REAL-TIME AIR QUALITY MONITORING, FORECASTING, AND INTELLIGENT NOTIFICATION SYSTEM

Abstract

This invention relates to a system (1) that enables the verification of air quality measurements, their synchronization in terms of time and location, their conversion into a dispersion graph using neighboring base station topology and wind direction / speed information, the generation of short and medium term pollution forecasts from this graph, the issuance of an alarm decision with a reliability-weighted composite score and its transmission to the institution / user channels with its technical justification.

Description

1 TARIFF REAL-TIME AIR QUALITY MONITORING, FORECASTING AND INTELLIGENT NOTIFICATION SYSTEM Technical Area 5 This invention enables the verification of air quality measurements, in terms of time and location. synchronization in the context of neighboring base station topology with wind Converting the direction / velocity information into a propagation graph, and this graph generating short and medium-term pollution forecasts, the alarm decision is 10 It should be given as a weighted composite score for reliability and distributed to institutional / user channels. It relates to a system that enables transmission for technical reasons. Previous Technique In the known state of the art, current air quality monitoring systems are mostly fixed measurement stations, independent sensor devices or general meteorology It relies on these services. In these structures, the number of measurement points may be limited, low. Errors in expensive sensors can produce false alarms, and pollution can be carried by the wind. Often, the decision-making process is not technically modeled. The previous technical 20 The first problem identified in the solutions is the reliability of the sensor measurement value. It is an assessment independent of the degree. Sensor calibration drift, temporary. Saturation, packet delay, or short-term measurement spikes are actual pollution events. This can be perceived as such. The second problem is the proximity of the measurement points and the wind direction. It is not sufficiently dependent on the decision-making mechanism. PM2.5 rising at a station 25 the value reaches the neighboring station in the windward direction with a certain delay. This is to be expected. Another problem is generating alarms with static threshold values. Fixed PM2.5 or PM10 thresholds depend on seasonality, regional baseline levels, and sensors. False alarm or late alarm when reliability and trend speed are not taken into account. It can produce. Another problem is that the machine learning model is affected by changes in data distribution. 2 It is due to becoming uncontrolled. Season, city traffic flow, sensor aging, or The model may deviate when the environmental conditions change. Therefore, considering the studies and shortcomings in the current technique... When considered, the windward propagation graph of the base station topology is 5. conversion, measurement confidence score, trend consistency, graph fit, and model. integrating the score into a combined anomaly confidence score, dynamic threshold and regression rollback controlled model management should be performed, with a range of 0 to 1 for each measurement. calculation of the normalized measurement confidence score and alarm decision. 10 used as weights, base stations in the form of graph nodes identification of wind direction and speed in a directional / weighted edge manner a system that enables the use of and calculation of pollution spread on a graph It is understood that the system is needed. Chinese patent CN109900865A, which is included in the prior art, 15 in the document, a neural network-based air pollution detection system It is mentioned that the system in question, which is the invention in question, consists of a sensor array, several sensors. It consists of a node and detects carbon monoxide and nitrogen in the environment. dioxide, ozone, carbon dioxide, smoke and particulate matter concentrations It is collecting data on air pollution in question. The assessment module evaluates the current air pollution situation. A Backpropagation Neural Network based on feature data to derive its results. It uses a backpropagation (BP) neural network model. Brief Description of the Invention 25 The purpose of this invention is to integrate environmental sensors into base stations. calculating the reliability of air quality data from the modules, wind direction propagation graph of data on base station topology 30 placement of neighboring stations included in the prediction model for lagged effects. 3 and the alarm decision should be made based on a combined anomaly confidence score. It is about implementing a system developed to provide this. Detailed Description of the Invention "Real-time Air Quality Monitoring" was carried out to achieve the purpose of this invention. The "Prediction and Smart Notification System" is shown in the attached diagram; Figure 1 shows a schematic view of the system that is the subject of the invention. The parts shown in the figure are individually numbered, and these numbers correspond to... The corresponding answers are given below. 1. System 2. Mobile Application 15 3. Sensor Module 4. Data Collection Module 5. Central Server B. Base Station 20 D. External Server H. Map System The invention was developed to enable the verification of air quality measurements. subject system (1); 25 - Real-time weather forecast for the area where users are located or will be moving. quality, direction of spread, estimated area of ​​impact, expected start the time, expected duration and alarm confidence score at least one mobile device configured to enable them to view it application (2), 30 4 - very fine particles with a diameter of 2.5 micrometers or smaller and the sizes of small solid and liquid particles suspended in the air PM2.5, the unit of measurement for air pollution, indicates this. the parameter, particles with a diameter of 10 micrometers or smaller. The size of small solid and liquid particles that are suspended in the air is 5 PM10, the unit of measurement for air pollution, indicates measuring parameters such as gas, humidity, temperature, and wind, each at least one configured to generate a sensor confidence score for measurement sensor module (2), - sensor measurements with timestamp, location and integrity tag 10 transmitting, aligning data to common time windows, missing data in this case, correlation of neighboring base stations, wind direction weighted by using alignment and measurement confidence score together at least one base configured to perform data completion operations station data collection module (3), 15 - accessing data packets, packet order, timestamp, and integrity. Checking data integrity via the tag, at a specific time creating a graph for the window, and then using that graph to create a station. At which neighboring station and for how many minutes was the observed increase in pollution? Calculating that it can be observed with a delay, the spread graph weights are 20 and confidence score components together as prediction input. running a predictive model using a combined anomaly confidence score to technically generate a single sensor failure versus a real propagation event to separate, dynamic alarm threshold; regional reference profile, final false alarm rate, missed critical event rate, sensor reliability, and 25 Updating using seasonal baseline information, when the combined anomaly confidence score is greater than the alarm threshold or An alarm is triggered when the predicted probability of exceeding the critical threshold surpasses the defined risk. to generate a deviation score to track changes in data distribution If the deviation score exceeds the defined threshold, the candidate will be re-selected with new data. at least one central server configured to train the model (4) It includes. The mobile application (2) included in the system (1) which is the subject of the invention, where users are located or the current air quality, direction of spread, and estimated impact for the area it will move through. 5 area, expected start time, expected duration, and alarm confidence score It is configured to enable them to view it. In the system that is the subject of the invention, the sensor module (3) located in (1) is on the base station (B) They are positioned and are very thin, with a diameter of 2.5 micrometers or less. the size of small solid and liquid particles that are suspended in the air The parameter expressed as PM2.5, which is the unit of measurement for air pollution, indicates... particles that are 10 micrometers or smaller in diameter and are suspended in the air. a unit of air pollution measurement that indicates the size of small solid and liquid particles The parameter expressed as PM10 is a combination of gas, humidity, temperature, and wind. to measure its parameters and generate a sensor confidence score for each measurement. It is configured. The sensor module (3) calculates the sensor confidence score: Cs_i(t) = with the formula w1*Kcal + w2*Ksat + w3*Kvar + w4*Kneighbor + w5*Kpacket It is calculated in the form of a weighted composition, where the Kcal value is used for calibration. validity, measurement saturation control with the Ksat value, short 20 with the Kvar value consistency of period variance, consistency with neighboring sensors with the Kneighbor value, The kpacket value represents packet delay and data loss, while w represents weights. to determine the weights; sensor type, field reliability or past verification It is structured to enable determination based on the results. The data collection module (4) located in the system (1) which is the subject of the invention, base station (B) It is positioned on it and takes measurements from the sensor modules (3). transmit to the central server (5) with timestamp, location and integrity tag, Aligning data to common time windows, using neighboring bases in case of missing data. station correlation, wind direction alignment, and measurement confidence score 30 6 to perform weighted data completion operations by using them together It is being structured. The central server (5), data collection module (4) in the system (1) which is the subject of the invention Data received via: station ID, sensor ID, parameter type, value, unit, 5 timestamp, latitude, longitude, sampling rate, confidence score, sequence number, Accessing data packets containing integrity tags and model versions, integrity tag Using the field as a hash-based validation tag, packet order, Checking data integrity via timestamp and integrity tag It is structured to clean the raw data. The central server (5) is used to clean the raw data. normalize and align according to the base station-based reference profile, each maintaining hourly, daily and seasonal baseline profiles for the station, this profile to calculate how much the measured value deviates from the regionally normal behavior. It is structured as follows: Central server (5), Z-score, Interquartile Width (Interquartile Range - IQR), Robust Scaler 15 Identifying outlier candidates using methods such as these, and assigning anomaly candidates to each outlier. labeling and evaluating this candidate's value based on sensor confidence score, trend consistency, and diffusion. It is configured to verify the graph compatibility. Central server (5), To create a graph of the form G_t=(V,E_t) for a specific time window, where The V cluster represents base stations, and the E_t cluster represents wind direction and neighborhood relationships. It is structured to represent the directional edges that are active according to the central function. server (5) as W_ij(t) = a*C_ij + b*A_ij(t) + c*Vwind(t) - d*D_ij To produce a normalized weight, where W_ij(t) is used with two base stations edge weight between them, historical correlation coefficient with C_ij, and A_ij(t) The wind direction alignment coefficient, D_ij, and the distance / transport delay factor are 25. and is structured to represent the wind speed component with Vwind(t). The central server (5) determines the value of A_ij(t) based on the wind direction and the j from station i. Calculating the wind direction based on the angle between the direction vector to the station and the direction vector j. Increase edge weight if aligned correctly with the station, decrease edge weight if aligned in the opposite direction. This graph shows which neighboring stations are experiencing an increase in pollution at one station. to calculate how many minutes of delay it will have to be observed at the station 7 is being configured. Central server (5), last 5 / 15 / 60 minute measurement windows, moving average, rate of change, derivative, sudden bounce coefficient, Time / day / season information, sensor reliability score, delayed data from neighboring stations. measurements, edge weight (W_ij(t)) spread weights between two base stations and using inputs consisting of parameters such as wind direction / speed, a 5 running the forecasting model, the expected results for the selected future time window PM2.5 / PM10 / gas levels, probability of exceeding the critical threshold, and expected area of ​​impact. It is configured to produce model outputs in the following way. Central server (5), Long Short-Term Memory (LSTM), Gated Repetitive Memory Unit (Gated Recurrent Unit - GRU), Temporal Convolution, 10 Graph Neural Network (GNN) or equivalent spatio-temporal network In the (spatio-temporal) prediction model, the dispersion graph weights and to use confidence score components together as prediction input is configured. The central server (5) generates a combined anomaly confidence score, 15 normalized as FAC_i(t)=f(M_i(t),Cs_i(t),T_i(t),G_i(t),P_i(t),N_i(t)) Generating a score between 0 and 1, where FAC_i(t) is the combined anomaly confidence score. M_i(t) represents the model anomaly score, Cs_i(t) represents the measurement confidence score, and T_i(t) represents the measurement confidence score. trend consistency, G_i(t) with dispersion graph fit, P_i(t) with estimated to represent the critical overshoot probability and the neighbor validation score with N_i(t) is being configured. Central server (5), low measurement confidence score (Cs_i(t)) 20 The combined anomaly confidence score in the case of a sudden rise in the value of a single measurement. (FAC) to suppress the value, while in neighboring stations, the wind direction If there is a compatible lagged increase, the propagation graph compatibility (G_i(t)) and neighboring Anomaly Confidence Score (FAC) combined with validation score (N_i(t)) components to increase its value, thus technically comparing the single sensor failure with the actual propagation event 25 It is configured to separate them as follows: Central server (5), dynamic alarm thresholds include: regional reference profile, last false alarm rate, missed critical event rate, using sensor reliability and seasonal baseline information It is configured to update. Central server (5), combined anomaly confidence score when greater than the alarm threshold or predicted critical overshoot 30 It is configured to generate an alarm when the probability exceeds a defined risk. 8 The central server (5) is designed to prevent repeated notifications for the same event. cooldown and deduplication policies implementing it, but the spread graph shows a new target area at risk. If it shows this, the same source event can generate separate alarms specific to the new target region. This helps reduce notification fatigue and minimizes the actual risk of transmission. It is configured to ensure that it is not lost. The central server (5) provides data. To track changes in drift, calculate the drift score. score; measurement distribution change, increase in error rate, prediction bias, sensor reliability using the average score and the deviation values ​​from the regional reference profile It is configured to produce. The central server (5) determines the deviation score. If the threshold is exceeded, the candidate model must first be trained with new data to activate the candidate model. In the validation window, the candidate model is compared with the currently active model. accuracy, false alarm rate, missed event rate, and spread prediction performance If it meets the acceptance criteria, activate the model, after activation. If performance degrades in the monitoring window, revert to the previous stable model version 15. to return, thus with the secure deployment logic of online adaptation It is configured to enable its implementation. Central server (5), current air quality, direction of spread, estimated area of ​​impact, expected start date time, expected duration and alarm confidence score to mobile applications (2) transmit and integrate the map system (H) by communicating with the external server (D), 20 Application Programming Interface (API) The event message transmitted includes the following: event ID, station ID, Event type, type of pollutant, current value, estimated value, confidence score, affected area, expected onset time, expected duration, direction of spread, recommended With areas such as action and model release, organizations can use 25 instead of raw data. to enable it to obtain a reasoned and actionable event outcome It is being structured. Industrial application of the invention 9 In the system that is the subject of the invention (1), the sensor module (3) measures the air quality parameters, It generates the measurement confidence score and adds it to the data packet. Data collection module (4), Measurement packets are centralized with timestamps, location tags, and integrity labels. It transmits to the server (5). The central server (5) handles packet ordering, hash-based verification and It performs data loss control, cleans the data, normalizes it, and adjusts it to a reference profile of 5. It generates alignments and anomalies candidate tags, base stations (B) node, wind By modeling the direction as a directed edge, W_ij(t) generates the diffusion coefficients. Using delayed measurements and propagation coefficients from neighboring stations, the future It generates pollution estimates for a given time window, model score, confidence score, and trend. It generates a combined anomaly confidence score from consistency and graph compatibility; dynamic 10 It makes an alarm decision based on the threshold, calculates the deviation score, and validates the candidate model. It blocks the unaccepted model and, if necessary, reverts it. Mobile application (2), It displays the current value, forecast, spread direction, area of ​​impact, and personalized alert. The central server (5) provides the alarm justification via an application programming interface. Confidence score, estimated time, scope, and recommended action information are included in standard event 15. It is sent to the relevant institutions in this format. Based on these fundamental concepts, the invention focuses on "Real-time Air Quality Monitoring and Forecasting." and development of a wide variety of applications related to the Smart Notification System (1)” It is possible, and the invention is not limited to the examples described here, but mainly 20 As stated in the requests.

Claims

REQUESTS 1. Developed to ensure the verification of air quality measurements; - Real-time weather forecast for the area where users are located or will be moving. quality, direction of spread, estimated area of ​​impact, expected start date 5 the time, expected duration and alarm confidence score at least one mobile device configured to enable them to view it application (2) includes and; - very fine particles with a diameter of 2.5 micrometers or smaller and the sizes of small solid and liquid particles suspended in the air are 10 PM2.5, the unit of measurement for air pollution, indicates this. the parameter, particles with a diameter of 10 micrometers or smaller. the size of small solid and liquid particles that are suspended in the air PM10, the unit of measurement for air pollution, indicates measuring parameters such as gas, humidity, temperature, and wind every 15 at least one configured to generate a sensor confidence score for measurement sensor module (3), - sensor measurements with timestamp, location and integrity tag transmitting, aligning data to common time windows, missing data In this case, the correlation of the neighboring base station, wind direction 20 weighted by using alignment and measurement confidence score together at least one base configured to perform data completion operations station data collection module (4), - accessing data packets, packet order, timestamp, and integrity. Checking data integrity via the tag, within a specific timeframe 25 creating a graph for the window, and then using that graph to create a station. At which neighboring station and for how many minutes was the observed increase in pollution? Calculating how long it will be observable with a delay, spread graph weights and confidence score components together as prediction input. Running a predictive model using a combined anomaly confidence score of 30 to technically generate a single sensor failure versus a real propagation event 11 to separate, dynamic alarm threshold; regional reference profile, final false alarm rate, missed critical event rate, sensor reliability, and Updating using seasonal baseline information, when the combined anomaly confidence score is greater than the alarm threshold or Alarm 5 when the predicted critical overshoot probability exceeds the defined risk. to generate a deviation score to track changes in data distribution to calculate, if the deviation score exceeds the determined threshold, the candidate will be given new data. with at least one central server (5) configured to train the model a characterized system (1).

2. Real-time weather for the area where users are located or will be moving. quality, direction of spread, estimated area of ​​impact, expected start They should be able to view the time, expected duration, and alarm confidence score. characterized by the mobile application (2) structured to provide A system like the one in Claim 1 (1). 15 3. It is located on base station (B) and has a diameter of 2.5 micrometers. or even smaller, very fine particles that are suspended in the air. Air pollution measures that indicate the size of small solid and liquid particles. The parameter, expressed in units such as PM2.5, has a diameter of 10 micrometers. or smaller particles, which are small solids suspended in the air. PM10 is a unit of air pollution measurement that indicates the size of liquid particles. The parameters expressed as follows are gas, humidity, temperature, and wind. to measure its parameters and generate a sensor confidence score for each measurement. The above 25 is characterized by the configured sensor module (3). a system like any of the requests (1).

4. Sensor confidence score: Cs_i(t) = w1*Kcal + w2*Ksat + w3*Kvar + As a weighted compound with the formula w4*Kneighbor + w5*Kpacket. It is calculating the calibration validity here with the Kcal value, Ksat 30 Measurement saturation control with the kvar value, short term with the kvar value. 12 consistency of variance, consistency with neighboring sensors using the Kneighbor value, The kPacket value indicates packet delay and data loss status, while w indicates... expressing weights, weights depending on sensor type, field reliability or to enable identification based on past verification results The above 5 is characterized by the configured sensor module (3). a system like any of the requests (1).

5. The sensor is located on base station (B). timestamps, positions and integrity of measurements from modules (3) transmitting the data to the central server (5) with the tag, in common time windows 10 aligning, correlation of neighboring base stations in case of missing data, by using wind direction alignment and measurement confidence score together Data collection structured to perform weighted data completion processing. any of the above requests characterized by module (4) a system like one of them (1). 15 6. Station ID, sensor ID, received via data collection module (4), parameter type, value, unit, timestamp, latitude, longitude, sampling including speed, confidence score, sequence number, integrity label, and model version. Accessing data packets, integrity tag field summary-based validation 20 using as a label, package order, timestamp and integrity label centralized system configured to check data integrity any of the above requests characterized by the server (5) such a system (1).

7. Cleaning, normalizing, and referencing raw data based on base station criteria. Adapting according to profile, hourly, daily and seasonal for each station. Maintaining a baseline profile, with this profile, allows the measured value to be compared with regional norms. a central structure designed to calculate how much the behavior deviates 13 any of the above requests characterized by the server (5) such a system (1).

8. Z-score, Interquartile Range, Robust Scaler Identifying outlier candidates using methods, assigning anomaly candidates to each outlier. assigning the label and valuing this candidate sensor confidence score, trend consistency, and centralized to verify with spread graph compatibility any of the above requests characterized by the server (5) such a system (1).

9. To create a graph of the form G_t=(V,E_t) for a specific time window. Here, V represents the base stations, and E_t represents the wind direction. to represent the active directed edges according to the neighborhood relationship the above characterized by the configured central server (5) a system like any of the requests (1). 15 10. Normalized as W_ij(t) = a*C_ij + b*A_ij(t) + c*Vwind(t) - d*D_ij to produce a weight, where W_ij(t) is between two base stations edge weight, historical correlation coefficient with C_ij, wind with A_ij(t) The alignment coefficient with respect to direction, and the distance / transport delay factor D_ij are 20. and configured to represent the wind speed component with Vwind(t) any of the above requests characterized by the central server (5) a system like one of them (1).

11. The value of A_ij(t) is calculated using the wind direction and the direction from station i to station j. Calculating based on the angle between the vectors, wind direction to station j. Increase edge weight if it's correctly aligned, decrease it if it's the opposite. The graph shows which neighboring station is experiencing an increase in pollution at one station. to calculate how many minutes of delay it will have to be observed at the station 14 the above characterized by the configured central server (5) a system like any of the requests (1).

12. Last 5 / 15 / 60 minute measurement windows, moving average, rate of change, derivative, sudden jump coefficient, time / day / season information, sensor confidence score, 5 Delayed measurements from neighboring stations, between two base stations edge weight spread weights and wind direction / speed running a predictive model using inputs consisting of parameters, Expected PM2.5 / PM10 / gas levels for the selected future time window, Model outputs in the form of critical threshold exceedance probability and expected area of ​​impact 10 characterized by a central server (5) configured to produce a system like any of the above requests (1).

13. Long-Term and Short-Term Memory, Gated Repetitive Unit, Temporal Convolution, Graphical Neural Network or equivalent spatio-temporal prediction 15 estimating dispersion graph weights and confidence score components in the model central server configured to be used together as input (5) as in any of the above claims characterized by system (1).

14. Generating a combined anomaly confidence score, FAC_i(t) = f(M_i(t), Cs_i(t), A normalized score between 0 and 1 in the form of T_i(t), G_i(t), P_i(t), N_i(t)) to generate, here the combined anomaly confidence score with FAC_i(t), and M_i(t) model anomaly score, measurement confidence score with Cs_i(t), trend with T_i(t) consistency, spread graph fit with G_i(t), estimation with P_i(t) 25 to represent the critical overshoot probability and the neighbor validation score with N_i(t) The above is characterized by the central server (5) which is configured to do so. a system like any of the requests (1).

15. Single measurement spike with low measurement confidence score. In this case, suppressing the combined anomaly confidence score value, conversely If there is a delayed increase in neighboring stations that is consistent with the wind direction, then the spread combined with graph compatibility and neighbor validation score components. Increasing the anomaly confidence score value, so that with a single sensor failure, 5 structured to technically isolate the actual spread event any of the above requests characterized by the central server (5) a system like one of them (1).

16. Dynamic alarm threshold; regional reference profile, last false alarm rate, 10 missed critical event rate, sensor reliability, and seasonal baseline. a central system configured to update using information in this format any of the above requests characterized by the server (5) such a system (1).

17. When the combined anomaly confidence score is greater than the alarm threshold, or An alarm is triggered when the predicted probability of exceeding the critical threshold surpasses the defined risk. characterized by a central server (5) configured to produce a system like any of the above requests (1).

18. Waiting period to prevent duplicate notifications for the same event and Implementing deduplication policies, but the diffusion graph is a new target. If the source indicates that the region is at risk, then a new target can be identified from the same event. Generating separate alarms specific to each region, thus reducing notification fatigue. 25 characterized by a central server (5) configured to provide a system like any of the above requests (1).

19. Calculating the deviation score to track changes in data distribution. score; measurement distribution change, increase in error rate, prediction bias, sensor 30 16 mean confidence score and deviation from regional reference profile values characterized by a central server (5) configured to produce using a system like any of the above-mentioned requests (1).

20. If the deviation score exceeds the defined threshold, train the candidate model with new data. 5 To activate the candidate model, first check the existing active one in the validation window. comparing with the model, the accuracy of the candidate model, the false alarm rate, missed event rate and spread prediction success are the acceptance criteria. If it meets the requirements, activate the model and monitor after activation. If performance deteriorates in the window, revert to the previous stable model version 10. to return, thus with the secure deployment logic of online adaptation central server (5) configured to enable its implementation as in any of the above characterized claims system (1).

21. Real-time air quality, direction of spread, estimated area of ​​impact, expected start time, expected duration and alarm confidence score mobile transmitting to applications (2) and communicating with the external server (D) to map to integrate into the system (H) via the application programming interface The transmitted event message includes: event ID, station ID, event 20. type, type of pollutant, current value, estimated value, confidence score, affected area, expected onset time, expected duration, direction of spread, With areas such as proposed action and model release, the raw materials of the institutions To enable it to receive reasoned and actionable event output instead of data. The above 25 is characterized by a central server (5) configured to do so. a system like any of the requests (1).

Images