IoT sensor data processing method and system based on smart building
Through the combination of graph network model and context generation unit, the IoT sensor network data is processed, which solves the problem of low sensor data processing efficiency, and realizes efficient and intelligent building environment status monitoring and early warning, improving prediction accuracy and Real-time.
Patent Information
- Application Number
- CN202411215971.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-09-02
AI Technical Summary
The existing technology is difficult to efficiently and accurately process and analyze massive IoT sensing network data, resulting in low efficiency in monitoring and abnormal warning of built environments, and it is difficult to fully reflect the actual state of the built environment.
Using graph network model, context generation unit and dynamic guided knowledge points, the high-dimensional relationship graph vector and cross-correlation features are generated through the combination of graph self-coding and multi-guided knowledge points, and the deep mining of IoT sensor network data is realized and Intelligent analysis, dynamically update model parameters to adapt to different building characteristics.
It significantly improves the accuracy and real-time performance of building environment status prediction, can process large-scale data in a short time, meets the real-time monitoring and rapid response needs of smart buildings, provides comprehensive environmental assessment, and improves data processing efficiency and robustness.
Smart Images

Figure CN118734050B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of smart building technology, and more specifically, to a method and system for processing Internet of Things sensor data based on smart buildings. Background Art
[0002] In the tide of smart city construction, smart buildings, as an important part of it, are gradually becoming a key link in improving urban management efficiency and optimizing residents' living environment. Smart buildings integrate advanced technologies such as the Internet of Things (IoT), big data, and artificial intelligence (AI) to achieve intelligent management and control of the building's internal environment, facilities, and energy consumption. However, in practical applications, how to efficiently and accurately process and analyze IoT sensor network data to achieve real-time monitoring of the building environment status and abnormal warning is still one of the difficulties in current technological development.
[0003] Traditional building environment monitoring methods mainly rely on manual inspections and data collection from fixed sensors. This method has problems such as low data collection efficiency, untimely analysis, and slow abnormal response. With the rapid development of Internet of Things technology, more and more sensors are deployed inside buildings to collect a variety of environmental parameters such as temperature, humidity, light, and air quality in real time, providing a rich data source for smart buildings. However, in the face of massive, multi-dimensional Internet of Things sensor network data, how to effectively process and analyze data to quickly and accurately identify abnormal paths in the building environment has become a technical problem that needs to be solved urgently.
[0004] Existing data processing methods mostly focus on statistical analysis of single-dimensional data, ignoring the correlation and coordinated change characteristics between data, making it difficult to fully and accurately reflect the actual state of the building environment. Summary of the invention
[0005] In view of this, the purpose of this application is to provide an Internet of Things sensor data processing method and system based on smart buildings.
[0006] According to the first aspect of the present application, a method for processing sensor data of the Internet of Things based on a smart building is provided, the method comprising:
[0007] Acquire example learning data, wherein the example learning data includes a plurality of example IoT sensor network data, wherein the example IoT sensor network data has an example building environment abnormal path under one or more building feature tags among a plurality of building feature tags;
[0008] Based on the first graph network model in the building environment state prediction network, the example Internet of Things sensor network data is subjected to graph self-encoding processing to generate an example Internet of Things sensor network relationship graph vector;
[0009] Based on the context generation unit in the building environment state prediction network, generating example context features corresponding to multiple guiding knowledge points respectively according to the example IoT sensor network relationship graph vector;
[0010] Based on the second graph network model in the building environment state prediction network, generating example cross-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels according to the basic context features respectively corresponding to the multiple guiding knowledge points and the example context features respectively corresponding to the multiple guiding knowledge points;
[0011] Determine a first network training cost according to the example mutual correlation features of the multiple guiding knowledge points under each building feature label in the multiple building feature labels, the example Internet of Things sensor network relationship graph vector, and the example building environment abnormal path of the example Internet of Things sensor network data under one or more building feature labels;
[0012] The network parameter information of the building environment state prediction network and the multiple guiding knowledge points are updated according to the first network training cost, and the updated guiding knowledge points are used to express the building environment state in the Internet of Things sensor network data from multiple guiding description fields.
[0013] In a possible implementation of the first aspect, the second graph network model based on the building environment state prediction network generates example cross-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels according to the basic context features respectively corresponding to the multiple guiding knowledge points and the example context features respectively corresponding to the multiple guiding knowledge points, including:
[0014] Performing cross-correlation analysis on the basic context features of each guiding knowledge point and the example context features corresponding to the guiding knowledge point to generate mutual integration context features corresponding to each guiding knowledge point;
[0015] The second graph network model based on the building environment state prediction network performs graph self-encoding processing on the mutual fusion context features corresponding to each of the guiding knowledge points to generate example cross-correlation features of each guiding knowledge point under each of the multiple building feature labels.
[0016] In a possible implementation of the first aspect, the method further includes:
[0017] Determine the deviation cost according to the feature distance between every two example cross-correlation features among the example cross-correlation features corresponding to multiple guiding knowledge points under the same building feature label;
[0018] The guiding knowledge point is updated according to the deviation cost.
[0019] In a possible implementation of the first aspect, determining the deviation cost according to the feature distance between every two example mutual correlation features in the example mutual correlation features respectively corresponding to the multiple guiding knowledge points under the same building feature label includes:
[0020] Calculate the feature distance between every two example cross-correlation features among the example cross-correlation features corresponding to multiple guiding knowledge points under the same building feature label;
[0021] The feature distances are accumulated to generate deviation costs.
[0022] In a possible implementation manner of the first aspect, determining a first network training cost according to the example mutual correlation features of the multiple guiding knowledge points under each building feature label in the multiple building feature labels, the example Internet of Things sensor network relationship graph vector, and the example building environment abnormal path of the example Internet of Things sensor network data under one or more building feature labels includes:
[0023] Calculate the vector distance between the example cross-correlation features of each of the guiding knowledge points under each building feature label and the example Internet of Things sensor network relationship graph vector;
[0024] Determine the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label according to the vector distance between the example cross-correlation features corresponding to each guiding knowledge point under the same building feature label and the example Internet of Things sensor network relationship graph vector;
[0025] Determine a first network training cost according to the example building environment abnormal paths of the example IoT sensor network data under one or more building feature tags and the predicted building environment abnormal paths of the example IoT sensor network data under each building feature tag;
[0026] In a possible implementation of the first aspect, the step of determining a predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label based on a vector distance between the example cross-correlation features corresponding to each guiding knowledge point under the same building feature label and the example Internet of Things sensor network relationship graph vector includes:
[0027] For each building feature label, the vector distance between the example cross-correlation feature corresponding to each guide knowledge point and the example IoT sensor network relationship graph vector is normalized. According to the normalized vector distance, a target weight is assigned to each guide knowledge point. The size of the target weight is inversely proportional to the vector distance. The calculation formula of the target weight is: target weight = (1 - normalized vector distance)^α, where α is an adjustment factor used to adjust the sensitivity of weight allocation;
[0028] After obtaining the target weight of each guiding knowledge point, the example cross-correlation features of each guiding knowledge point under the building feature label are weighted fused, and the weighted fused features are input into the regression network in the building environment state prediction network for feature mapping to obtain the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label.
[0029] In a possible implementation manner of the first aspect, determining a first network training cost according to the example building environment abnormal path of the example IoT sensor network data under one or more building feature tags and the predicted building environment abnormal path of the example IoT sensor network data under each building feature tag includes:
[0030] The loss function value is calculated based on the abnormal building environment path of the example Internet of Things sensor network data under one or more building feature labels and the predicted abnormal building environment path of the example Internet of Things sensor network data under each building feature label to generate a first network training cost.
[0031] In a possible implementation of the first aspect, updating the network parameter information of the building environment state prediction network and the plurality of guiding knowledge points according to the first network training cost includes:
[0032] The network parameter information and the guiding knowledge points of the context generation unit are updated according to the first network training cost.
[0033] In a possible implementation of the first aspect, before the example Internet of Things sensor network data is subjected to graph self-encoding processing based on the first graph network model in the building environment state prediction network to generate an example Internet of Things sensor network relationship graph vector, the method further includes:
[0034] Acquire a plurality of sensor data-description combinations, wherein the sensor data-description combinations include template Internet of Things sensor network data and template guided description data;
[0035] The plurality of template guided description data are loaded into the second graph network model, and the plurality of template IoT sensor network data are loaded into the first graph network model, and comparative training is performed to generate feature distances between template IoT sensor network data and template guided description data belonging to the same sensor data-description combination, and feature distances between template IoT sensor network data and template guided description data belonging to different sensor data-description combinations;
[0036] generating a second network training cost according to a feature distance between the template IoT sensor network data and the template guided description data belonging to the same sensor data-description combination, and a feature distance between the template IoT sensor network data and the template guided description data belonging to different template IoT sensor network data pairs;
[0037] The second graph network model and the first graph network model are trained according to the second network training cost until model convergence requirements are met.
[0038] In a possible implementation of the first aspect, the method further includes:
[0039] Obtain target IoT sensor network data;
[0040] Performing graph self-encoding processing on the target IoT sensor network data based on the first graph network model in the building environment state prediction network to generate a target IoT sensor network relationship graph vector;
[0041] A context generation unit in the building environment state prediction network generates target context features corresponding to a plurality of guide knowledge points respectively according to the target IoT sensor network relationship graph vector, wherein the plurality of guide knowledge points are used to describe the building environment state in the target IoT sensor network data from a plurality of guide description fields;
[0042] Based on the second graph network model in the building environment state prediction network, according to the basic context features corresponding to the multiple guiding knowledge points and the target context features corresponding to the multiple guiding knowledge points, the target cross-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels are generated;
[0043] Calculate the characteristic distance between each target cross-correlation feature and the target IoT sensor network relationship graph vector, and determine the abnormal path prediction data of the building environment corresponding to each guiding knowledge point under each building feature label according to the characteristic distance between the target cross-correlation feature and the target IoT sensor network relationship graph vector;
[0044] The abnormal building environment path prediction data of each guiding knowledge point under the building feature label are fused to generate the abnormal building environment path of the target IoT sensor network data under the building feature label.
[0045] According to the second aspect of the present application, a smart building-based Internet of Things sensor data processing system is provided, wherein the smart building-based Internet of Things sensor data processing system comprises a machine-readable storage medium and a processor, wherein the machine-readable storage medium stores machine-executable instructions, and when the processor executes the machine-executable instructions, the smart building-based Internet of Things sensor data processing system implements the aforementioned smart building-based Internet of Things sensor data processing method.
[0046] According to the third aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed, the aforementioned method for processing IoT sensor data based on smart buildings is implemented.
[0047] According to any one of the above aspects, the technical effect of the present application is:
[0048] The embodiment of the present application realizes efficient and intelligent processing and analysis of IoT sensor network data by integrating graph network model, context generation unit and dynamic guidance knowledge points, and significantly improves the accuracy and real-time performance of building environment state prediction. Specifically, by introducing graph self-encoding processing technology, the sample IoT sensor network data is deeply mined, and a high-dimensional sensor network relationship graph vector rich in relationship information is generated, which provides a richer and more accurate data basis for subsequent state prediction. At the same time, combined with the context generation unit and multiple guidance knowledge points, the characterization of the building environment state is further refined, and the accuracy of the prediction model is effectively improved. By using the dynamically updated guidance knowledge points, the model can automatically adjust the recognition ability of the environmental state under different building feature labels according to the feedback in the learning process. This adaptive mechanism ensures that the model can maintain a high degree of accuracy and robustness when processing sensor data in different building areas and different environmental conditions. Through the efficient computing power of the graph network model, the processing and analysis of large-scale IoT sensor network data can be completed in a short time, meeting the needs of smart buildings for real-time monitoring and rapid response. In addition, the optimized network parameters and guidance knowledge points also further improve the overall efficiency of data processing. Moreover, it not only focuses on the abnormal detection of a single environmental parameter, but also realizes a comprehensive assessment of the building environment status through the comprehensive role of multiple guiding knowledge points. This helps managers to more intuitively understand the overall situation of the building's internal environment and take corresponding measures to optimize the environmental quality in a timely manner. As a result, it provides strong technical support for the intelligent management of smart buildings and promotes the deepening of the application of Internet of Things technology in the construction field. Through continuous learning and optimization, it will bring more intelligent and efficient environmental monitoring and control solutions to smart buildings. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0050] Figure 1 A flowchart of a method for processing IoT sensor data based on smart buildings provided in an embodiment of the present application.
[0051] Figure 2 A schematic diagram of the component structure of an Internet of Things sensor data processing system based on a smart building provided in an embodiment of the present application is shown. DETAILED DESCRIPTION
[0052] In order to make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below according to the drawings in the embodiment of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description, and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the embodiment of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add multiple other operations to the flowchart under the guidance of the content of the present application, and can also destroy multiple operations from the flowchart.
[0053] Figure 1 The flowchart of the method and system for processing sensor data of the Internet of Things based on smart buildings provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps of the method for processing sensor data of the Internet of Things based on smart buildings in this embodiment can be shared with each other according to actual needs, or some steps can be omitted or maintained. The detailed steps of the method for processing sensor data of the Internet of Things based on smart buildings include:
[0054] Step S110, obtaining example learning data, wherein the example learning data includes a plurality of example IoT sensor network data, wherein the example IoT sensor network data has an example building environment abnormal path under one or more building feature tags among a plurality of building feature tags.
[0055] In detail, the example IoT sensor network data refers to a set of IoT sensor data collected to train or verify a certain building environment state prediction model, which comes from various sensors deployed inside the building, including but not limited to temperature sensors, humidity sensors, air pressure sensors, light sensors, air quality sensors, etc. Each sensor periodically collects environmental parameters and transmits these data to the data center. The example IoT sensor network data not only contains the readings of each sensor, but may also include metadata such as the location information of the sensor and the collection time. These data are the basis for the model to learn the building environment state.
[0056] The building feature labels are used to distinguish different areas or functions inside the building. These labels help the model understand and identify environmental characteristics and abnormal patterns in different areas. For example, in a large hospital, the building feature labels may include inpatient department, outpatient building, operating room, pharmacy, cafeteria, etc. Each label corresponds to one or more specific areas in the building, with unique environmental requirements and possible environmental anomalies.
[0057] The example building environment anomaly paths are a series of sensor reading changes observed in the example IoT sensor network data that indicate anomalies in the building environment. These anomaly paths may be caused by some kind of failure or external event, such as equipment failure, environmental changes, etc. The identification of anomaly paths helps the model learn how to detect and predict future environmental anomalies.
[0058] In this embodiment, taking the building environment data of a large comprehensive hospital as an example, the example learning data includes multiple example IoT sensor network data of various areas of the hospital.
[0059] The hospital is divided into multiple building feature label areas, such as the inpatient department, outpatient building, operating room, pharmacy, cafeteria, etc. In the example IoT sensor network data of the inpatient department, the following example building environment abnormal path was found: in a certain time period, the temperature sensors of multiple wards showed that the temperature was continuously lower than the set comfort range, while the humidity sensor detected that the humidity was high, and the wind speed sensor data of the ventilation system indicated that the ventilation volume was insufficient. In addition, the air quality sensors in some wards detected that the carbon dioxide concentration and PM2.5 concentration exceeded the standard, while these indicators in the adjacent wards were relatively normal. This series of data constitutes the building environment abnormal path of the inpatient department in a specific time period.
[0060] In the sample data of the outpatient building, different abnormal paths were found. In the registration area, the flow sensor showed that there were too many people in line, resulting in too high a density of people, and the temperature adjustment of the air conditioning system failed to keep up, causing the temperature in the area to rise. At the same time, the brightness sensor of the lighting system detected uneven brightness, with some areas too bright and some areas too dark. In the consulting room area, the sound sensor detected that the noise level was beyond the normal range, which may be due to multiple consulting rooms conducting diagnosis and treatment activities at the same time and poor sound insulation.
[0061] In the example data of the operating room, it was found that during the operation, the pressure sensor in the room showed that the pressure fluctuations exceeded the allowable range, and the control of temperature and humidity also became unstable, which may affect the sterile environment of the operation and the comfort of the patient.
[0062] In the sample data from the pharmacy, it was found that the temperature sensors in the refrigerated medicine storage area recorded temperatures outside the specified low temperature range many times, which may affect the quality and efficacy of the medicines.
[0063] In the example data of the cafeteria, the smoke sensors in the cooking area frequently triggered alarms, and the speed of the exhaust fans failed to increase accordingly, resulting in the smoke not being discharged in time, affecting the indoor air quality.
[0064] Step S120, performing graph self-encoding processing on the example Internet of Things sensor network data based on the first graph network model in the building environment state prediction network to generate an example Internet of Things sensor network relationship graph vector.
[0065] In detail, the example IoT sensor network relationship graph vector is a high-dimensional vector representing the relationship between sensors obtained by performing graph autoencoding on the example IoT sensor network data. This vector captures the spatial position relationship, data dependency, and potential collaborative change patterns between sensors, providing rich relationship information for subsequent model learning and prediction.
[0066] In this embodiment, the first graph network model of the building environment state prediction network begins to process these complex example IoT sensor network data. Taking the data of the inpatient department as an example, the first graph network model regards the sensors of each ward (such as temperature sensors, humidity sensors, air quality sensors, etc.) as nodes in the graph, and the associations between sensors (such as sensors in adjacent wards, sensors controlled by the same ventilation system, etc.) as edges.
[0067] The first graph network model analyzes the time series relationships and mutual influences between various sensor data. For example, when the temperature in a ward drops, the temperature in the adjacent ward may also be affected to a certain extent; changes in ventilation volume may affect the air quality and humidity of multiple related wards at the same time. Through in-depth analysis and encoding of these relationships, the first graph network model transforms the example IoT sensor network data of the inpatient department into a highly abstract example IoT sensor network relationship graph vector.
[0068] For the data of the outpatient building, the first graph network model will consider the relationship between the flow sensor and the air conditioning system and lighting system sensors, as well as the association between the sound sensors in different consulting rooms. In the data processing of the operating room, the focus will be on the coordinated change relationship between the pressure sensor, temperature sensor and humidity sensor. The data of the pharmacy and cafeteria will also be processed by graph autoencoding in a similar way to generate the corresponding example IoT sensor network relationship graph vector.
[0069] Step S130, based on the context generation unit in the building environment state prediction network, generates example context features corresponding to multiple guiding knowledge points respectively according to the example Internet of Things sensor network relationship graph vector, and based on the second graph network model in the building environment state prediction network, generates example cross-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels according to the basic context features corresponding to the multiple guiding knowledge points and the example context features corresponding to the multiple guiding knowledge points.
[0070] In detail, guiding knowledge points refer to a series of knowledge units or concepts that play a guiding role in the process of building environment state prediction. These guiding knowledge points may be based on the experience summary of domain experts, the statistical laws of historical data, or the preliminary learning results of the model. Guiding knowledge points help the model quickly locate key information in a complex data environment and improve the accuracy and efficiency of prediction.
[0071] The example context features refer to feature representations related to the guided knowledge points generated based on the example IoT sensor network relationship graph vector. These example context features not only contain the information of the sensor data itself, but also incorporate the context information related to the guided knowledge points, which helps the model to more comprehensively understand the meaning and patterns behind the data.
[0072] The example cross-correlation features refer to features that are generated based on the basic context features and example context features of the guiding knowledge points under multiple building feature labels, and represent the correlation between knowledge points. These example cross-correlation features reflect the coordinated changes or mutual influences of different knowledge points in the same building environment state, which helps the model to more accurately predict the overall state of the building environment.
[0073] In this embodiment, after receiving the example IoT sensor network relationship graph vector generated from the first graph network model, the context generation unit starts to generate example context features for multiple guide knowledge points. Assume that the guide knowledge points include "coordinated change of temperature and humidity", "ventilation effect evaluation", "association between human flow and environmental comfort", "influence of lighting uniformity", "sound noise control", "pressure stability analysis", etc.
[0074] For the inpatient department, for the guiding knowledge point of "coordinated changes in temperature and humidity", the context generation unit analyzes the relationship graph vector to derive the specific situation and change trend of the increase in humidity while the temperature drops, and generates corresponding example context features.
[0075] The second graph network model obtains the basic context features of each guiding knowledge point (such as typical features learned from a large amount of historical data) and the newly generated example context features.
[0076] Taking the operating room as an example, for the two guiding knowledge points of "pressure stability analysis" and "temperature and humidity control effect", the second graph network model analyzes the relationship between them. If the control of temperature and humidity also shows obvious abnormalities when the pressure fluctuates, then an example cross-correlation feature reflecting this complex relationship will be generated.
[0077] For the guiding knowledge points of "association between human flow and environmental comfort" and "impact of lighting uniformity" in the outpatient building, the model will generate example cross-correlation features that reflect the decrease in environmental comfort due to excessive human flow and the impact of uneven lighting on the experience of patients and medical staff.
[0078] Step S140, determining a first network training cost based on the example inter-correlation features of the multiple guiding knowledge points under each of the multiple building feature tags, the example Internet of Things sensor network relationship graph vector, and the example building environment abnormal path of the example Internet of Things sensor network data under one or more building feature tags.
[0079] In this embodiment, the server starts to determine the first network training cost. First, the vector distance between the example cross-correlation feature of each guiding knowledge point under each building feature label and the example Internet of Things sensor network relationship graph vector is calculated.
[0080] Taking the guiding knowledge point of “Coordinated changes in temperature and humidity” in the inpatient department as an example, the difference between its example cross-correlation features and the relationship graph vector in reflecting the relationship between temperature and humidity is calculated to obtain the vector distance.
[0081] Then, according to the vector distances corresponding to each guiding knowledge point under the same building feature label, the predicted building environment abnormal path of the example IoT sensor network data under the building feature label is determined.
[0082] For the inpatient department, if the vector distances of multiple guiding knowledge points (such as “coordinated changes in temperature and humidity” and “ventilation effectiveness evaluation”) collectively indicate that there are problems with the environmental conditioning system, it is predicted that there may be abnormal paths caused by blockage in the ventilation ducts or failure of the air-conditioning equipment.
[0083] The predicted building environment abnormal path is compared with the actual labeled example building environment abnormal path (such as the abnormalities in the temperature, humidity, ventilation, etc. of the inpatient department obtained previously). This difference is calculated through a loss function (such as a mean square error function) to determine the first network training cost.
[0084] Step S150, updating the network parameter information of the building environment state prediction network and the multiple guiding knowledge points according to the first network training cost, and the updated guiding knowledge points are used to express the building environment state in the Internet of Things sensor network data from multiple guiding description fields.
[0085] In this embodiment, the server updates the building environment state prediction network according to the calculated first network training cost.
[0086] For example, if the judgment of ventilation problems in the prediction of the inpatient department is inaccurate, resulting in a high training cost, the server will adjust the sensor data processing weights related to ventilation, or optimize certain connection parameters in the first graph network model and the second graph network model.
[0087] At the same time, multiple guiding knowledge points are updated. The definition and scope of the guiding knowledge point "ventilation effect evaluation" may be further refined, or new guiding knowledge points that are more closely related to air quality monitoring may be added to enable the network to more accurately capture and describe abnormal conditions in the building environment.
[0088] In subsequent data analysis, the updated guidance knowledge points can more effectively express the building environment status in the IoT sensor network data from multiple guidance description fields, providing more accurate support for the hospital's environmental management and optimization.
[0089] Based on the above steps, the embodiment of the present application realizes efficient and intelligent processing and analysis of IoT sensor network data by integrating graph network model, context generation unit and dynamic guidance knowledge points, and significantly improves the accuracy and real-time prediction of building environment state. Specifically, by introducing graph self-encoding processing technology, the sample IoT sensor network data is deeply mined, and a high-dimensional sensor network relationship graph vector rich in relationship information is generated, which provides a richer and more accurate data basis for subsequent state prediction. At the same time, combined with the context generation unit and multiple guidance knowledge points, the characterization of the building environment state is further refined, and the accuracy of the prediction model is effectively improved. By using the dynamically updated guidance knowledge points, the model can automatically adjust the recognition ability of the environmental state under different building feature labels according to the feedback in the learning process. This adaptive mechanism ensures that the model can maintain a high degree of accuracy and robustness when processing sensor data in different building areas and different environmental conditions. Through the efficient computing power of the graph network model, the processing and analysis of large-scale IoT sensor network data can be completed in a short time, meeting the needs of smart buildings for real-time monitoring and rapid response. In addition, the optimized network parameters and guiding knowledge points also further improve the overall efficiency of data processing. Moreover, it not only focuses on the abnormal detection of a single environmental parameter, but also realizes a comprehensive assessment of the building environment status through the comprehensive effect of multiple guiding knowledge points. This helps managers to more intuitively understand the overall situation of the building's internal environment and take corresponding measures to optimize the environmental quality in a timely manner. As a result, it provides strong technical support for the intelligent management of smart buildings and promotes the deepening of the application of Internet of Things technology in the construction field. Through continuous learning and optimization, it will bring more intelligent and efficient environmental monitoring and control solutions to smart buildings.
[0090] In a possible implementation, step S130 includes:
[0091] Step S131 , performing cross-correlation analysis on the basic context features of each of the guiding knowledge points and the example context features corresponding to the guiding knowledge points, to generate mutual integration context features corresponding to each of the guiding knowledge points.
[0092] Step S132, based on the second graph network model of the building environment state prediction network, graph self-encoding processing is performed on the mutual fusion context features corresponding to each of the guiding knowledge points to generate example cross-correlation features of each guiding knowledge point under each of the multiple building feature labels.
[0093] In this embodiment, it is further assumed that the server is processing the building environment data of the large comprehensive hospital mentioned above.
[0094] For the guiding knowledge point of "coordinated changes in temperature and humidity" in the inpatient department, the server first obtains its basic context features, such as the general change pattern learned from a large amount of historical inpatient temperature and humidity data. At the same time, it obtains the example context features generated for this guiding knowledge point in the currently processed inpatient department example data, that is, the specific situation and change trend of temperature drop and humidity increase in a specific time period.
[0095] The server performs a cross-correlation analysis on these two features. Through complex calculations and comparisons, it is found that the temperature and humidity in the basic context usually change relatively smoothly, while there is a clear reverse change in the example context. This difference and correlation are comprehensively considered to generate the mutual integration context feature corresponding to the guiding knowledge point "coordinated change of temperature and humidity", which may highlight the abnormal degree and characteristics of the temperature and humidity changes in the current example.
[0096] Next, the second graph network model of the building environment state prediction network performs graph autoencoding on this mutual context feature.
[0097] In the graph structure, each element in the mutual context feature is regarded as a node, and the relationship between them (such as the relationship between the temperature change amplitude and the humidity change amplitude) is regarded as an edge. The model analyzes the relationship between these nodes and edges and performs encoding and compression.
[0098] For example, it is found that when the temperature drops by a large amount and the humidity increases by a small amount, it may be related to a specific ventilation system failure mode. After encoding, an example cross-correlation feature of the "temperature and humidity co-variation" guiding knowledge point under the building feature label of the inpatient department is generated. This feature may represent the special pattern of temperature and humidity co-variation in the current inpatient department environment and its potential association with other factors in a highly abstract and concise way.
[0099] For the guiding knowledge point of "the relationship between human flow and environmental comfort" in the outpatient building, the basic context features may include the common impact patterns of human flow changes on environmental comfort under normal circumstances. The example context features reflect the specific situation that the excessive human flow in the registration area of the current outpatient building leads to increased temperature and decreased environmental comfort.
[0100] The server performs cross-correlation analysis and finds that the flow of people this time exceeds the scale commonly seen in the basic context, which has a more serious impact on environmental comfort. The generated cross-context features highlight the relationship between this abnormal flow of people and changes in environmental comfort.
[0101] The second graph network model performs graph self-encoding on the mutual context features. In the graph, the size of the flow of people, the increase in temperature, the decrease in environmental comfort, etc. may be used as nodes, and the associated edges between them are established. Through encoding, the example mutual correlation features of the guiding knowledge point of "the association between the flow of people and the environmental comfort" in the outpatient building are obtained, which may reveal the complex relationship between the flow of people and the environmental comfort in this area that is different from the usual one.
[0102] For the "pressure stability analysis" guidance knowledge point in the operating room, the basic context features include the typical patterns of pressure stability during normal surgery. The example context features show the specific situation in which the pressure fluctuation exceeds the allowable range during the current surgery.
[0103] The server finds the differences and connections between the basis and the examples through cross-correlation analysis, generates mutual context features, and highlights the abnormal characteristics of pressure fluctuations. The second graph network model performs graph autoencoding on it, establishes the association of nodes such as pressure fluctuation values, temperature and humidity changes in the graph structure, and finally generates example cross-correlation features, reflecting the specific relationship between abnormal pressure in the operating room and other environmental factors.
[0104] For the pharmacy's "temperature control of refrigerated drug storage" guidance knowledge point, the basic context features cover the temperature standards and common fluctuation ranges for refrigerated drug storage under normal circumstances. The example context features show that the current refrigerated area temperature has exceeded the specified low temperature range multiple times.
[0105] The server performs cross-correlation analysis to generate mutual context features, emphasizing the current abnormal temperature situation. The second graph network model encodes it, associates nodes such as temperature sensor data and drug storage locations in the graph, and generates example cross-correlation features, indicating possible refrigeration system problems or unreasonable drug storage layout.
[0106] For the cafeteria's "smoke emission and air quality" guidance knowledge point, the basic context features include the general relationship between smoke emission and air quality during normal cooking. The example context features reflect the current situation where the smoke sensor frequently alarms and the exhaust fan speed is insufficient, resulting in a decrease in air quality.
[0107] The server performs cross-correlation analysis to generate mutual context features, highlighting the severity of smoke emissions and air quality deterioration. The second graph network model performs graph autoencoding processing on it, associates nodes such as smoke concentration, exhaust fan speed, and air quality indicators in the graph, generates example cross-correlation features, and reveals the failure mode of the cafeteria's smoke emission system.
[0108] Through the above processing of each guiding knowledge point under each building feature label, the server can generate comprehensive and targeted example cross-correlation features, which provides a key basis for the subsequent accurate evaluation of the building environment status and determination of the training cost.
[0109] In a possible implementation, the method further includes:
[0110] Step A110 , determining a deviation cost according to a feature distance between every two example mutual correlation features among example mutual correlation features corresponding to a plurality of guiding knowledge points under the same building feature label.
[0111] Step A120: updating the guiding knowledge point according to the deviation cost.
[0112] In this embodiment, it is further assumed that the server is processing the building environment data of the aforementioned large comprehensive hospital.
[0113] Taking the inpatient department as an example, the server has obtained example cross-correlation features corresponding to multiple guiding knowledge points such as "coordinated changes in temperature and humidity" and "ventilation effect evaluation".
[0114] For the example cross-correlation feature of the guiding knowledge point "Temperature and humidity co-variation", it can be expressed as a series of numerical vectors about the relationship between temperature and humidity changes, such as [0.5, -0.3, 0.2, -0.1]. The example cross-correlation feature of the guiding knowledge point "Ventilation effect evaluation" may be expressed as [0.3, 0.1, -0.2, 0.4].
[0115] The server starts calculating the feature distance between the two example cross-correlation features. It may use Euclidean distance, cosine distance, or other suitable distance metric.
[0116] Similarly, the server calculates the characteristic distances between other guiding knowledge points in the inpatient department, such as "air quality monitoring" and "coordinated changes in temperature and humidity", "air quality monitoring" and "ventilation effect evaluation", etc.
[0117] Then, the server accumulates these calculated feature distances. Assuming that the calculated feature distances are 0.61, 0.55, and 0.48, the server accumulates them:
[0118] Deviation cost = 0.61 + 0.55 + 0.48;
[0119] After obtaining the total deviation cost, if the deviation cost is large, it means that the correlation and consistency between the guiding knowledge points are poor, and the guiding knowledge points may need to be updated.
[0120] For example, if it is found that the deviation cost of the guidance knowledge point "Ventilation effect evaluation" from other guidance knowledge points is large, the server may review and adjust the definition and calculation method of this guidance knowledge point. Perhaps some factors more directly related to temperature, humidity or air quality will be added, or the weight distribution will be adjusted to make it more closely related to other guidance knowledge points and more reasonable.
[0121] For other building feature label areas such as outpatient buildings, operating rooms, pharmacies, and cafeterias, the server also calculates the feature distances between the guiding knowledge points in the same way, determines the deviation costs, and updates the guiding knowledge points based on the deviation costs to improve the accuracy and consistency of the building environment status prediction.
[0122] In a possible implementation, step A110 includes:
[0123] Step A111 , calculating the feature distance between every two example cross-correlation features among the example cross-correlation features corresponding to multiple guiding knowledge points under the same building feature label.
[0124] Step A112, accumulating the feature distances to generate a deviation cost.
[0125] Let's continue to assume that the server is still processing the building environment data of the aforementioned large comprehensive hospital, taking the inpatient department as an example.
[0126] The server has obtained example cross-correlation features corresponding to multiple guiding knowledge points of the inpatient department. For example, the example cross-correlation feature of "coordinated changes in temperature and humidity" is the vector [0.2, -0.3, 0.1, 0.4], the example cross-correlation feature of "ventilation effect evaluation" is the vector [0.1, 0.2, -0.2, 0.3], and the example cross-correlation feature of "air quality monitoring" is the vector [0.3, -0.1, 0.2, -0.2].
[0127] The server first calculates the feature distance between the two example cross-correlation features of “temperature and humidity co-variation” and “ventilation effect evaluation”.
[0128] Next, the server calculates the characteristic distance between “temperature and humidity coordinated changes” and “air quality monitoring”, and then the server calculates the characteristic distance between “ventilation effect evaluation” and “air quality monitoring”.
[0129] After calculating the feature distance between each two example cross-correlation features, the server accumulates these feature distances.
[0130] For example, in the above example, deviation cost = d1 + d2 + d3 = 0.6 + 0.648 + 0.735 = 1.983
[0131] Through such calculation and accumulation, the server obtains the deviation cost of the guiding knowledge points of the inpatient department. For other building feature label areas, such as outpatient building, operating room, pharmacy and cafeteria, the server will also follow the same steps to calculate to determine the deviation cost of the guiding knowledge points in each area.
[0132] In a possible implementation, step S140 includes:
[0133] Step S141, calculating the vector distance between the example mutual correlation features of each of the guiding knowledge points under each building feature label and the example Internet of Things sensor network relationship graph vector.
[0134] Step S142, determining the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label based on the vector distance between the example mutual correlation features corresponding to each guiding knowledge point under the same building feature label and the example Internet of Things sensor network relationship graph vector.
[0135] Step S143, determining a first network training cost according to the example building environment abnormal path of the example Internet of Things sensor network data under one or more building feature tags and the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature tag.
[0136] First, taking the inpatient department as an example, the server has obtained example cross-correlation features of multiple guiding knowledge points (such as "coordinated changes in temperature and humidity", "ventilation effect evaluation", etc.) under the building feature label of the inpatient department, as well as an example IoT sensor network relationship graph vector of the inpatient department.
[0137] The server starts to calculate the vector distance between the example cross-correlation feature of each guide knowledge point and the example IoT sensor network relationship graph vector. For example, the example cross-correlation feature of the guide knowledge point "Temperature and humidity co-variation" is the vector [0.2, -0.3, 0.1, 0.4], and the example IoT sensor network relationship graph vector is [0.1, -0.2, 0.3, 0.5].
[0138] Similarly, the server calculates the vector distance between the example cross-correlation features of other guided knowledge points such as "ventilation effect evaluation" and the example IoT sensor network relationship graph vector.
[0139] Next, the server determines the predicted building environment abnormal path of the example IoT sensor network data under the inpatient department based on the vector distance between the example cross-correlation features corresponding to each guiding knowledge point under the same building feature label (inpatient department) and the example IoT sensor network relationship graph vector.
[0140] Assume that the server sets some thresholds and rules for judgment. If the vector distances of multiple guiding knowledge points are large and exceed the preset threshold, it may indicate that there are serious environmental anomalies in the inpatient department. For example, it is predicted that the overall failure of the ventilation system may cause abnormal temperature, humidity and air quality.
[0141] Then, the server compares the actual example building environment abnormal path of the example IoT sensor network data under the building feature label of the inpatient department (for example, the abnormal conditions of temperature, humidity, ventilation, etc. of multiple wards obtained before) and the building environment abnormal path just predicted.
[0142] Assume that the actual example building environment abnormal path indicates that the environmental abnormality in a specific area is caused by the blockage of some ventilation ducts, while the prediction is that the ventilation system is faulty as a whole. The server calculates this difference through a loss function (such as a mean square error function) to determine the first network training cost.
[0143] For other building feature label areas such as outpatient buildings, operating rooms, pharmacies, and cafeterias, the server also performs calculations and comparisons according to the same steps, determines the predicted building environment abnormal path for each area, and compares it with the actual example building environment abnormal path to calculate the corresponding first network training cost.
[0144] Through this process, the performance of the building environment state prediction network in each building feature label area can be comprehensively evaluated, and the accurate first network training cost can be determined, providing a basis for subsequent network optimization and improvement.
[0145] In a possible implementation, step S142 includes:
[0146] Step S1421, for each building feature label, the vector distance between the example mutual correlation feature corresponding to each guiding knowledge point and the example Internet of Things sensor network relationship graph vector is normalized, and a target weight is assigned to each guiding knowledge point according to the normalized vector distance. The size of the target weight is inversely proportional to the vector distance. The calculation formula of the target weight is: target weight = (1-normalized vector distance)^α, where α is an adjustment factor used to adjust the sensitivity of weight allocation.
[0147] Step S1422, after obtaining the target weight of each guiding knowledge point, weighted fusion is performed on the example cross-correlation features of each guiding knowledge point under the building feature label, and the weighted fusion features are input into the regression network in the building environment state prediction network for feature mapping, so as to obtain the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label.
[0148] In this embodiment, the server first calculates the vector distances between the example cross-correlation features corresponding to several guiding knowledge points in the inpatient department (such as "temperature and humidity coordinated changes", "ventilation effect evaluation", and "air quality monitoring") and the example IoT sensor network relationship graph vector. Assume that these distances are: d1=0.3, d2=0.4, and d3=0.2.
[0149] Next, the server normalizes these vector distances. Assuming the sum of the total distances is D=0.3+0.4+0.2=0.9, the normalized distances are:
[0150] d1'=0.3 / 0.9≈0.33
[0151] d2'=0.4 / 0.9≈0.44
[0152] d3'=0.2 / 0.9≈0.22
[0153] Assuming the adjustment factor α=2, assign a target weight to each guided knowledge point according to the target weight calculation formula:
[0154] For "co-variation of temperature and humidity", weight w1=(1-0.33)^2≈0.44
[0155] For "ventilation effect evaluation", weight w2=(1-0.44)^2≈0.31
[0156] For "air quality monitoring", weight w3=(1-0.22)^2≈0.60
[0157] Then, the server weights and fuses the example cross-correlation features of each guiding knowledge point under the label of inpatient department. Assume that the example cross-correlation features of "Temperature and humidity coordinated change" are [0.2, -0.1, 0.3, 0.4], "Ventilation effect evaluation" are [0.1, 0.2, -0.3, 0.2], and "Air quality monitoring" are [0.3, -0.2, 0.1, -0.1].
[0158] The features after weighted fusion are:
[0159] [0.2×0.44+0.1×0.31+0.3×0.60, -0.1×0.44+0.2×0.31-0.2×0.60, 0.3×0.44-0.3×0.31+0.1×0.60, 0.4×0.44+0.2×0.31-0.1×0.60]
[0160] The server inputs the weighted fusion features into the regression network in the building environment state prediction network for feature mapping. Through complex calculations and mapping, the regression network finally obtains the predicted building environment abnormality path under the building feature label of the inpatient department. For example, it predicts that the abnormal temperature and humidity in some wards may be caused by the failure of some ventilation equipment.
[0161] For other building feature label areas in the hospital, such as outpatient buildings, operating rooms, pharmacies, cafeterias, etc., the server also follows the same steps to obtain the predicted building environment abnormal paths for each area.
[0162] In a possible implementation, step S143 includes: calculating a loss function value based on abnormal building environment paths of example IoT sensor network data under one or more building feature labels and predicted abnormal building environment paths of example IoT sensor network data under each building feature label to generate a first network training cost.
[0163] In a possible implementation, step S150 includes: updating the network parameter information and the guiding knowledge points of the context generation unit according to the first network training cost.
[0164] In this embodiment, the server has obtained the actual example building environment abnormal paths of the example IoT sensor network data under these building feature tags, such as the actual situation records of poor ventilation, abnormal temperature and humidity in some wards of the inpatient department; at the same time, the predicted building environment abnormal paths under each building feature tag are also obtained. For example, the prediction for the inpatient department is that the air-conditioning system failure in a specific area causes temperature imbalance.
[0165] The server calculates the loss function value based on these actual example building environment abnormal paths and predicted building environment abnormal paths. It is assumed that the mean square error (MSE) is used as the loss function.
[0166] For the inpatient department, the actual abnormal path is represented by the vector [1, 0, 1, 0] (where 1 indicates abnormality and 0 indicates normality), and the predicted abnormal path is represented by [0.8, 0.1, 0.9, 0.2].
[0167] Then, the mean square error between the abnormal building environment path of the example IoT sensor network data under one or more building feature labels and the predicted abnormal building environment path of the example IoT sensor network data under each building feature label can be calculated as the loss function value, and similar calculations are performed for other areas such as outpatient buildings and operating rooms. The loss values of each area are combined to obtain the first network training cost.
[0168] Next, the server updates the building environment state prediction network based on the first network training cost. Specifically, the server updates the network parameter information and guiding knowledge points of the context generation unit.
[0169] For example, if the prediction results of the inpatient department have large deviations, resulting in high training costs, the server will adjust the parameter weights related to temperature, humidity, and ventilation in the context generation unit so that it can generate context features more accurately in subsequent processing.
[0170] At the same time, for guiding knowledge points, such as redefining the specific calculation method and considerations of "ventilation effect evaluation", or adding guiding knowledge points that are more relevant to air-conditioning system fault detection, the accuracy of the prediction can be improved.
[0171] In this way, the building environment status prediction network is continuously optimized so that it can more accurately predict abnormal situations in the building environment.
[0172] In a possible implementation, before step S120, the method further includes:
[0173] Step B110, obtaining a plurality of sensor data-description combinations, wherein the sensor data-description combinations include template Internet of Things sensor network data and template guided description data.
[0174] Step B120, loading the multiple template-guided description data into the second graph network model, and loading the multiple template IoT sensor network data into the first graph network model, performing comparative training, and generating feature distances between template IoT sensor network data and template-guided description data belonging to the same sensor data-description combination, and feature distances between template IoT sensor network data and template-guided description data belonging to different sensor data-description combinations.
[0175] Step B130, generating a second network training cost based on the feature distance between the template IoT sensor network data and the template guided description data belonging to the same sensor data-description combination, and the feature distance between the template IoT sensor network data and the template guided description data belonging to different template IoT sensor network data pairs.
[0176] Step B140: training the second graph network model and the first graph network model according to the second network training cost until the model convergence requirement is met.
[0177] In this embodiment, before the server performs graph self-encoding processing on the example Internet of Things sensor network data based on the first graph network model of the building environment state prediction network, it first performs the following operations.
[0178] The server obtains multiple sensor data-description combinations. For example, one set of sensor data-description combinations is about the situation of the hospital inpatient department at a certain moment. The template IoT sensor network data includes the measurement values of the temperature, humidity, ventilation volume and other sensors of each ward at that moment, while the template guided description data is a text description of the environmental status at that moment, such as "the temperature in some wards is slightly low, the overall ventilation is good, and the humidity is moderate."
[0179] Then, the server loads a plurality of such template guided description data into the second graph network model, and loads the corresponding template IoT sensor network data into the first graph network model for comparative training.
[0180] For the cases belonging to the same sensor data-description combination, the server calculates the characteristic distance between them. For example, for the above inpatient department combination, the server analyzes the difference between the temperature distribution characteristics in the template IoT sensor network data and the description of temperature in the template guided description data to obtain a characteristic distance value. Similarly, similar comparisons and distance calculations are performed for other aspects such as humidity and ventilation.
[0181] For situations belonging to different sensor data-description combinations, the server also performs similar processing. Assuming that there is another group of situations about the hospital outpatient building at a certain moment, the server calculates the feature distance between the template IoT sensor network data of the inpatient department and the template guided description data of the outpatient building.
[0182] Next, the server generates a second network training cost based on the feature distance between the template IoT sensor network data and the template guided description data belonging to the same sensor data-description combination, and the feature distance between the template IoT sensor network data and the template guided description data belonging to different template IoT sensor network data pairs.
[0183] Assume that the server sets certain weights and rules to integrate these distances and calculate the value of the second network training cost.
[0184] Afterwards, the server trains the second graph network model and the first graph network model based on the second network training cost. During the training process, the parameters of the model are continuously adjusted, such as the weight distribution of sensor data in the first graph network model and the processing method of describing data in the second graph network model.
[0185] The server continues this training process and continuously evaluates the performance of the model until the model meets the convergence requirements. For example, when the value of the second network training cost no longer decreases significantly after multiple consecutive trainings, or the model can give relatively accurate and stable results for new test data, the server considers that the model has converged and can be used for subsequent processing of the example IoT sensor network data.
[0186] In a possible implementation, the method further includes:
[0187] Step C110, obtaining target IoT sensor network data.
[0188] Step C120, performing graph self-encoding processing on the target Internet of Things sensor network data based on the first graph network model in the building environment state prediction network to generate a target Internet of Things sensor network relationship graph vector.
[0189] Step C130, based on the context generation unit in the building environment state prediction network, generates target context features corresponding to multiple guidance knowledge points according to the target Internet of Things sensor network relationship graph vector, and the multiple guidance knowledge points are used to describe the building environment state in the target Internet of Things sensor network data from multiple guidance description fields.
[0190] Step C140, based on the second graph network model in the building environment state prediction network, according to the basic context features corresponding to the multiple guiding knowledge points and the target context features corresponding to the multiple guiding knowledge points, generate the target inter-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels.
[0191] Step C150, calculate the characteristic distance between each target mutual correlation feature and the target Internet of Things sensor network relationship graph vector, and determine the abnormal path prediction data of the building environment corresponding to each guiding knowledge point under each building feature label based on the characteristic distance between the target mutual correlation feature and the target Internet of Things sensor network relationship graph vector.
[0192] Step C160, the building environment abnormal path prediction data of each guiding knowledge point under the building feature label are integrated to generate the building environment abnormal path of the target Internet of Things sensor network data under the building feature label.
[0193] In this embodiment, the server obtains the target IoT sensor network data during operation. Assume that this is the real-time data of a newly built hospital, including sensor information in various areas.
[0194] The server uses the first graph network model in the building environment state prediction network to perform graph self-encoding processing on these target IoT sensor network data. For example, the sensor data of temperature, humidity, air quality, etc. in each ward, consulting room, corridor, etc. are regarded as nodes, and the associations between them are regarded as edges. Through the analysis and encoding of the model, the target IoT sensor network relationship graph vector is generated.
[0195] Next, the context generation unit in the building environment state prediction network generates target context features for multiple guiding knowledge points based on the target IoT sensor network relationship graph vector. Assuming that the guiding knowledge points include "temperature change trend", "humidity fluctuation range", "air quality level", etc., the context generation unit analyzes the specific conditions of these aspects in the current data based on the graph vector and generates corresponding target context features.
[0196] Then, the second graph network model in the building environment state prediction network generates target inter-correlation features of multiple guiding knowledge points under various building feature labels (such as inpatient department, outpatient building, etc.) based on the basic context features (typical features learned from a large amount of previous data) corresponding to multiple guiding knowledge points and the newly generated target context features. For example, under the label of inpatient department, the relationship between "temperature change trend" and "humidity fluctuation range" is analyzed.
[0197] After that, the server calculates the feature distance between each target cross-correlation feature and the target IoT sensor network relationship graph vector. Taking the "temperature change trend" guidance knowledge point of the inpatient department as an example, the feature distance between its target cross-correlation feature and the graph vector reflecting the temperature change is calculated.
[0198] For the characteristic distance between the target cross-correlation feature corresponding to each guiding knowledge point under each building feature label and the target IoT sensor network relationship graph vector, the server determines the building environment abnormal path prediction data corresponding to each guiding knowledge point under each building feature label. For example, if the characteristic distance of the "humidity fluctuation amplitude" of the inpatient department is large, it may be predicted that there is an abnormal humidity regulation in this area.
[0199] Finally, the server integrates the prediction data of the abnormal building environment paths of each guiding knowledge point under the building feature tag. For example, in the inpatient department, the prediction data of guiding knowledge points such as "temperature change trend", "humidity fluctuation range", and "air quality level" are comprehensively considered to generate the abnormal building environment path of the inpatient department as a whole, and it may be concluded that the environmental conditioning system of some wards needs to be maintained or optimized.
[0200] Through such a series of processing, the newly acquired target IoT sensor network data can be comprehensively and accurately analyzed for the building environment status and abnormal path prediction.
[0201] Figure 2 The IoT sensor data processing system 100 based on smart buildings shown includes: a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the IoT sensor data processing system 100 based on smart buildings may also include a transceiver 1004, which may be used for data interaction between the server and other servers, such as data transmission and / or data reception. It should be noted that the transceiver 1004 is not limited to one in actual scheduling, and the structure of the IoT sensor data processing system 100 based on smart buildings does not constitute a limitation on the embodiments of the present application.
[0202] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.
[0203] The bus 1002 may include a path to transmit information between the above components. The bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 1002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0204] The memory 1003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage medium, other magnetic storage devices, or any other medium that can be used to carry or store program code and can be read by a computer, without limitation herein.
[0205] The memory 1003 is used to store program codes for executing the embodiments of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the program codes stored in the memory 1003 to implement the steps shown in the above method embodiments.
[0206] An embodiment of the present application provides a computer-readable storage medium having program code stored thereon. When the program code is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.
[0207] It should be understood that, although each operation step is characterized by directed lines in the flow chart of the embodiment of the present application, the implementation order of these steps is not limited to the order covered by the directed lines. Unless there is clear explanation in this article, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flow chart can be performed in other orders based on demand. In addition, some or all of the steps in each flow chart may include multiple sub-steps or multiple stages according to actual implementation scenarios, and some or all of these sub-steps or stages may be performed in the same stage, and each sub-step or stage in these sub-steps or stages may also be performed in different stages respectively. Under different scenarios in the execution stage, the execution order of these sub-steps or stages can be flexibly configured based on demand, and the embodiment of the present application does not limit this.
[0208] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the scheme of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.
Claims
1. A method for processing IoT sensor data based on smart buildings. It is characterized in that The method comprises: Acquire example learning data, wherein the example learning data includes a plurality of example IoT sensor network data, wherein the example IoT sensor network data has an example building environment abnormal path under one or more building feature tags among a plurality of building feature tags; Based on the first graph network model in the building environment state prediction network, the example Internet of Things sensor network data is subjected to graph self-encoding processing to generate an example Internet of Things sensor network relationship graph vector; Based on the context generation unit in the building environment state prediction network, generating example context features corresponding to multiple guiding knowledge points respectively according to the example Internet of Things sensor network relationship graph vector; Based on the second graph network model in the building environment state prediction network, generating example cross-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels according to the basic context features respectively corresponding to the multiple guiding knowledge points and the example context features respectively corresponding to the multiple guiding knowledge points; Determine a first network training cost according to the example mutual correlation features of the multiple guiding knowledge points under each building feature label in the multiple building feature labels, the example Internet of Things sensor network relationship graph vector, and the example building environment abnormal path of the example Internet of Things sensor network data under one or more building feature labels; The network parameter information of the building environment state prediction network and the multiple guiding knowledge points are updated according to the first network training cost, and the updated guiding knowledge points are used to express the building environment state in the Internet of Things sensor network data from multiple guiding description fields.
2. According to the method for processing IoT sensor data based on smart buildings according to claim 1, It is characterized in that The second graph network model in the building environment state prediction network generates example cross-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels according to the basic context features respectively corresponding to the multiple guiding knowledge points and the example context features respectively corresponding to the multiple guiding knowledge points, including: Performing cross-correlation analysis on the basic context features of each guiding knowledge point and the example context features corresponding to the guiding knowledge point to generate mutual integration context features corresponding to each guiding knowledge point; The second graph network model based on the building environment state prediction network performs graph self-encoding processing on the mutual fusion context features corresponding to each of the guiding knowledge points to generate example cross-correlation features of each guiding knowledge point under each of the multiple building feature labels.
3. According to the method for processing IoT sensor data based on smart buildings as claimed in claim 1, It is characterized in that The method further comprises: Determine the deviation cost according to the feature distance between every two example cross-correlation features among the example cross-correlation features corresponding to multiple guiding knowledge points under the same building feature label; The guiding knowledge point is updated according to the deviation cost.
4. According to the method for processing sensor data of the Internet of Things based on smart buildings according to claim 3, It is characterized in that According to the feature distance between every two example cross-correlation features in the example cross-correlation features corresponding to multiple guiding knowledge points under the same building feature label, the deviation cost is determined, including: Calculate the feature distance between every two example cross-correlation features among the example cross-correlation features corresponding to multiple guiding knowledge points under the same building feature label; The feature distances are accumulated to generate deviation costs.
5. According to the method for processing sensor data of the Internet of Things based on smart buildings according to claim 1, It is characterized in that The determining of a first network training cost according to the example inter-correlation features of the multiple guiding knowledge points under each of the multiple building feature tags, the example IoT sensor network relationship graph vector, and the example building environment abnormal path of the example IoT sensor network data under one or more building feature tags includes: Calculate the vector distance between the example cross-correlation features of each of the guiding knowledge points under each building feature label and the example Internet of Things sensor network relationship graph vector; Determine the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label according to the vector distance between the example cross-correlation features corresponding to each guiding knowledge point under the same building feature label and the example Internet of Things sensor network relationship graph vector; Determine a first network training cost according to the example building environment abnormal paths of the example IoT sensor network data under one or more building feature tags and the predicted building environment abnormal paths of the example IoT sensor network data under each building feature tag; The step of determining the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label according to the vector distance between the example cross-correlation features corresponding to each guiding knowledge point under the same building feature label and the example Internet of Things sensor network relationship graph vector comprises: For each building feature label, the vector distance between the example cross-correlation feature corresponding to each guide knowledge point and the example IoT sensor network relationship graph vector is normalized. According to the normalized vector distance, a target weight is assigned to each guide knowledge point. The size of the target weight is inversely proportional to the vector distance. The calculation formula of the target weight is: target weight = (1 - normalized vector distance)^α, where α is an adjustment factor used to adjust the sensitivity of weight allocation; After obtaining the target weight of each guiding knowledge point, the example cross-correlation features of each guiding knowledge point under the building feature label are weighted fused, and the weighted fused features are input into the regression network in the building environment state prediction network for feature mapping to obtain the predicted building environment abnormal path of the example Internet of Things sensor network data under each building feature label.
6. According to the method for processing sensor data of the Internet of Things based on smart buildings according to claim 5, It is characterized in that The determining of the first network training cost according to the example abnormal building environment path of the example IoT sensor network data under one or more building feature labels and the predicted abnormal building environment path of the example IoT sensor network data under each building feature label comprises: The loss function value is calculated based on the abnormal building environment path of the example Internet of Things sensor network data under one or more building feature labels and the predicted abnormal building environment path of the example Internet of Things sensor network data under each building feature label to generate a first network training cost.
7. The method for processing IoT sensor data based on smart buildings according to claim 1, It is characterized in that Updating the network parameter information of the building environment state prediction network and the plurality of guiding knowledge points according to the first network training cost includes: The network parameter information and the guiding knowledge points of the context generation unit are updated according to the first network training cost.
8. The method for processing IoT sensor data based on smart buildings according to claim 1, It is characterized in that Before the example Internet of Things sensor network data is subjected to graph self-encoding processing based on the first graph network model in the building environment state prediction network to generate an example Internet of Things sensor network relationship graph vector, the method further includes: Acquire a plurality of sensor data-description combinations, wherein the sensor data-description combinations include template Internet of Things sensor network data and template guided description data; The plurality of template guided description data are loaded into the second graph network model, and the plurality of template IoT sensor network data are loaded into the first graph network model, and comparative training is performed to generate feature distances between template IoT sensor network data and template guided description data belonging to the same sensor data-description combination, and feature distances between template IoT sensor network data and template guided description data belonging to different sensor data-description combinations; generating a second network training cost according to a feature distance between the template IoT sensor network data and the template guided description data belonging to the same sensor data-description combination, and a feature distance between the template IoT sensor network data and the template guided description data belonging to different template IoT sensor network data pairs; The second graph network model and the first graph network model are trained according to the second network training cost until model convergence requirements are met.
9. The method for processing IoT sensor data based on smart buildings according to claim 1, It is characterized in that The method further comprises: Obtain target IoT sensor network data; Performing graph self-encoding processing on the target IoT sensor network data based on the first graph network model in the building environment state prediction network to generate a target IoT sensor network relationship graph vector; A context generation unit in the building environment state prediction network generates target context features corresponding to a plurality of guide knowledge points respectively according to the target IoT sensor network relationship graph vector, wherein the plurality of guide knowledge points are used to describe the building environment state in the target IoT sensor network data from a plurality of guide description fields; Based on the second graph network model in the building environment state prediction network, according to the basic context features corresponding to the multiple guiding knowledge points and the target context features corresponding to the multiple guiding knowledge points, the target cross-correlation features of the multiple guiding knowledge points under each of the multiple building feature labels are generated; Calculate the characteristic distance between each target cross-correlation feature and the target IoT sensor network relationship graph vector, and determine the abnormal path prediction data of the building environment corresponding to each guiding knowledge point under each building feature label according to the characteristic distance between the target cross-correlation feature and the target IoT sensor network relationship graph vector; The abnormal building environment path prediction data of each guiding knowledge point under the building feature label are fused to generate the abnormal building environment path of the target IoT sensor network data under the building feature label.
10. An IoT sensor data processing system based on smart buildings, It is characterized in that It includes a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by the processor, the method for processing IoT sensor data based on smart buildings as described in any one of claims 1 to 9 is implemented.
Citation Information
Patent Citations
Production process operation state monitoring method and related equipment
CN113919540A
Building model data analysis method and system based on BIM
CN118114127A
Public safety monitoring method and system based on networking collaborative awareness of Internet of Things
CN118353917A
BIM system informatization processing method and system based on artificial intelligence
CN118468409A
