A method for recognizing behavior patterns of indoor personnel

Abstract

The application provides an indoor personnel environmental behavior mode recognition method, aiming at the preference type behavior, first, the collected data is cleaned, then the data compression is carried out according to the characteristics of time series data, including segmenting the data according to important points and classifying the data segments by using a clustering algorithm, so as to realize the symbolic representation of time series, finally, the environment-behavior sequence is constructed by using the processed data, the time series association rules are formulated and the main association rules are extracted. The main association rules express that a behavior usually occurs after a series of state changes, so as to describe the long-time influence of environmental parameters on personnel environmental behavior. For the habit type behavior, only the collected data is cleaned. According to the characteristics of the behavior data, the behavior sequence is constructed and the main association rules are extracted. The corresponding main association rules express a series of behaviors produced by the personnel continuously, so as to describe the long-time influence of behavior factors on personnel environmental behavior.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of digital simulation processing, and particularly relates to an indoor personnel ring behavior mode recognition method. BACKGROUND

[0002] The building field consumes nearly one-third of the global energy consumption, accounting for 15% of direct carbon dioxide emissions, and has great potential for energy saving and emission reduction[1]. During the operation of the building, the energy saving design usually cannot achieve the ideal effect because the occupants do not use the building as expected. Liang[2] proposed in the research on green commercial buildings that the main reasons for the difference between the actual and the design of building energy consumption are: (1) the occupants in the building use more energy than expected; (2) the number of occupants in the building increases; (3) the energy saving technology does not work. Among them, the energy consumption difference caused by the behavior model that does not conform to the actual situation exceeds 30%. This is because the building designers usually do not realize the preferences and habitual needs of the occupants, so the indoor personnel behavior is often ignored or improperly simplified in the process of energy consumption prediction, which reduces the reliability of building performance evaluation. The behavior of the indoor personnel to adjust the environmental state is the ring behavior. Clevenger[3] et al. found that the energy consumption difference can reach 150% by using different personnel in-room schedules and different environmental preferences for energy consumption simulation. Therefore, the ring behavior of the indoor personnel plays a fundamental role in building energy saving design and evaluating building performance. Fully understanding the indoor personnel behavior is the key to bridging the gap between the predicted energy consumption and the actual situation, and is an important means to develop personalized control strategies and explore building energy saving potential.

[0003] The existing researches on ring behavior mainly predict the personnel ring behavior by constructing a model. The modeling methods mainly include Agent-based modeling, statistical analysis modeling, data mining modeling and random process modeling. For example, Mahdavi et al. established the statistical relationship between action behavior, outdoor meteorological parameters and personnel displacement, and summarized the long-term mode of personnel behavior expressed by the function relationship of indoor and outdoor environmental parameters[4]. Zaraket et al. proposed a probability mapping of family profile and family energy consumption, and obtained the energy consumption caused by various behaviors in different families according to the estimation of the "activity amount" of each family[5]. Zhao et al. predicted personnel behavior by using data mining method according to the operation mode of the power consumption equipment in the office[6]. The prediction of personnel behavior can improve the accuracy of the overall energy consumption prediction of the building, but the result still has great uncertainty when applied to the control of the ring control equipment. Therefore, the intelligent control of the ring control equipment in the building should be made from the perspective of describing the common behavior mode of the personnel.

[0004] The circadian behavior includes preference behavior and habit behavior. In the aspect of "preference behavior", most of the existing researches are about the key environmental parameters affecting the ventilation behavior, shading behavior, air conditioning behavior and lighting behavior of indoor personnel, or the "threshold range" of these parameters when the personnel generate the behavior. In the aspect of "habit behavior", most of the existing researches are about the behavior habit from the perspective of time table. For example, Bavaresco et al. [7] found that the interaction between personnel and blinds often occurs when personnel arrive at the workspace in the morning, and among all the personnel who open the interior blinds, 63.3% open the blinds when they arrive. Li Zhengrong [8] found that the daily activity of the office group participating in shading adjustment is not high, but the participation degree is obviously improved when the switching time such as work, off work and lunch break occurs.

[0005] According to the above content, it can be found that the current research on the circadian behavior of personnel is mainly to realize the prediction of behavior by constructing a model, which can improve the prediction accuracy of the overall energy consumption of the building in a statistical sense. However, the idea of realizing dynamic behavior prediction in a real scene through a model to improve the effect of intelligent control of the building still needs to be discussed. As for the description of the circadian behavior, the description of the habit behavior is less, and the key parameters of the preference behavior are mostly proposed. Moreover, these descriptions mainly focus on the influence of the factors at the current time or the previous time on the state at the current time, and the long-term influence is less researched, and the influence caused by the change trend of the parameters cannot be described. SUMMARY

[0006] To solve the above problems, a method for describing the circadian behavior pattern of indoor personnel is provided, and the technical scheme adopted by the present application is as follows:

[0007] The present application provides a method for recognizing the circadian behavior pattern of indoor personnel, which is used for describing the circadian behavior of indoor personnel. The circadian behavior is divided into preference behavior and habit behavior, and the method comprises the following steps: step S1, collecting environmental parameters and personnel circadian behavior data by using a sensor; step S2, respectively pre-processing the environmental parameters and personnel circadian behavior data; step S3, constructing an environment-behavior sequence based on the pre-processed environmental parameters and personnel circadian behavior data, taking the time factor as a state parameter, and formulating corresponding time sequence association rules for the preference behavior and the habit behavior; and step S4, analyzing the preference behavior pattern or the habit behavior pattern based on the time sequence association rules, extracting corresponding main association rules, and describing the long-term influence of the environmental parameters or the behavior factors on the personnel circadian behavior based on the main association rules.

[0008] The indoor personnel environment behavior mode recognition method provided by the application can also have the following technical features: the environment parameters at least include indoor illumination, indoor temperature, carbon dioxide concentration, outdoor temperature and weather, and the personnel environment behavior data at least include sunshade behavior, lighting behavior, ventilation behavior and air conditioning behavior.

[0009] The indoor personnel environment behavior mode recognition method provided by the application can also have the following technical features: in step S2, the data cleaning operation is performed on the environment parameters and the personnel environment behavior data, and the preference type behavior data in the environment parameters and the personnel environment behavior data is compressed based on the TD algorithm and the k-medoids algorithm, that is, only the data cleaning operation is performed on the habit type behavior.

[0010] The indoor personnel environment behavior mode recognition method provided by the application can also have the following technical features: the following data cleaning operation is performed on the environment parameters and the personnel environment behavior data:

[0011] The missing values are processed by using the MICE chain equation multiple imputation method, specifically: m complete equivalent imputation sets are generated by filling the missing values, and the most suitable imputation combination is selected according to the evaluation standard analysis result; the data is denoised by using the convolution based on the signal window, specifically: each element is allocated to a window of a specified size, then the average value of the data in the window is calculated, the average value is used to replace the original data, and all data is traversed by rolling on the complete data; the abnormal value is processed by using the Lambda criterion, that is: assuming that a group of data only contains random errors, the sample data mean and standard deviation are calculated, and if:

[0012] |x j -μ|≥3σ

[0013] x j is regarded as an abnormal value and is removed.

[0014] The indoor personnel environment behavior mode recognition method provided by the application can also have the following technical features: the time series compression data compression method is used for the environment parameters, the time series compression refers to the process of segmenting and mode representing the time series, including time series segmentation, clustering analysis on the segmented sequence and symbolic representation to form a mode sequence, and the specific process is as follows: first, the top-down TD algorithm is used to segment the environment parameter time series, then the k-medoids algorithm is used to cluster the segmented sequence into different categories, and finally, when the symbolic representation is considered, the line segment properties include the average value, slope and horizontal intercept of the line segment, which respectively correspond to the absolute value of the environment parameter, the change trend of the environment parameter and the time length contained in the environment parameter time series data.

[0015] The indoor personnel ring behavior mode recognition method provided by the application can also have the following technical features: the specific process of formulating time sequence association rules for the preference type behavior is as follows: the time factor is taken as a state parameter, the time at which the transaction occurs is taken as an event in the time sequence association rules, the time of a day is divided into a predetermined number of intervals, and the SPADE algorithm based on the improved Apriori is used to mine the time sequence association rules: when the association rules are formulated, the frequent item sets are first screened by setting the minimum support, and then the main association rules are screened by setting the minimum confidence, the minimum support of 5% commonly used in the related field is used, and the total number of rule occurrences is not less than 3 times; the confidence threshold of 50% is used, that is, the occurrence of the former item will probably be followed by the occurrence of the latter item, and the rules with the lift greater than 1 are also included in the main association rules.

[0016] The indoor personnel ring behavior mode recognition method provided by the application can also have the following technical features: the specific process of formulating time sequence association rules for the habit type behavior is as follows: for the discrete personnel ring behavior data, a column of non-empty data is supplemented for the empty data in the data, which plays a role in filling the time interval when the association rules are formed, that is, in the process of formulating the association rules, when no time interval is set between the events, the data is filled in the interval between the occurrence of other behaviors to represent the time elapsed between the behaviors, the column of data uses the time parameter, the time parameter has no vacancy, and can represent the approximate events of the occurrence of behaviors.

[0017] The indoor personnel ring behavior mode recognition method provided by the application can also have the following technical features: in the analysis based on the time sequence association rules in step S4, the sequence without the time parameter is used for analysis first, the main association rules are identified, and then the sequence with the time parameter is used for analysis, and the time characteristics of the main association rules are extracted as the frequent item sets.

[0018] Invention action and effect

[0019] According to the indoor personnel environmental behavior pattern recognition method of the present application, for the preference type behavior, the present application first cleans the collected data, then compresses the data according to the characteristics of the time series data, including segmenting the data according to the important points and classifying the data segments using the clustering algorithm, so as to realize the symbolic representation of the time series, and finally constructs the environment-behavior sequence using the processed data, formulates the time series association rules and extracts the main association rules. The main association rules express that a certain behavior usually occurs after a series of state changes, thereby describing the long-term influence of the environmental parameters on the personnel environmental behavior. For the habit type behavior, the behavior pattern is consistent with the behavior mode affected by the preference, the present application first cleans the collected data, but does not need to compress the data. According to the characteristics of the behavior data, the behavior sequence is constructed, and the main association rules are extracted. The corresponding main association rules express a series of behaviors produced by the personnel in succession, thereby describing the long-term influence of the behavior factors on the personnel environmental behavior.

[0020] The indoor personnel environmental behavior pattern recognition method of the present application can realize the description of the personnel preference type behavior and the habit type behavior under the long-term influence, solves the problem that the habit type behavior and the long-term influence are less described in the prior art, and lays a foundation for the subsequent research on the personnel environmental behavior under the long-term influence. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart of the indoor personnel environmental behavior pattern recognition method in the embodiment of the present application;

[0022] Figure 2 is an environmental parameter and personnel environmental behavior data mapping diagram in the embodiment of the present application;

[0023] Figure 3 is an illumination data segmentation situation diagram in the embodiment of the present application;

[0024] Figure 4 is a habit affected behavior mode diagram of the indoor personnel to be tested in the embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the technical means, creative features, purposes and effects realized by the present application easy to understand, the indoor personnel environmental behavior pattern recognition method of the present application is specifically described below in combination with the embodiments and the drawings.

[0026] The technical terms related to the present application are explained as follows:

[0027] Looping behavior: the behavior of people to control the indoor environment unconsciously and consciously by comparing the perceived environment with the sum of past experience, which is divided into preference behavior and habit behavior. Preference behavior: the looping behavior caused by the deviation of the indoor environment state from the individual comfort interval of the person, such as adjusting the air conditioner when the person feels cold or hot. Habit behavior: the looping behavior mainly affected by the habit of the person, not caused by the stimulation of the surrounding environment change, such as adjusting the air conditioner or lighting due to the person leaving the room. Behavior pattern: the generalization of the behavior of the person repeatedly occurring, which refers to the specific tendency or stable coherence shown by the external behavior. Behavior pattern affected by preference: a certain behavior usually occurs after a series of environmental state changes. Habit-affected behavior pattern: a series of behaviors produced by the person in sequence, i.e. behavior sequence.

[0028] <Embodiment>

[0029] Figure 1 is a flowchart of the indoor personnel looping behavior pattern recognition method in the embodiment of the present application.

[0030] As shown in Figure 1 , the specific process of the behavior pattern recognition of the indoor personnel behavior to be tested based on the indoor personnel looping behavior pattern recognition method in the embodiment is as follows:

[0031] Step S1, collect environmental parameter and personnel looping behavior data as original data set.

[0032] In the embodiment, various sensors are used to monitor the environmental parameters and behavior data required for extracting typical behavior patterns. Personnel looping behavior includes shading behavior, lighting behavior, ventilation behavior and air conditioning behavior. According to existing research, the environmental parameters affecting behavior include indoor illuminance, temperature and carbon dioxide concentration, outdoor temperature and weather, and the corresponding relationship between environmental parameters and behavior is shown in the mapping diagram of Figure 2 . Therefore, the personnel looping behavior to be monitored includes lighting behavior, shading behavior, ventilation behavior and air conditioning behavior, and the environmental parameters to be monitored include indoor and outdoor temperature, weather, illuminance and carbon dioxide concentration. The sensors required to achieve the above parameter monitoring are shown in Table 1.

[0033] Table 1 Environmental parameter and behavior data monitoring method

[0034]

[0035]

[0036] Step S2, data preprocessing is performed on the collected original data set.

[0037] In this embodiment, first, the environmental parameters and personnel environmental behavior data are subjected to the following data cleaning operations:

[0038] Because the data is susceptible to equipment failure, communication interruption and other influences during transmission, the data needs to be cleaned before analysis, that is, the missing values are processed by MICE chain equation multiple imputation based on multiple imputation method. The implementation process includes: filling in the missing values to generate m completely equivalent imputation sets; analyzing the results according to the evaluation criteria; and selecting the most appropriate imputation combination.

[0039] Because the sensor measurement value contains random errors and is sensitive to environmental changes, the data obtained by actual measurement of the sensor usually shows the characteristics of being not smooth and not clean. Therefore, the data is denoised by using convolution based on signal (i.e. data point) window. The basic algorithm is based on the rolling window mode: each element is assigned to a window of a specified size, then the average value of the data in the window is calculated, the average value is used instead of the original data, and all data is traversed by rolling on the complete data.

[0040] Outlier processing is a process of finding behaviors significantly different from expected objects and processing accordingly. Outliers are different from noise data, and their generation is usually caused by reasons. The application of lambda criterion for processing is that a set of data only contains random errors, the sample data mean, standard deviation are calculated, and if:

[0041] |x j -μ|≥3σ (1)

[0042] x j it is considered as an outlier and is removed.

[0043] Then, the preference behavior data in the environmental parameters and personnel environmental behavior data are subjected to data compression based on TD algorithm and k-medoids algorithm, and the consistent compression degree is maintained:

[0044] The amount of time series data of environmental parameters obtained in the field monitoring is large and data-intensive, and a large amount of calculation is required when data mining is performed. Meanwhile, according to the accuracy of data collection, the number of states of environmental parameters can reach dozens to hundreds, and it is difficult to extract regularity information using accurate numerical values and behaviors, and it does not match the actual situation and application requirements. For example, before the person generates the open air conditioner behavior, the indoor temperature will not always remain the same accurate value, but within a certain range (due to the randomness and complexity of human behavior, the indoor temperature triggering the behavior will fluctuate within a certain range). In addition to the influence on the person, the trend of the change of the environmental parameter also has a certain influence on the human body, and the original data cannot directly reflect the trend information. Therefore, before mining the association rules, the measured data needs to be processed to reduce the calculation amount when mining. Time series compression can achieve this process. Time series compression refers to the process of segmenting and patternizing the time series, including time series segmentation, clustering analysis of segmented sequences, and symbolic representation to form pattern sequences.

[0045] In this embodiment, the top-down TD algorithm is used to segment the time series of environmental parameters, and k-medoids is used to cluster them into several categories. According to the existing research [9] , the behavior of the person is related to the time, intensity and intensity-time change rate of the stimulus, therefore, when clustering and symbolizing in this embodiment, the line segment properties considered include the average value, slope and transverse intercept (the number of data points included) of the line segment, which represent the absolute value of the environmental parameter, the change trend of the environmental parameter and the time length included by the segment data, respectively.

[0046] Figure 3 is a schematic diagram of the segmentation of the illumination data in the embodiment of the present application.

[0047] For example, the parameters listed in Table 1 of a certain residence in Suzhou are monitored, and the illumination data is taken as an example. After data preprocessing and data compression, the original data points are 6048, and are compressed into 2529 line segments, and the segmentation situation is shown in Table 2 as shown in Figure 3 After segmentation, clustering analysis is performed, and the symbolic typical change curve segment obtained is shown in Table 2.

[0048] Table 2 Symbolic summary of illumination data

[0049]

[0050] Step S3, constructing time sequence association rules and analyzing: based on the pre-processed environmental parameters and personnel environmental behavior data, constructing the sequence of environment-behavior, taking the time factor as the state parameter, and formulating the corresponding time sequence association rules for the preference type behavior and the habit type behavior; based on the time sequence association rules, analyzing the preference type behavior mode or the habit type behavior mode, extracting the corresponding main association rules, and describing the long-term influence of the environmental parameters or the behavior factors on the personnel environmental behavior based on the main association rules.

[0051] The specific analysis of the time sequence association rules is as follows according to the preference type behavior and the habit type behavior.

[0052] The analysis process of the time sequence association rules for the preference type behavior is as follows:

[0053] In addition to the environmental parameters and the personnel environmental behavior data, the time factor is taken as a state parameter, and the "transaction time is in a certain time period of a day" is taken as an event in the time sequence association rules. First, the time of a day is divided into four intervals: [0:00, 6:00) for early morning, [6:00, 12:00) for morning, [12:00, 18:00) for afternoon, and [18:00, 24:00) for night.

[0054] Then, the SPADE algorithm based on the Apriori is used for time sequence association rule mining. When formulating the association rules, the frequent item sets are first screened by setting the minimum support, and then the main association rules are screened by setting the minimum confidence. The commonly used 5% in the related field is used as the minimum support, and the total occurrence number of the rules is not less than 3 times. The confidence threshold of 50% is used, that is, the occurrence of the former item will probably occur after the latter item, and the rules with the lift greater than 1 are also included in the main association rules.

[0055] In the analysis, the sequence without the time parameter is used for analysis first, and the main association rules are identified, and then the time characteristics of the main association rules are analyzed by adding the time parameter. Similarly, the time sequence association rule analysis is performed on the personnel lighting behavior, and the main time sequence association rules obtained by mining are shown in Table 3.

[0056] Table 3 Main time sequence association rules of indoor illuminance and lighting behavior (without time period parameter item)

[0057]

[0058] After adding the time parameter, the time sequence association analysis of the indoor illuminance and the lighting behavior is performed again to extract the time characteristics of the main association rules, and the frequent item sets of the indoor illuminance and the lighting behavior containing the time period parameter item are summarized in Table 4.

[0059] Table 4 Frequent item sets of indoor illuminance and lighting behavior (with time period parameter item)

[0060]

[0061] The rule 1 in the above table indicates that the light-on behavior occurs after a period of time (about 8.00 lux, 10-20 minutes) of keeping low illuminance unchanged. The rules 2 and 3 indicate that the light-on and then light-off behavior occurs after a period of time (about 12.32 lux, slope 0.27, total 45-50 minutes) of increasing illuminance after the indoor illuminance slowly decreases (about 7.96 lux, slope -0.26) in the early morning.

[0062] The analysis process of the time sequence association rules of the habit type behavior is as follows:

[0063] The data preprocessing is also needed before the time sequence association rule analysis of the habit type behavior pattern. The behavior data are discrete data, and thus do not need to be segmented. The time length of a single data point is 5 minutes. Since there are a large number of null data points in the behavior sequence, there is a certain time interval between events in the time sequence association analysis result, and thus a column of null data needs to be supplemented to fill the time interval in the formation of the association rules, that is, in the process of formulating the association rules, the data is filled in the interval between other behaviors to represent the time elapsed between behaviors. The column of data uses the time parameter, the time parameter has no null, and can represent the approximate time of the behavior.

[0064] The time sequence association rule analysis is also performed on the above example. The analysis objects include the air conditioning behavior, the lighting behavior, the ventilation behavior and the shading behavior. The main association rules obtained by the time sequence association rule mining are summarized in Table 5 below.

[0065] Table 5 Main time sequence association rules between personnel behaviors (without time period parameter item)

[0066]

[0067] The time sequence association analysis of the indoor personnel behavior is performed again to extract the time characteristics of the main association rules by adding the time parameter. The frequent item sets of the personnel behavior with the time period parameter item are summarized in Table 6 below.

[0068] Table 6 Frequent item sets of personnel behavior (with time period parameter item)

[0069]

[0070] In the table above, rule 1 indicates that if the window and air conditioner are closed in the afternoon, the air conditioner will be turned on within 0-10 minutes. Its confidence is 1, and its lift is greater than 1. Rule 2 indicates that if the air conditioner is turned on within 5-15 minutes after the air conditioner is turned off in the evening, the lights will be turned off within 0-10 minutes. Its lift is greater than 1, making it a recommended rule. Rule 3 indicates that if the air conditioner is turned off in the evening, the lights will be turned off within 5-15 minutes. Its lift is greater than 1, making it a recommended rule. Rule 4 indicates that if the air conditioner is turned on in the early morning, the lights will be turned on within 15-25 minutes. Its lift is greater than 1, making it a recommended rule.

[0071] According to the main time series association rules extracted above, the behavior pattern of indoor personnel at this measurement point affected by habits can be found in Figure 4 .

[0072] Example Function and Effect

[0073] According to the method for identifying indoor environmental control behavior patterns of personnel, for preference-based behaviors, the collected data is first cleaned. Data compression is then performed based on the characteristics of the time series data. This involves segmenting the data according to key points and classifying the data segments using a clustering algorithm, thereby achieving a symbolic representation of the time series. Finally, the processed data is used to construct an environment-behavior sequence, formulate time series association rules, and extract key association rules. These key association rules express the behaviors that typically occur after a series of state changes, thereby describing the long-term impact of environmental parameters on personnel environmental control behaviors. For habitual behaviors, consistent with the behavior patterns influenced by preferences, the collected data is first cleaned, but data compression is not required. Behavior sequences are constructed based on the characteristics of the behavioral data, and key association rules are extracted. The corresponding key association rules are used to express a series of coherent behaviors of personnel, thereby describing the long-term impact of behavioral factors on personnel environmental control behaviors.

[0074] In this embodiment, data cleaning operations such as interpolation of missing values, noise reduction, and outlier processing are performed on the data, effectively addressing data errors and unevenness, thereby improving the accuracy of behavioral descriptions. Furthermore, data compression using the TD algorithm and the k-medoids algorithm reduces the computational effort involved in mining association rules, improving efficiency.

[0075] In summary, the indoor personnel environmental control behavior pattern recognition method of this embodiment can respectively describe the personnel preference-type behavior and habitual behavior under long-term influence, solving the problem of insufficient description of habitual behavior and long-term influence in the prior art, and laying the foundation for subsequent research on personnel environmental control behavior under long-term influence.

[0076] The above examples are only used to illustrate the specific embodiments of the present application, and the present application is not limited to the description range of the above examples.

[0077] The above references are:

[0078] [1] Agency I E. Buildings: A Source of Enormous Untapped Efficiency Potential [J]. 2021.

[0079] [2] Liang X, Hong T, Shen G Q. Occupancy data analytics and prediction: A case study [J]. Building and Environment, 2016, 102: 179-192.

[0080] [3] Clevenger C M, Haymaker J. The impact of the building occupant on energy modeling simulations [C] / / Joint International Conference on Computing and Decision Making in Civil and Building Engineering, Montreal, Canada.

[0081] [4] Mahdavi A, Tahmasebi F. Predicting people’s presence in buildings: An empirically based model performance analysis [J]. Energy and Buildings, 2015, 86: 349-355.

[0082] [5] Zaraket T, Yannou B, Leroy Y, et al. A Stochastic activity-based approach for forecasting occupant-related energy consumption in residential buildings [C] / / International Design Engineering Technical Conferences and Computers and Information in Engineering Conference.

[0083] [6] Zhao J, Lasternas B, Lam K P, et al. Occupant behavior and schedule modeling for building energy simulation through office appliance power consumption data mining [J]. Energy and Buildings, 2014, 82: 341-355.

[0084] [7] Bavaresco M V, Ghisi E. A low-cost framework to establish internal blind control patterns and enable simulation-based user-centric design [J / OL]. Journal of Building Engineering, 2020, 28: 101077.

[0085] [8] Li Z, Zhao Y, Su W. Research on environmental regulation behavior of office buildings in Shanghai area [J]. Building Energy Efficiency, 2019, 47(10): 4.

[0086] [9] Li Z, Zhu H, Dong B, et al. Development of a systematic procedure to establish customized shading behavior identification model [J / OL]. Energy and Buildings, 2021, 239: 110793.

Claims

1. A method for identifying indoor personnel environmental control behavior patterns, used to describe indoor personnel environmental control behaviors, wherein the environmental control behaviors are divided into preference-based behaviors and habit-based behaviors, characterized in that: The following steps are involved: Step S1, using sensors to collect environmental parameters and personnel environmental control behavior data; Step S2, performing data preprocessing on the environmental parameters and the personnel environmental control behavior data respectively; Step S3: construct an environment-behavior sequence based on the preprocessed environmental parameters and personnel environmental control behavior data, taking the time factor as the state parameter, and formulating corresponding time series association rules for preference-based behavior and habitual behavior respectively; Step S4: Analyze the preference behavior pattern or habitual behavior pattern based on the time series association rules, extract the corresponding main association rules, and describe the long-term impact of environmental parameters or behavioral factors on personnel environmental control behavior based on the main association rules.

2. The method for recognizing indoor personnel environmental control behavior patterns according to claim 1, characterized in that: in, The environmental parameters include at least indoor illumination, indoor temperature, carbon dioxide concentration, outdoor temperature and weather, and the personnel environmental control behavior data includes at least sunshade behavior, lighting behavior, ventilation behavior and air conditioning behavior.

3. The method for recognizing indoor personnel environmental control behavior patterns according to claim 1, characterized in that: in, In step S2, data cleaning operations are performed on both the environmental parameters and the personnel environmental control behavior data, and data compression is performed on the environmental parameters and the preference behavior data in the personnel environmental control behavior data based on the TD algorithm and the k-medoids algorithm, that is, only data cleaning is performed on habitual behavior.

4. The method for recognizing indoor personnel environmental control behavior patterns according to claim 3, characterized in that: in, Perform the following data cleaning operations on the environmental parameters and the personnel environmental control behavior data: The MICE chain equation multiple imputation method based on the multiple imputation method is used to handle missing values. Specifically, the missing values ​​are filled to generate m completely equivalent imputation sets, and the most appropriate imputation combination is selected based on the evaluation criteria analysis results; Use signal window-based convolution to perform data noise reduction. Specifically, each element is assigned to a window of a specified size, and then the average value of the data in the window is calculated. The average value is used to replace the original data, and all data are rolled over the complete data. Apply the Lambda criterion to handle outliers, that is, assume that a set of data contains only random errors, calculate the mean and standard deviation of the sample data, if: |x j -μ|≥3σ Then x j are considered as outliers and removed.

5. The method for recognizing indoor personnel environmental control behavior patterns according to claim 3, characterized in that: in, A data compression method of time series compression is used for the environmental parameters. Time series compression refers to the process of segmenting the time series and performing pattern representation, including time series segmentation, cluster analysis of the segmented sequence, and symbolic representation to form a pattern sequence. The specific process is as follows: First, the top-down TD algorithm is used to segment the environmental parameter time series, and then the k-medoids algorithm is used to cluster them into different categories. Finally, the segment attributes considered in the symbolic representation include the average value, slope and horizontal intercept of the segment, which correspond to the absolute value of the environmental parameter, the changing trend of the environmental parameter and the time length of the environmental parameter time series data.

6. The method for recognizing indoor personnel environmental control behavior patterns according to claim 1, characterized in that: in, The specific process of formulating time series association rules for preference-based behaviors is as follows: Take the time factor as a state parameter, make "the transaction time is in a certain time period of the day" an event in the time series association rule, and divide the day into a predetermined number of intervals. The SPADE algorithm based on the improvement of Apriori is used to mine temporal association rules: when formulating association rules, the frequent item sets are first filtered out by setting the minimum support, and then the main association rules are filtered out by setting the minimum confidence. We use 5% as the minimum support commonly used in related fields, and the total number of rule occurrences is no less than 3 times; we use 50% as the confidence threshold, which means that the subsequent term is likely to occur after the preceding term occurs. At the same time, we also include rules with a lift greater than 1 into the main association rules.

7. The method for recognizing indoor personnel environmental control behavior patterns according to claim 1, characterized in that: in, The specific process of formulating time series association rules for habitual behaviors is as follows: For discrete personnel environmental control behavior data, a column of non-empty data is added to the empty data in the data to fill the time interval when forming association rules. That is, in the process of formulating association rules, when there is no time interval between events, this data will be filled in the interval where other behaviors occur to represent the time elapsed between behaviors. This column of data uses time parameters, the time parameters have no gaps, and can represent the approximate events where the behaviors occur.

8. The method for recognizing indoor personnel environmental control behavior patterns according to claim 6 or 7, characterized in that: in, When analyzing based on time series association rules in step S4, first use the sequence without time parameters to analyze and identify the main association rules, then use the sequence with time parameters to analyze and extract the time features of the main association rules as frequent itemsets.

Images