label, two kinds of misjudgment rates, namely, too loose and too strict, are respectively counted, and the two demarcation thresholds of comfort / general and general / discomfort are dynamically adjusted accordingly;Finally output personalized evaluation result.The application realizes the convergence of evaluation threshold to user's real tolerance through self-adapting learning, has self-evolution ability, and can realize low delay, thousand people and thousand faces of residential vibration comfort degree personalized early warning.

Description

Technical Field

[0001] This invention relates to the field of real-time monitoring and artificial intelligence assessment of building vibration comfort, and particularly to a personalized assessment method for residential vibration comfort that integrates multi-dimensional indicators and adaptive learning. Background Technology

[0002] Living comfort is one of the core indicators of modern building performance. Floor vibration issues are becoming increasingly prominent due to the combined effects of lightweight building structures, large spans, and internal and external vibration sources (such as traffic, equipment, and human activity). Excessive vibration not only reduces the living experience but may also cause psychological discomfort and health risks for users. Therefore, in the context of intelligent building development, establishing an accurate and dynamic vibration comfort evaluation system is of great significance for ensuring the quality of human habitation.

[0003] Current vibration comfort assessments largely rely on fixed thresholds specified in international or national standards such as ISO and GB, including ISO 2631-1:1997, ANSI S3.29 (1983), GB 10070-1988, GB / T 50355-2018, BS 6472-1:2008, and DIN 4150-2:1999. These thresholds are based on population statistics and cannot accommodate individual differences in perception due to age, health status, and psychological expectations. Technically, traditional assessments primarily rely on numerical simulation and short-term field testing: the former is limited by idealized model assumptions and the singularity of the excitation source, making it difficult to apply to the dynamic assessment of buildings in service; the latter suffers from fragmented monitoring, high costs, and difficulty in sustainability. Commercial monitoring systems that have emerged in recent years mostly adopt a centralized architecture, with raw vibration data uploaded to the cloud for processing throughout the process, resulting in high network bandwidth pressure, poor real-time performance, and significant consumption of computing resources.

[0004] The rise of edge computing and adaptive learning technologies has provided new solutions to the above problems. Edge computing, by processing data near the data source, can greatly reduce the amount of data transmission and improve the real-time response of the system; adaptive algorithms can dynamically adjust model parameters based on user feedback, enabling personalized evaluation criteria.

[0005] However, research combining the two and applying them to continuous monitoring and intelligent evaluation of residential vibration comfort is still relatively rare. Summary of the Invention

[0006] To address the aforementioned technical issues, this invention proposes an adaptive learning-based personalized assessment method for residential vibration comfort, which can monitor residential vibration comfort in real time and provide real-time personalized early warnings for different residents, with a delay of no more than 3ms.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] An adaptive learning-based personalized assessment method for residential vibration comfort includes the following steps:

[0009] Step 1: Data Acquisition: Utilize the accelerometer built into the vibration monitoring system to measure the vibration acceleration inside the building / residential space. ;

[0010] Step 2, Filtering: Filter the collected vibration acceleration... Twenty sets of time history data corresponding to each frequency band were obtained by second-order Butterworth bandpass filtering. ;

[0011] Step 3: Index Extraction: Using the raw vibration data through edge computing nodes, calculate multiple evaluation indices. With a time period of 1 second, calculate the following vibration comfort evaluation indices for each frequency band time history data after filtering: maximum Z-level, weighted root mean square value, human vibration perception ratio, vertical fourth power vibration dose value, and annoyance rate.

[0012] Step 4: Adaptive Threshold Update: Input the vibration comfort evaluation index obtained in Step 3 into the adaptive learning model based on misjudgment statistics. The initial threshold is set according to the current national standards. Collect user subjective comfort feedback through mobile terminals and use this feedback as a supervision label. Calculate the misjudgment rates for two categories: too lenient and too strict. Based on the difference between the two misjudgment rates, dynamically adjust the comfort / normal dividing threshold b1 and the normal / uncomfortable dividing threshold b2.

[0013] Step 5: Personalized assessment output: Within a specified time period, the updated comfort / normal threshold b1 and normal / uncomfortable threshold b2 are converged and trained to the current user's actual tolerance. Different vibration level thresholds are used for daytime and nighttime, and personalized assessment results are output. The daytime period is from 6:00 to 18:00, and the rest of the time is nighttime.

[0014] Furthermore, in step one, the sampling frequency is set to 200Hz based on the human body's sensitive frequency band of 1~80Hz and the Nyquist theorem.

[0015] Furthermore, in step three, the maximum value of the Z-level vibration... The calculation method is as follows: for the first The maximum value of the Z-level is calculated using the impact vibration algorithm based on the time history data of each center frequency.

[0016] ;

[0017] In the formula: For the first The time histories corresponding to each 1 / 3 octave band are obtained by bandpass filtering and spectral transformation of the vibration data. This is the time integration constant, usually taken as 1 second; The effective value of the vibration reference acceleration. ; For 1 / 3 octave band Z-weighting factor corresponding to each center frequency; It is the Z-level vibration.

[0018] Furthermore, in step three, the method for calculating the weighted root mean square (RMS) value is as follows:

[0019] ;

[0020] In the formula: The acceleration value at a certain time t, in m / s². 2 T represents the test duration, in seconds.

[0021] The Human Vibration Perception Ratio (WODL) is defined as the percentage of samples that exceed the vibration perception threshold.

[0022] Furthermore, in step three, the vertical fourth-power vibration dose value The calculation method is as follows:

[0023] ;

[0024] In the formula, This is the vertical fourth power dose value, in m / s. 1.75 ; W for basic frequency weighting k Instantaneous vertical acceleration, in m / s² 2 T represents the duration, which is usually 1 second.

[0025] Furthermore, in step three, the annoyance rate is calculated using a fuzzy membership function based on frequency-weighted root mean square acceleration.

[0026] ;

[0027] In the formula, Let i be the annoyance rate at the i-th vibration intensity; The number of people who exhibit the j-th subjective response at the i-th vibration intensity; Let be the membership degree of the j-th subjective reaction belonging to the unacceptable category; The number of levels reflects subjective feelings; This reflects the differences in the degree of human perception.

[0028] Furthermore, the specific process of adaptive threshold update in step four includes:

[0029] 401 Sliding Window Management: Maintain a sliding window with a fixed capacity of N. To manage recent interaction data, store various indicators from the most recent N observations and corresponding user subjective comfort feedback, with feedback levels including comfortable, average, and uncomfortable;

[0030] ;

[0031] in, This represents the dataset at time t of the data collection. Let i represent the vibration comfort evaluation index for the t-th observation, where i takes values ​​in the range [1,4], meaning that it traverses the four evaluation indices. This represents the subjective comfort feedback provided by the user in the t-th instance;

[0032] 402 Boundary Region Definition: For the current threshold Define its boundary region as ,in This is the boundary width parameter; samples within the boundary region are used to trigger threshold adjustment.

[0033] 403 Misjudgment Statistics:

[0034] Type A and Type I misjudgments, overly lenient misjudgments: User feedback is "moderate" or "uncomfortable," but a certain indicator is judged as "comfortable." Calculate the misjudgment rate for this type. :

[0035] Type B, Type II False Positives: Overly strict judgments: User feedback indicates comfort, but a certain indicator is judged as uncomfortable. Statistical analysis of this type of false positive rate is required. :

[0036] 404 Adaptive Step Size Update:

[0037] Calculate a normalized misclassification rate difference to quantify the direction and urgency of threshold adjustment, expressed as:

[0038] ;

[0039] Where ε=0.1 is a constant used to avoid the denominator being zero; This indicates that there are more false positives in Type II, and the threshold needs to be adjusted upwards to reduce false alarms; This indicates that there are more false positives in Type I, and the threshold needs to be adjusted downwards to reduce false negatives;

[0040] 405 Validity Physical Constraint: The updated threshold must satisfy: and ;

[0041] ;

[0042] ;

[0043] in, The minimum distance between the two thresholds; The comfort / general boundary threshold at the final moment after iterative calculation; The general / uncomfortable boundary threshold at the final moment after iterative calculation;

[0044] Otherwise, project to the nearest feasible point to ensure the distinguishability of the general comfort level.

[0045] Furthermore, in step five, the system completes the calculation and evaluation of all indicators within a 3ms delay through edge computing nodes, and supports automatic reconvergence of thresholds after different users have changed them.

[0046] Furthermore, in step two, the acceleration time history data Second-order Butterworth bandpass filtering was performed on 20 frequency bands with center frequencies of 1Hz, 1.25Hz, 1.6Hz, 2Hz, 2.5Hz, 3.15Hz, 4Hz, 5Hz, 6.3Hz, 8Hz, 10Hz, 12.5Hz, 16Hz, 20Hz, 25Hz, 31.5Hz, 40Hz, 50Hz, 63Hz, and 80Hz to obtain 20 sets of time history data for each frequency band.

[0047] Beneficial effects:

[0048] Compared with the prior art, the advantages of the present invention are as follows:

[0049] 1) This invention creatively introduces an adaptive learning model, breaking the "one-size-fits-all" fixed threshold limitation in traditional standards. By continuously receiving subjective feedback tags from residents through mobile terminals, the system can use machine learning algorithms to dynamically update the judgment boundaries of multi-dimensional features in real time, so that the evaluation threshold automatically converges to the true tolerance of a specific user or a specific group of people, truly realizing the self-evolution and personalized service of the comfort evaluation system;

[0050] 2) Based on the Internet of Things gateway and adaptive learning cloud architecture, the system can not only complete complex multi-dimensional index calculations, but also adapt to environmental changes through algorithms. It can maintain a very high evaluation accuracy without human intervention. It can also update the system according to the changed threshold after the residential user is replaced, which greatly reduces the cost of large-scale deployment and long-term operation and maintenance, and has extremely high engineering promotion and application value. Attached Figure Description

[0051] Figure 1 This is a flowchart of an adaptive learning-based personalized assessment method for residential vibration comfort according to the present invention.

[0052] Figure 2 This is an adaptive effect diagram of the personalized assessment method for residential vibration comfort based on adaptive learning according to the present invention. Detailed Implementation

[0053] The technical solution of the present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0054] like Figure 1 As shown, the specific implementation of the adaptive learning-based personalized assessment method for residential vibration comfort of the present invention adopts the following technical solution:

[0055] Step 1: Data Acquisition: Utilizing the built-in accelerometer of the EdgeVibMS system to measure vibration data within the building's interior. The data is transmitted to the edge computing node via a wired connection. Since the impact of indoor vibration on human comfort is mainly concentrated in the range of 1~80Hz, the sampling frequency is determined to be 200Hz according to the Nyquist sampling theorem.

[0056] Step 2: Filtering: Filter the collected vibration acceleration time history data. A second-order Butterworth bandpass filter was applied to 20 frequency bands with center frequencies of 1 Hz, 1.25 Hz, 1.6 Hz, 2 Hz, 2.5 Hz, 3.15 Hz, 4 Hz, 5 Hz, 6.3 Hz, 8 Hz, 10 Hz, 12.5 Hz, 16 Hz, 20 Hz, 25 Hz, 31.5 Hz, 40 Hz, 50 Hz, 63 Hz, and 80 Hz to obtain 20 sets of time history data for each frequency band. The specific process is as follows:

[0057] 1) Design filter parameters:

[0058] First, it is necessary to determine the cutoff frequency, quality factor, and other parameters of the bandpass filter. Then, these parameters are used to calculate the coefficients b0, b1, b2, a1, and a2 of the difference equation.

[0059] 2) Initialization state:

[0060] Initialize the values ​​of y[n-1], y[n-2], x[n-1], and x[n-2] to 0.

[0061] 3) Processing input signals:

[0062] For each input sample x[n], calculate the output y[n] according to the difference equation:

[0063] (1)

[0064] Where y[n] is the output sample; x[n] is the input sample; b0, b1, b2 are the input coefficients; and a1, a2 are the output coefficients.

[0065] 4) Update status:

[0066] Update the state variables by assigning the current input and output values ​​to the corresponding state variables for calculation of the next sample.

[0067] 5) Repeat steps 3 and 4 until all input samples have been processed.

[0068] This process filters each sample of the input signal, retaining the signal components within the frequency range selected by the bandpass filter while removing components at other frequencies.

[0069] The original vibration data can be obtained through formula (1). After bandpass filtering of each frequency band (1 / 3 octave), 20 sets of time history data corresponding to each frequency band were obtained. .

[0070] Step 3: Index Extraction: Multiple evaluation indicators are calculated using the raw vibration data through edge computing nodes. Considering the special residential area limits for residential buildings, the time history data corresponding to the 1 / 3 octave band is extracted with a time period of 1 second. Find the maximum Z-level value after Z-weighting factor correction for whole-body vibration. Considering the comfort of whole-body vibration exposure, the time-history data will be used. Solve for the weighted root mean square (RMS) and the human vibration perception ratio (WODL); consider the impact of subway operation on human comfort, and use time-history data. Solving for the vertical fourth power vibration dose value Considering the different levels of acceptance among different groups, the percentage of people who find the vibration "unacceptable" out of the total population is calculated as the annoyance rate.

[0071] Step 4, Threshold Adaptation: The obtained indicators are input into the adaptive learning model. The initial benchmark is combined with the current national standards to set the basic thresholds for multi-dimensional features. The misjudgment rate of the basic thresholds is statistically analyzed through the stored recent user feedback. The adaptive threshold update is trained with user subjective feelings as the supervision label.

[0072] Step 5: Analysis and Evaluation: Within a specified time period, the evaluation weight thresholds are trained to converge with the actual tolerance of different users. The vibration level thresholds for some indicators are different during the day and at night. The daytime range is set to 6:00 AM to 6:00 PM, and the rest of the time is set to nighttime. Based on the boundary thresholds for different users, the model is trained for each user's tolerance level, and the current residential vibration state is output in real time to determine the result for that user, thus achieving a personalized evaluation that is tailored to each individual.

[0073] Preferably, in step three, the specific indicator extraction process is as follows:

[0074] A. Determine the vibration level within the building: Use the time history data corresponding to the 1 / 3 octave band. The maximum Z-level value under impact vibration load was obtained by processing the data using the impact vibration algorithm and other algorithm modules. The specific method is as follows:

[0075] (2)

[0076] (3)

[0077] (4)

[0078] In the formula: For the first The effective value of vibration acceleration at each center frequency, in units of ; For the first The time histories corresponding to each 1 / 3 octave band are obtained by bandpass filtering and spectral transformation of the vibration data. This is the time integration constant, usually taken as 1 second; The effective value of the vibration reference acceleration. ; The constant of time integration Calculated Vertical vibration acceleration level at a center frequency; For 1 / 3 octave band Z-weighting factor corresponding to each center frequency; It is the Z-level vibration.

[0079] B. Solving for Human Vibration Comfort: According to human anatomy, the three mutually perpendicular axes of the human body are the back-chest axis (X-axis), the right-left axis (Y-axis), and the foot-head axis (Z-axis). When the human body experiences vibration, the acceleration along the X-axis is expressed as... The acceleration along the Y-axis is expressed as The acceleration along the Z-axis is expressed as Rotational vibrations about each axis are called angular vibrations; angular vibrations about the X-axis are called roll, angular vibrations about the Y-axis are called pitch, and angular vibrations about the Z-axis are called yaw. The specific methods are as follows:

[0080] (5)

[0081] In the formula: Let be the acceleration value at a certain time t, in m / s². 2 T represents the test duration, in seconds.

[0082] The Human Vibration Perception Ratio (WODL) is the percentage of vibrations exceeding the vibration perception threshold. The specific method is as follows:

[0083] (6)

[0084] C. Determining the comfort level of building vibrations caused by subway operation: The time history data is processed using a basic frequency weighting algorithm module to obtain the vertical fourth-order vibration dose value. The specific method is as follows:

[0085] (7)

[0086] In the formula, Vertical fourth power dose value (m / s) 1.75 ); W for basic frequency weighting k instantaneous vertical acceleration (m / s²) 2 ); T is the duration, usually taken as 1 second.

[0087] D. Solving for the annoyance rate: The root mean square acceleration weighted by frequency is used as the basis for vibration comfort, and the annoyance rate A of the population at this intensity is calculated based on it. The specific method is as follows:

[0088] (8)

[0089] (9)

[0090] (10)

[0091] In the formula, Let f be the peak acceleration of the vibration signal at frequency f. Peak factor, Let i be the annoyance rate at the i-th vibration intensity; The number of people who exhibit the j-th subjective response at the i-th vibration intensity; Let be the degree of membership of the j-th subjective reaction belonging to the "unacceptable" category; The number of levels reflects subjective feelings; This reflects the differences in the degree of human perception.

[0092] Preferably, in step four, the specific adaptive weight update process is as follows:

[0093] Acceleration time history data Second-order Butterworth bandpass filtering was performed on 20 frequency bands with center frequencies of 1 Hz, 1.25 Hz, 1.6 Hz, 2 Hz, 2.5 Hz, 3.15 Hz, 4 Hz, 5 Hz, 6.3 Hz, 8 Hz, 10 Hz, 12.5 Hz, 16 Hz, 20 Hz, 25 Hz, 31.5 Hz, 40 Hz, 50 Hz, 63 Hz, and 80 Hz to obtain 20 sets of time history data for each frequency band.

[0094] 1) Learning mechanism and key sample extraction:

[0095] The algorithm maintains a sliding window with a fixed capacity of N. To manage recent interaction data, enabling dynamic tracking and adaptation to changes in user preferences:

[0096] (11)

[0097] in, This represents the dataset at time t of the data collection. Let i represent the vibration comfort evaluation index for the t-th observation, where i takes values ​​in the range [1,4], meaning that it traverses the four evaluation indices. This represents the subjective comfort feedback provided by the user in the t-th instance.

[0098] 2) Intelligent threshold adjustment:

[0099] For the current threshold Define its boundary region as ,in This is the boundary width parameter. Samples within the boundary region have the highest information content for threshold adjustment.

[0100] 3) Criteria for Correcting Misjudgments:

[0101] Regarding the inconsistency between the classification results and actual user feedback, for each threshold b1 and b2, two types of misclassifications are statistically analyzed:

[0102] A. Too lenient: User feedback is "neutral" or "uncomfortable," but based on the current threshold, a certain evaluation indicator is judged as comfortable. This suggests that the current threshold is set too high, leading to underreporting of uncomfortable vibrations. The formula for calculating false positives can be expressed as:

[0103] (12)

[0104] (13)

[0105] B. Overly stringent: User feedback indicates "comfortable," but a certain evaluation metric is judged as uncomfortable. This suggests that the current threshold is set too low, leading to false positives for comfortable states. The formula for calculating false positives is as follows:

[0106] (14)

[0107] (15)

[0108] 4) Adaptive step size update:

[0109] Based on the misjudgment statistics, a normalized misjudgment rate difference is calculated to quantify the direction and urgency of threshold adjustment, which can be expressed as:

[0110] (16)

[0111] Here, ε=0.1 is a constant used to avoid the denominator being zero. This indicates that there are more false positives in Type II, and the threshold needs to be adjusted upwards to reduce false alarms; This indicates that there are more false positives in Class I, and the threshold needs to be adjusted downwards to reduce false negatives.

[0112] 5) Physical constraints on effectiveness:

[0113] To ensure that the updated thresholds always have clear physical meaning and classification utility, the algorithm enforces constraints: ensuring the order and minimum interval between the two thresholds to maintain the discriminative power of the "general" comfort levels. If the updated threshold violates the above constraints, it is projected to the nearest feasible point. The corresponding formula is shown below:

[0114] and (17)

[0115] ;

[0116] ;

[0117] in, The minimum distance between the two thresholds; The comfort / general boundary threshold at the final moment after iterative calculation; This is the general / uncomfortable boundary threshold at the final moment after iterative calculation.

[0118] Step 5: Analysis and Evaluation: Within a specified time period, the evaluation weight thresholds of different indicators are trained to converge on the actual tolerance of different users. The thresholds for some indicators are different during the day and at night. The daytime range is set from 6:00 AM to 6:00 PM, and the rest of the time is set as nighttime, so as to achieve personalized evaluation for each user.

[0119] like Figure 2 As shown, the effectiveness of the adaptive learning-based personalized assessment method for residential vibration comfort of the present invention is verified as follows: the comfort / general boundary threshold b1 of sensitive users converges to 60.6 dB after the 91st feedback (threshold deviation 0.6 dB, relative error 1.0%), and the general / uncomfortable boundary threshold b2 of sensitive users converges to 70.1 dB after the 106th feedback (threshold deviation 0.1 dB, relative error 0.14%).

[0120] The comfort / normal threshold b1 for standard users converged to 70 dB after the 106th feedback (threshold deviation of 0), and the normal / uncomfortable threshold b2 for standard users converged to 78 dB after the 1st feedback (threshold deviation of 2 dB, relative error of 2.5%). Since the initial thresholds (73, 78 dB) are close to the true thresholds for standard users, the algorithm only needs minor adjustments.

[0121] The comfort / moderate threshold b1 for insensitive users converged to 76.9 dB after the 60th feedback (threshold deviation of 3.1 dB, relative error of 3.88%), while the moderate / uncomfortable threshold b2 for insensitive users converged to 89.9 dB after the 76th feedback (threshold deviation of 0.1 dB, relative error of 0.11%). This demonstrates that the algorithm can automatically converge the thresholds to a reasonable range based on the actual feelings of different user types.

[0122] In summary, the algorithm can automatically converge the threshold to a reasonable range based on the actual experiences of different user types. Standard users converge the fastest due to their similar initial values, while sensitive and insensitive users require more feedback to overcome larger initial biases. The maximum relative error of the threshold across all user types is 3.88%, indicating that the algorithm performs well in threshold convergence.

[0123] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An adaptive learning-based personalized assessment method for residential vibration comfort, characterized in that, The steps include the following: Step 1: Data Acquisition: Utilize the accelerometer built into the vibration monitoring system to measure the vibration acceleration inside the building / residential space. ; Step 2, Filtering: Filter the collected vibration acceleration. Twenty sets of time history data corresponding to each frequency band were obtained by second-order Butterworth bandpass filtering. ; Step 3: Index Extraction: Using the raw vibration data through edge computing nodes, calculate multiple evaluation indices. With a time period of 1 second, calculate the following vibration comfort evaluation indices for each frequency band time history data after filtering: maximum Z-level, weighted root mean square value, human vibration perception ratio, vertical fourth power vibration dose value, and annoyance rate. Step 4: Adaptive Threshold Update: Input the vibration comfort evaluation index obtained in Step 3 into the adaptive learning model based on misjudgment statistics. The initial threshold is set according to the current national standards. Collect user subjective comfort feedback through mobile terminals and use this feedback as a supervision label. Calculate the misjudgment rates for two categories: too lenient and too strict. Based on the difference between the two misjudgment rates, dynamically adjust the comfort / normal dividing threshold b1 and the normal / uncomfortable dividing threshold b2. Step 5: Personalized Assessment Output: Within a specified time period, the updated comfort / normal threshold b1 and normal / uncomfortable threshold b2 are converged and trained to the current user's actual tolerance. Different vibration level thresholds are used for daytime and nighttime to achieve personalized assessment. The daytime period is from 6:00 to 18:00, and the rest of the time is nighttime.

2. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, In step one, the sampling frequency is set to 200Hz based on the human body's sensitive frequency band of 1~80Hz and the Nyquist theorem.

3. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, In step three, the maximum value of the Z-level vibration. The calculation method is as follows: for the first The maximum value of the Z-level is calculated using the impact vibration algorithm based on the time history data of each center frequency. ； In the formula: For the first The time histories corresponding to each 1 / 3 octave band are obtained by bandpass filtering and spectral transformation of the vibration data. This is the time integration constant, usually taken as 1 second; The effective value of the vibration reference acceleration. ; For 1 / 3 octave band Z-weighting factor corresponding to each center frequency; It is the Z-level vibration.

4. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, In step three, the weighted root mean square (RMS) value is calculated as follows: ； In the formula: The acceleration value at a certain time t, in m / s². 2 T represents the test duration in seconds; the Human Vibration Perception Ratio (WODL) is defined as the percentage of samples exceeding the vibration perception threshold.

5. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, In step three, the vertical fourth power vibration dose value The calculation method is as follows: ； In the formula, Vertical fourth power dose value (m / s) 1.75 ); W for basic frequency weighting k Instantaneous vertical acceleration, in m / s² 2 T represents the duration, which is 1 second.

6. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, In step three, the annoyance rate is calculated using a fuzzy membership function based on frequency-weighted root mean square acceleration. ； In the formula, Let i be the annoyance rate at the i-th vibration intensity; The number of people exhibiting the j-th subjective response at the i-th vibration intensity; Let be the membership degree of the j-th subjective reaction belonging to the "unacceptable" category; The number of levels reflects subjective feelings; This reflects the differences in the degree of human perception.

7. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, The specific process of adaptive threshold update in step four includes: 401 Sliding Window Management: Maintain a sliding window with a fixed capacity of N. To manage recent interaction data, store a certain evaluation index and the corresponding user subjective comfort feedback for the most recent N observations, with feedback levels including comfortable, average, and uncomfortable; ； in, This represents the dataset at time t of the data collection. Let i represent the vibration comfort evaluation index for the t-th observation, where i takes values ​​in the range [1,4]. This means that the four evaluation indices are traversed. This represents the subjective comfort feedback provided by the user in the t-th instance; 402 Boundary Region Definition: For the current threshold Define its boundary region as ,in This is the boundary width parameter; samples within the boundary region are used to trigger threshold adjustment. 403 False Positive Statistics: Type A and Type I misjudgments, overly lenient misjudgments: User feedback is "neutral" or "uncomfortable," but a certain evaluation metric is judged as "comfortable." Calculate the misjudgment rate for this type. : Type B, Type II False Judgment: Overly strict judgment: The user reports comfort, but a certain evaluation indicator is judged as discomfort. Calculate the false judgment rate for this type. : 404 Adaptive Step Size Update: Calculate a normalized false positive rate difference to quantify the direction and urgency of threshold adjustment, expressed as: ； Where ε=0.1 is a constant used to avoid the denominator being zero; This indicates that there are more false positives in Type II, and the threshold needs to be adjusted upwards to reduce false alarms; This indicates that there are more false positives in Type I, and the threshold needs to be adjusted downwards to reduce false negatives; 405 Validity Physical Constraint: The updated threshold satisfies: and ; ； ； in, The minimum distance between the two thresholds; The comfort / general boundary threshold at the final moment after iterative calculation; The general / uncomfortable boundary threshold at the final moment after iterative calculation; Otherwise, project to the nearest feasible point to ensure the differentiation of general comfort levels.

8. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, In step five, the edge computing node is configured to process the vibration data for each 1-second time period in real time and output the evaluation results.

9. The adaptive learning-based personalized assessment method for residential vibration comfort according to claim 1, characterized in that, In step two, acceleration time history data Second-order Butterworth bandpass filtering was performed on 20 frequency bands with center frequencies of 1 Hz, 1.25 Hz, 1.6 Hz, 2 Hz, 2.5 Hz, 3.15 Hz, 4 Hz, 5 Hz, 6.3 Hz, 8 Hz, 10 Hz, 12.5 Hz, 16 Hz, 20 Hz, 25 Hz, 31.5 Hz, 40 Hz, 50 Hz, 63 Hz, and 80 Hz to obtain 20 sets of time history data for each frequency band.

Images