A cloud-machine combined air quality intelligent decision system and method

Abstract

The application discloses a cloud-combined air quality intelligent decision system and method, the method comprises the following steps: designing an adaptive target function, using a moss growth optimization algorithm improved by coupling a chaos mapping with a population adaptive adjustment strategy to perform optimal path planning on a UAV, performing data processing through a multi-source heterogeneous data processing model, taking air pollution data and geographical positions collected by the UAV in real time as inputs of a BiLSTM-Informer air quality detection model, performing spatio-temporal multi-scale detection, and providing auxiliary decision support; the application can better adapt to environmental changes, realizes real-time monitoring and intelligent and effective decision-making of air quality in a harsh environment, and through a multi-objective optimization method, the UAV can consider both efficiency and safety during path planning.

Description

Technical Field

[0001] This invention relates to an intelligent air quality decision-making system and method, specifically to a cloud-based intelligent air quality decision-making system and method. Background Technology

[0002] Current air pollution monitoring technologies mainly rely on fixed ground-based monitoring stations, which suffer from problems such as delayed data updates, inability to reflect real-time air quality changes, and difficulty in responding to extreme weather conditions. Although significant achievements have been made in air quality forecasting, such as the independent development of air quality models and improved forecasting capabilities, domestic research methods generally lack specificity and there is a lack of technical specifications to guide the selection of parameterized schemes. This results in poor comparability of forecast results and an inability to effectively provide accurate and real-time information to environmental protection and emergency management departments.

[0003] To address the aforementioned issues, traditional air pollution monitoring methods have limitations. Ground-based monitoring stations have long data update cycles, making it difficult to reflect real-time air quality trends. Furthermore, environmental factors such as strong winds, heavy rains, and extreme weather can affect sampling and data transmission at ground-based monitoring points, leading to difficulties in data acquisition. While existing air quality forecasting technologies can provide early warning information, the lack of localized parameterization based on high-quality observation data and the guidance of technical specifications limits the accuracy of model predictions, making it difficult to meet the application requirements of high precision and rapid response.

[0004] With their small size, flexibility, and ease of deployment, micro unmanned aerial vehicles (UAVs) can play an important role in scenarios where traditional monitoring methods are inadequate, such as accident site monitoring, high-altitude monitoring, and hazardous area monitoring. Combined with the computing and data processing capabilities of cloud platforms, micro UAVs can achieve real-time monitoring, data transmission, and predictive analysis of air pollutants, providing environmental protection departments and emergency management departments with more accurate, timely, and targeted information for air pollution prevention and emergency response. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a cloud-based intelligent air quality decision-making system to improve the accuracy of air quality detection and the efficiency and safety of UAV trajectory planning. On the other hand, it also provides a cloud-based intelligent air quality decision-making method.

[0006] Technical solution: The cloud-based intelligent air quality decision-making system of the present invention includes:

[0007] The adaptive path planning module is used to design an adaptive objective function and utilizes a moss growth optimization algorithm improved by coupling chaotic mapping with a population adaptive adjustment strategy to perform optimal path planning for UAVs.

[0008] A multi-source heterogeneous data acquisition module is used to collect and process air pollutant data and aerial image data. The multi-source heterogeneous data acquisition module includes a gas sensor, an ultrasonic sensor, and an accelerometer.

[0009] The central processing unit module is used to receive data from the UAV's multi-source heterogeneous data acquisition module and transmit the data to the cloud platform using the wireless fidelity communication module.

[0010] The wireless fidelity communication module is used to realize data transmission between the central processing unit module and the cloud platform;

[0011] A cloud platform is used to provide auxiliary decision support, and the cloud platform specifically includes:

[0012] The data management module is used to clean and integrate the collected data, remove duplicate and erroneous data to improve the accuracy of the final results, integrate time series data from different sensors and drone aerial image data to facilitate subsequent retrieval and analysis, and back up data regularly to ensure that the system can recover data in case of loss or damage.

[0013] The air quality prediction module drives the BiLSTM-Informer model to predict future air quality based on the current air quality conditions.

[0014] The performance evaluation module is used to evaluate the performance of the gas sensors carried by the drone.

[0015] The cloud platform display module is used to receive and display the predicted results of future air quality.

[0016] The air quality remote control module is used to remotely control the flight path and monitoring tasks of drones, enabling flexible coverage of the monitoring area, assessing environmental quality, and issuing early warnings when pollutants exceed standards.

[0017] Preferably, the formula for calculating the adaptive objective function in the adaptive path planning module is as follows:

[0018] f min = (1-p1-p2)f1+p1f2+p2f3;

[0019]

[0020] Where f1 represents the range cost, f2 represents the threat cost, f3 represents the altitude cost, p1 and p2 represent the importance coefficients of the objective function, and p2 is dynamically adjusted based on the data collected by the ultrasonic sensor;

[0021] λ1 and λ2 represent the cost weighting coefficients of power consumption during drone flight time, t(n,n+1) represents the flight time of the drone from the nth waypoint to the (n+1)th waypoint, and q represents the proportionality coefficient between unit power consumption and flight distance. This represents the total flight distance of the drone between any two points in the three-dimensional search space.

[0022] d n This indicates the danger level of the nth flight node, sum_obstacle represents the number of obstacle grids around the waypoint, and sum_surround represents the total number of grids around the waypoint.

[0023] n num φ represents the total number of waypoints. max and φ min φ represents the weighting factor for heights exceeding the maximum and below the minimum. rate The weighting coefficient h represents the rate of change of height. max and h min This indicates the set maximum and minimum flight altitude, a n b n c n This represents the nth track node of the drone flight aircraft;

[0024] θ max β represents the maximum pitch angle of the drone. max Indicates the maximum turning angle.

[0025] Preferably, the formula for the BiLSTM-Informer model in the air quality prediction module is:

[0026]

[0027] Q m =MultiHeadAttention(X′) t );

[0028] C m =Softmax{ReLU|(W q Q m ) T W O Q m ]};

[0029]

[0030] Among them, X t denoted as the input to the BiLSTM-Informer model, p represents the spatial sampling point, m represents the type of air pollutant, and n represents the sample size of the concentration of the m-th air pollutant.

[0031] Q represents the query matrix, O represents the key-value matrix, L represents the value matrix, and W represents the value matrix. q W O W L Let d represent the trainable weight matrix. k This represents the number of columns in the Q, O, and L matrices, i.e., the vector dimensions. Q m X' represents a local feature in the head attention mechanism. t This represents the different time series segments into which the input time series data is decomposed, C m S represents the feature weights of different time series. (m) This represents the self-attention weights between each time segment. U represents the output pollutant concentration at time t. m This represents the hyperparameter matrix in the BiLSTM-Informer model, which can be optimized to improve model accuracy and generalization ability. Softmax is a function in deep learning that can transform the original output of the model into a probability distribution. MultiHeadAttention is a multi-head self-attention mechanism used to process sequential data.

[0032] Preferably, the performance in the performance evaluation module includes:

[0033] Used to evaluate the accuracy, sensitivity, response time, and stability of a drone's data acquisition capabilities;

[0034] The data acquisition frequency, data real-time performance, and data processing efficiency are used to evaluate the data transmission and processing capabilities of drones.

[0035] The cloud-based intelligent air quality decision-making method of the present invention includes the following steps:

[0036] (1) An adaptive objective function for the UAV is designed through an adaptive path planning module. The optimal path planning for the UAV is performed by using a moss growth optimization algorithm improved by a chaotic mapping coupled with a population adaptive adjustment strategy.

[0037] (2) Collect and process air pollutant data and aerial image data through a multi-source heterogeneous data processing module;

[0038] (3) The central processing unit module receives data from the UAV multi-source heterogeneous data acquisition module and transmits the data to the cloud platform using the wireless fidelity communication module;

[0039] (4) The cloud platform uses the BiLSTM-Informer prediction model to predict future air quality based on the current air quality conditions, evaluates the performance of the gas sensor carried by the UAV, displays the final prediction results of future air quality, and provides auxiliary decision support.

[0040] (5) Real-time monitoring data is achieved through the air quality remote control module. When the concentration of pollutants exceeds the preset threshold, an early warning mechanism is triggered, and pollutant hotspots and warning areas are displayed.

[0041] Preferably, the optimal path planning for the UAV in step 1 includes the following specific steps:

[0042] (11) Initialize the UAV position using chaotic mapping, and determine the optimal path search direction based on the adaptive objective function value;

[0043] (12) Update the UAV position according to the search direction and search step size;

[0044] (13) Locally update the drone's position to realize the planning of the drone's optimal path.

[0045] Preferably, the chaotic mapping formula in step 1 is:

[0046]

[0047] Where R represents the algorithm search space variable, r represents the control parameter, R∈[0,1], r∈(0,4), M(0) represents the optimized initial position of the UAV, α and β represent the lower and upper limits of the algorithm search range, and mod 1 represents the modulo operation with respect to 1;

[0048] The formula for the population adaptive adjustment strategy is:

[0049]

[0050] Among them, M bad M represents the worst-case drone position. bes t represents the optimal UAV position, v represents the current objective function evaluation, % represents the remainder or modulo operator, FE s MaxFE represents the current iteration number. s This indicates the maximum number of iterations.

[0051] Preferably, the formula for determining the optimal path search direction in step 11 is:

[0052]

[0053] Where f represents the adaptive objective function, D dir This indicates the calculated search direction, and num represents the total number of alternative paths for the UAV in divV.

[0054] Preferably, the formula for updating the UAV position based on the search direction and search step size in step 22 is as follows:

[0055]

[0056] Among them, M i D indicates the current location of the drone. dir This indicates the calculated search direction, and the search step size is determined based on the relative magnitudes of random numbers p1 and d1.

[0057] When p1 > d1, a long-distance search is used: F1 = w·(p2-0.5)·E;

[0058] Where w represents a constant parameter, w is set to 2, p2 represents a random variable in the range (0,1), and E is the standard search step size;

[0059] When p1≤d1, use short-distance search:

[0060]

[0061] Here, p3 is a random vector in the range (0,1) that controls the search step size;

[0062] FE s MaxFE represents the current iteration number. s This indicates the maximum number of iterations.

[0063] Preferably, the formula for locally updating the UAV position in step 13 is:

[0064]

[0065] in, Let d1 represent the position of the i-th drone, and d2 and p4 represent random numbers between (0,1).

[0066] Determine whether to utilize M based on the location update factor act. best The formula for updating the drone's location is as follows:

[0067]

[0068] F3 = 0.1·(p6-0.5)·E;

[0069] Here, p5 represents a random vector in the range (0,1), and p6 represents a random number in the range (0,1).

[0070] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. By considering three important factors, namely range cost, threat cost, and altitude cost, the flexibility and adaptability of the algorithm are improved; 2. The moss growth optimization algorithm improved by using chaotic mapping coupled with population adaptive adjustment strategy ensures uniform distribution of UAVs, thereby improving work efficiency and data acquisition accuracy; 3. By constructing a BiLSTM-Informer prediction model, the accuracy of prediction is improved. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the system framework of the present invention;

[0072] Figure 2 This is a schematic diagram of the optimal path planning of the present invention;

[0073] Figure 3 This is a schematic diagram of the air quality prediction model of the present invention;

[0074] Figure 4 This is a schematic diagram of the BiLSTM-Informer air quality prediction model of the present invention. Detailed Implementation

[0075] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0076] (1) Through the adaptive path planning module, the UAV's flight altitude is dynamically guided based on ultrasonic sensor data. An adaptive objective function that simultaneously considers the UAV's safety, flight efficiency, and mission completion is designed to perform optimal trajectory planning for the UAV and achieve dynamic trajectory adjustment. The formula for calculating the adaptive objective function is as follows:

[0077] f min = (1-p1-p2)f1+p1f2+p2f3;

[0078]

[0079]

[0080] Here, f1 represents the range cost, emphasizing efficiency and minimizing flight distance; f2 represents the threat cost; f3 represents the altitude cost, with f2 and f3 focusing on safety, requiring the UAV to avoid threat areas and maintain a safe flight altitude; p1 and p2 represent the importance coefficients of the objective function, with p2 dynamically adjusted based on data collected by the ultrasonic sensor, thereby dynamically adjusting the weights of the other two cost functions, achieving dynamic adjustment of the weights of these three costs, balancing the three to ensure efficient and safe mission execution;

[0081] λ1 and λ2 represent the cost weighting coefficients of power consumption during drone flight time, t(n,n+1) represents the flight time of the drone from the nth waypoint to the (n+1)th waypoint, and q represents the proportionality coefficient between unit power consumption and flight distance. This represents the total flight distance of the drone between any two points in the three-dimensional search space.

[0082] d n The sum_obstacle represents the danger level of the nth flight node, the sum_obstacle represents the number of obstacle grids around the waypoint, and the sum_surround represents the total number of grids around the waypoint. num Indicates the total number of waypoints;

[0083] φ max and φ min φ represents the weighting factor for heights exceeding the maximum and below the minimum. rate The weighting coefficient h represents the rate of change of height. max and h min This indicates the set maximum and minimum flight altitude, a n b n c n a represents the nth track node of the drone flight aircraft. n+1 b n+1 c n+1 θ represents the (n+1)th track node of the drone flight aircraft. max β represents the maximum pitch angle of the drone. max Indicates the maximum turning angle.

[0084] (2) An improved moss growth optimization algorithm using a chaotic mapping coupled with a population adaptive adjustment strategy is used to plan the optimal path for the UAV.

[0085] 21. Initializing the UAV position M using chaotic mapping ensures a more uniform UAV position in the search space, preventing the optimization process from prematurely converging to a local optimum, thereby improving global search capability and helping the optimization process better adapt to environmental changes. The mathematical formula for chaotic mapping is expressed as follows:

[0086]

[0087] Where R represents the algorithm search space variable, r represents the control parameter, R∈[0,1], r∈(0,4), M(0) represents the optimized initial position of the UAV, α and β represent the lower and upper limits of the algorithm search range, mod 1 represents the modulo operation of 1. Since the chaotic mapping is performed in the interval [0,1], it is necessary to ensure that the search results are always within this interval.

[0088] The optimal path search direction D is determined based on the adaptive objective function value. dir :

[0089]

[0090] Where f represents the adaptive objective function, D dir This indicates the calculated search direction, and num represents the total number of alternative paths for the UAV in divV.

[0091] 22. Update the UAV position based on the search direction and search step size. After the position is updated, a new UAV position is formed. The wind intensity decreases as the iteration progresses, thus simulating the process of the UAV gradually adapting to the environment and finding a better path. In this way, individual UAVs can better adapt to environmental changes, improving the efficiency and accuracy of their path planning. The updated position formula is as follows:

[0092]

[0093] Among them, D dir Indicates the search direction, M i This indicates the current position of the drone, and the search step size is determined based on the relative magnitudes of random numbers p1 and d1.

[0094] When p1 > d1, use long-distance search:

[0095] F1 = w·(p2-0.5)·F;

[0096] Where w represents a constant parameter, w is set to 2, and p2 represents a random variable in the range (0,1);

[0097] When p1≤d1, use short-distance search:

[0098]

[0099] Where p3 represents a random vector in the range (0,1), which can control the search step size;

[0100]

[0101] E is the standard search step size, FE s MaxFE represents the current iteration number. s This indicates the maximum number of iterations.

[0102] A population-adaptive adjustment strategy is adopted, which balances global and local searches by dynamically updating the moss positions during the optimization process. This dynamic adjustment helps ensure that the UAV finds the optimal path while accelerating the search efficiency. The method is as follows:

[0103]

[0104] Among them, M bad M represents the worst-case drone position. best This represents the optimal UAV position, v represents the current objective function evaluation, % represents the remainder or modulo operator, and FE s MaxFE represents the current iteration number. s This indicates the maximum number of iterations.

[0105] 23. Locally update the drone's position to plan the optimal drone path.

[0106] This allows the drone to focus more on the area near the current optimal path during path planning, in order to find more possible better paths and thus enhance the drone's local area exploration capabilities. The formula is as follows:

[0107]

[0108] in, Let d2 and p4 represent the position of the i-th drone, where d2 and p4 are random numbers between (0,1).

[0109] Position update factor act determines whether to use M best The formula for updating the drone's location is as follows:

[0110]

[0111] F3 = 0.1·(p6-0.5)·E;

[0112] Here, p5 represents a random vector in the range (0,1), and p6 represents a random number in the range (0,1).

[0113] (3) Process time series data and aerial image data acquired by pollutant detection sensors.

[0114] The multi-source heterogeneous data acquisition module processes time-series data from the pollutant detection sensors on the drone and aerial image data. The gas sensor detects harmful gases in the air; the accelerometer provides the acceleration force experienced by the drone in the x, y, and z axes, determining the drone's tilt angle when stationary and providing linear acceleration data in both horizontal and vertical directions; the ultrasonic sensor utilizes the property of ultrasound to bounce off other substances for altitude control, especially near the ground, and, in conjunction with a barometric pressure sensor, enables stable flight of the drone at both high and low altitudes.

[0115] (4) Collect air pollutant data and geographical location data from the UAV as input data, and input the input data into the BiLSTM-Informer air quality detection model.

[0116] During path optimization, the historical location of each drone's path planning is recorded. By selecting the best drone location from the historical records, the initial drone location is updated. This avoids redundant searches and improves search efficiency. The BiLSTM-Informer air quality detection model is constructed using the following formula:

[0117]

[0118] Q m =MultiHeadAttention(X′) t );

[0119] C m =Softmax{ReLU(W q Q m ) T W O Q m ”};

[0120]

[0121] Among them, i t This represents the input gate output, σ represents the sigmoid activation function, and W... xi The input weights of the input gate, x t W represents the input at time t. hi h represents the weights between the hidden state and the input gate. t-1 Let b represent the hidden state at time t-1. i f represents the bias between the hidden state and the input gate. t Indicates the output of the forget gate, W xf b represents the weight between the output gate and the forget gate. f This represents the bias between the hidden state and the forget gate. W represents the output of the neuron. xc W represents the bias between the input gate and the neuron. hc b represents the weight between the hidden state and the neuron. c C represents the bias between the hidden state and the neuron. t O represents the cell state at time t. t W represents the output of the output gate. xo W represents the weights between the input and output gates. ho b represents the weight between the hidden state and the output gate. ho This indicates the bias between the hidden state and the output gate;

[0122] X t Let p represent the input to the BiLSTM-Informer model, m represent the types of air pollutants, n represent the sample size of the m-th air pollutant concentration, Q represent the query matrix, O represent the key-value matrix, L represent the value matrix, and W represent the input to the BiLSTM-Informer model. q W O W L Let d represent the trainable weight matrix. k This represents the number of columns in the Q, O, and L matrices, i.e., the vector dimensions. Q m X' represents a local feature in the head attention mechanism. t This represents the different time series segments into which the input time series data is decomposed, C m S represents the feature weights of different time series. (m) This represents the self-attention weights between each time segment. U represents the output pollutant concentration at time t. m This represents the hyperparameter matrix in the BiLSTM-Informer model, which can be optimized to improve model accuracy and generalization ability. Softmax is a function in deep learning that can transform the original output of the model into a probability distribution. MultiHeadAttention is a multi-head self-attention mechanism used to process sequential data.

[0123] The air quality prediction module uses the BiLSTM-Informer model, which consists of an encoder, a decoder, and an output layer.

[0124] Sparse multi-head probabilistic sub-attention and distillation are key technologies for improving efficiency and performance in encoders. Sparse multi-head probabilistic sub-attention is an optimized attention mechanism that improves model efficiency by reducing unnecessary computation, especially when processing long sequences. Its core is to focus only on the most important parts of the sequence, reducing computation and mapping input features to different subspaces, enabling the model to simultaneously focus on information from different representation subspaces and enhancing the model's ability to capture information. Distillation is a model compression technique that transfers information from a complex model to a simpler one.

[0125] The BiLSTM-Informer model's decoder uses BiLSTM to process sequence data through two LSTM networks, forward and backward. This allows the model to simultaneously acquire information before and after each time point in the sequence, enhancing its understanding and prediction capabilities of time series data. The sparse multi-head probabilistic self-attention mechanism reduces computation by introducing a mask, helping the model concentrate resources on processing key information. The multi-head self-attention mechanism allows the model to capture information in different representation subspaces. The weights learned by each "head" represent the importance of different features, and these weights can be viewed as probability distributions, guiding the model to concentrate resources on processing the most relevant information.

[0126] In the output layer, the linear layer and the soft max function are two key components. Together, they are responsible for transforming the model's internal representation into the final classification result. The linear layer receives the output of the BiLSTM layer and maps its extracted features to the final prediction space. The soft max function is typically applied after the fully connected layer, which fuses and transforms the output features of the BiLSTM and Informer layers. Then, the soft max function transforms these transformed features into the final air quality prediction module result, using MAE, RMSE, and R... 2 As an evaluation metric for the model, it is used to quantify the difference between the model's predicted values ​​and the actual observed values, thereby assessing the model's accuracy. Performance evaluation metrics are not only used to evaluate model performance, but also to identify key factors that affect model performance and adjust the model accordingly to improve prediction accuracy.

[0127] (5) The cloud platform monitors the data of air quality sensors and monitoring stations in real time. Once the concentration of pollutants exceeds the preset threshold, the early warning mechanism is immediately triggered. Based on the air quality index or the concentration of specific pollutants, three warning levels are set: yellow warning, orange warning, and red warning. After the warning is issued, the information will be automatically sent to users via email, SMS, mobile application push, etc. The user interface of the cloud platform will intuitively display pollution hotspots and warning areas through visualization tools such as maps, charts and dashboards, so that users can plan their future travel methods or routes in advance and avoid the harm of air pollutants to themselves. In the event of an emergency, such as a toxic gas leak, the emergency management department can take emergency measures, formulate corresponding evacuation routes, provide auxiliary decision support, and ensure personnel safety.

Claims

1. A cloud-machine combined air quality intelligent decision system, characterized in that, include: The adaptive path planning module is used to design an adaptive objective function and utilizes a moss growth optimization algorithm improved by coupling chaotic mapping with a population adaptive adjustment strategy to perform optimal path planning for UAVs. A multi-source heterogeneous data acquisition module is used to collect and process air pollutant data and aerial image data. The multi-source heterogeneous data acquisition module includes a gas sensor, an ultrasonic sensor, and an accelerometer. The central processing unit module is used to receive data from the UAV's multi-source heterogeneous data acquisition module and transmit the data to the cloud platform using the wireless fidelity communication module. The wireless fidelity communication module is used to realize data transmission between the central processing unit module and the cloud platform; A cloud platform is used to provide auxiliary decision support. Specifically, the cloud platform includes: a data management module for cleaning and integrating collected data; an air quality prediction module for driving the BiLSTM-Informer model to predict future air quality based on current air quality conditions; a performance evaluation module for evaluating the performance of the gas sensors mounted on the UAV; a cloud platform display module for receiving and displaying the predicted future air quality results; and an air quality remote control module for remotely controlling the UAV's flight path and monitoring tasks, assessing environmental quality, and issuing warnings when pollutants exceed standards. The formula for calculating the adaptive objective function in the adaptive path planning module is as follows: ； ； ； ； ； ； in, Indicates the cost of the journey. Indicates the cost of threats. This indicates a high cost. and The coefficient represents the importance of the objective function. Dynamically adjust based on data collected by the ultrasonic sensor; and This represents the cost weighting coefficient for the drone's flight time and power consumption, where (n, n+1) represents the flight time from the nth waypoint to the (n+1th)th waypoint, and q represents the proportionality coefficient between unit power consumption and flight distance. This represents the total flight distance of the drone between any two points in the three-dimensional search space. This indicates the danger level of the nth flight node, sum_obstacle represents the number of obstacle grids around the waypoint, and sum_surround represents the total number of grids around the waypoint. This indicates the total number of waypoints. and This represents the weighting coefficients for exceeding the maximum height and falling below the minimum height. Weighting coefficients representing the rate of change of height. and This indicates the set maximum and minimum flight altitudes. This represents the nth track node of the drone flight aircraft; Indicates the maximum pitch angle of the drone. Indicates the maximum turning angle.

2. The intelligent air quality decision-making system according to claim 1, characterized in that, The formula for the BiLSTM-Informer model in the air quality prediction module is: ； ； ； ； ； ； ； in, This represents the input to the BiLSTM-Informer model, where p represents the spatial sampling point. m Let represent the types of air pollutants, and n represent the sample size of the concentration of the m-th air pollutant; Q represents the query matrix, O represents the key-value matrix, and L represents the value matrix. This represents a trainable weight matrix. This represents the number of columns in the Q, O, and L matrices, i.e., the vector dimension. This represents local features in the head attention mechanism. This represents the different time series segments into which the input time series data is decomposed. The feature weights represent the features of different time series. This represents the self-attention weights between each time segment. express t Output pollutant concentration at any given time This represents the hyperparameter matrix in the BiLSTM-Informer model.

3. The intelligent air quality decision-making system according to claim 1, characterized in that, The performance in the performance evaluation module includes: Used to evaluate the accuracy, sensitivity, response time, and stability of a drone's data acquisition capabilities; The data acquisition frequency, data real-time performance, and data processing efficiency are used to evaluate the data transmission and processing capabilities of drones.

4. A cloud-based intelligent decision-making method for air quality, characterized in that, Includes the following steps: (1) An adaptive objective function for the UAV is designed through an adaptive path planning module. The optimal path planning for the UAV is performed by using a moss growth optimization algorithm improved by a chaotic mapping coupled with a population adaptive adjustment strategy. (11) Initialize the UAV position using chaotic mapping, and determine the optimal path search direction based on the adaptive objective function value; (12) Update the UAV position according to the search direction and search step size; (13) Locally update the drone's position to realize the planning of the drone's optimal path; (2) Collect and process air pollutant data and aerial image data through a multi-source heterogeneous data processing module; (3) The central processing unit module receives data from the UAV multi-source heterogeneous data acquisition module and transmits the data to the cloud platform using the wireless fidelity communication module; (4) The cloud platform uses the BiLSTM-Informer prediction model to predict future air quality based on the current air quality conditions, evaluates the performance of the gas sensor carried by the UAV, displays the final prediction results of future air quality, and provides auxiliary decision support. (5) Real-time monitoring of air quality data is achieved through the air quality remote control module. When the concentration of pollutants exceeds the preset threshold, an early warning mechanism is triggered to display pollutant hotspots and warning areas. The chaotic mapping formula is: ； Where R represents the algorithm's search space variables, and r represents the control parameters. , , This indicates the optimized initial position of the drone. and This indicates the lower and upper limits of the algorithm's search range, and mod 1 indicates the modulo operation with respect to 1; The formula for the population adaptive adjustment strategy is: ； in, Indicates the worst-case drone location. This represents the optimal drone position, v represents the current objective function evaluation, and % represents the remainder or modulo operator. Indicates the current iteration number. This indicates the maximum number of iterations.

5. The intelligent air quality decision-making method according to claim 4, characterized in that, The formula for determining the optimal path search direction in step 11 is: ； ； Where f represents the adaptive objective function, This indicates the calculated search direction, and num represents the total number of alternative paths for the UAV in divV.

6. The intelligent air quality decision-making method according to claim 4, characterized in that, The formula for updating the UAV position based on the search direction and search step size, as described in step 12, is as follows: ； in, Indicates the current location of the drone. This indicates the calculated search direction, with the search step size determined by a random number. and Determined by their relative size; when At that time, a long-distance search is used: ; Here, w represents a constant parameter, which is set to 2. Let E represent a random variable in the range (0,1), and let E represent the standard search step size. when At that time, a short-distance search is used: ; in, It is a random vector within the range (0,1), which can control the search step size; , Indicates the current iteration number. This indicates the maximum number of iterations.

7. The intelligent air quality decision-making method according to claim 4, characterized in that, The formula for locally updating the UAV position in step 13 is as follows: ； in, Indicates the position of the i-th drone. and Represents a random number between (0,1) Determine whether to utilize the location update factor (act) to determine whether to use it. The formula for updating the drone's location is as follows: ； ； in, Let (represent) be a random vector in the range (0,1). It represents a random number in the range (0,1).

Images