Forest fire monitoring and early warning method and system based on artificial intelligence

Abstract

According to the forest fire monitoring and early warning method and system based on artificial intelligence provided by the invention, the forest state vector mapping model, the fire risk description model and the classification prediction model are constructed and integrated to form the fire risk prediction network, so that efficient and accurate prediction of the forest fire risk is realized; the real-time performance and accuracy of fire monitoring are remarkably improved, specifically, by converting complex forest state data into high-dimensional vector representation and combining fire risk description of category knowledge data, fire risk features can be carefully captured, and therefore the accuracy of fire risk prediction is improved. By constructing and optimizing the classification prediction model, when the classification prediction model receives new forest state data, prediction of fire risk categories can be rapidly completed, and possibility is provided for rapid response. Through automatic generation and application of the fire risk prediction network, intelligence of fire monitoring and early warning is realized, and manual intervention is reduced.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based forest fire monitoring and early warning method and system. Background Art

[0002] As a highly destructive disaster in nature, forest fire not only seriously threatens the safety of forest resources, but also has a huge impact on the ecological environment and human society. Traditional forest fire monitoring methods mainly rely on manual patrols, satellite remote sensing monitoring, and ground meteorological stations. Although these methods can monitor forest fires to a certain extent, they have problems such as limited monitoring range, poor real-time performance, and low early warning accuracy. Summary of the invention

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present application, an embodiment of the present application provides a forest fire monitoring and early warning method based on artificial intelligence, the method comprising: Obtain template forest status collection data and set fire risk category sequence; According to the forest state vector mapping model, the template forest state collection data is represented by a forest state vector to generate a forest state vector sequence of the template forest state collection data; For each set fire risk category corresponding to the category knowledge data, obtain the knowledge point description distribution corresponding to the category knowledge data, perform fire risk description on the knowledge point description distribution according to the fire risk description model, and generate a category description vector sequence of the set fire risk category; According to the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each of the category description vector sequences, parameter learning is performed on the classification prediction model to generate a classification prediction model that meets the model convergence requirements; Based on the forest state vector mapping model, the fire risk description model and the classification prediction model that meets the model convergence requirements, a fire risk prediction network is generated, and the fire risk prediction network is used to predict the fire risk category of the acquired target forest state collection data.

[0004] In a possible implementation manner of the first aspect, the forest state vector mapping model includes a plurality of feature space mapping units, and the template forest state collection data includes forest state collection data of a plurality of feature spaces; The step of performing forest state vector representation on the template forest state collection data according to the forest state vector mapping model to generate a forest state vector sequence of the template forest state collection data includes: Acquire the plurality of feature space mapping units, each feature space mapping unit corresponding to a feature space; Using the plurality of feature space mapping units, respectively perform state vector mapping on the forest state collection data of the feature space corresponding to the template forest state collection data, to generate a forest state vector sequence corresponding to the feature space of the template forest state collection data; In a possible implementation manner of the first aspect, the acquiring the multiple feature space mapping units includes: Acquire the reference feature space mapping units corresponding to the test template data and each feature space, each reference feature space mapping unit corresponds to a feature space; Through the reference feature space mapping units corresponding to the respective feature spaces, state vector mapping is performed on the forest state acquisition data of the corresponding feature space of the test template data to generate a forest state vector sequence of the independent feature space corresponding to the corresponding feature space of the test template data; For each forest state vector sequence in an independent feature space, classification prediction is performed based on the forest state vector sequence in the independent feature space and the category description vector sequence of each set fire risk category to generate a corresponding fire risk classification prediction vector, and parameter learning is performed on the classification prediction model based on each of the fire risk classification prediction vectors to generate multiple classification prediction models; Based on the forest state vector mapping model, the fire risk description model and the classification prediction model, generating multiple fire risk prediction networks; Based on the prediction performance indicators of the plurality of fire risk prediction networks, a plurality of feature space mapping units are selected from reference feature space mapping units corresponding to the respective feature spaces.

[0005] In a possible implementation of the first aspect, the method further includes: When the template forest status collection data includes the forest status collection data of the environment status feature space, but does not include the forest status collection data of other feature spaces, the forest status collection data of other feature spaces are generated based on the forest status collection data of the environment status feature space, and the forest status collection data of multiple feature spaces included in the template forest status collection data are generated, wherein the forest status collection data of other feature spaces include at least one of the forest status collection data of the soil feature space, the forest status collection data of the meteorological feature space and the forest status collection data of the biological feature space.

[0006] In a possible implementation of the first aspect, the method further includes: For each set fire risk category in the set fire risk category sequence, according to the correspondence between the fire risk category and the category knowledge data defined previously, based on the set fire risk category, the category knowledge data corresponding to the set fire risk category is retrieved.

[0007] In a possible implementation of the first aspect, the method further includes: For each set fire risk category in the set fire risk category sequence, generating guidance data for search query based on the set fire risk category; A neural network model is called based on the guidance data, and the category knowledge data corresponding to the set fire risk category is generated using the neural network model.

[0008] In a possible implementation of the first aspect, the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each of the category description vector sequences, performing parameter learning on the classification prediction model to generate a classification prediction model that meets the model convergence requirements, includes: The forest state vector sequence is used as a reference feature, and each of the category description vector sequences is used as a retrieval feature, and loaded into a classification prediction model based on pattern recognition in the fire risk prediction network, and the classification prediction model is used to perform a correlation analysis on the reference feature and each of the retrieval features to generate a fire risk classification prediction vector corresponding to each set fire risk category; The fire risk classification prediction vector corresponding to each set fire risk category is input into an independent fire risk category prediction task or a combined fire risk category prediction task, and the model weight information of the classification prediction model is optimized based on the prediction result to generate a classification prediction model that meets the model convergence requirements.

[0009] In a possible implementation of the first aspect, the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each of the category description vector sequences, performing parameter learning on the classification prediction model to generate a classification prediction model that meets the model convergence requirements, includes: Acquiring a fire risk category carried by the template forest state collection data, wherein the fire risk category carried by the template forest state collection data belongs to the set fire risk category sequence; Based on the fire risk category carried by the template forest state collection data, determining the fire risk category supervision data corresponding to each set fire risk category of the template forest state collection data; Calculating the category confidence of each set fire risk category corresponding to the template forest state collection data according to the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each category description vector sequence; Based on the fire risk category supervision data and the category confidence, a training error parameter is constructed, and the model weight information of the classification prediction model is optimized with minimizing the training error parameter as the optimization direction to generate a classification prediction model that meets the model convergence requirements.

[0010] In a possible implementation of the first aspect, the method includes: Obtain target forest status data and complete the trained fire risk prediction network; According to the forest state vector mapping model in the fire risk prediction network, the target forest state collection data is represented by a forest state vector to generate a forest state vector sequence; According to the fire risk description model in the fire risk prediction network, the category knowledge data corresponding to each set fire risk category in the set fire risk category sequence are respectively described for fire risk, and a category description vector sequence of each set fire risk category is generated; According to the classification prediction model in the fire risk prediction network, the forest state vector sequence and the category description vector sequence of each set fire risk category are classified and predicted to generate a fire risk classification prediction vector, and the fire risk category of the target forest state collection data is determined based on the fire risk classification prediction vector.

[0011] In a possible implementation of the first aspect, the method further includes: When an extended fire risk category appears in the set fire risk category sequence, obtaining category knowledge data corresponding to the extended fire risk category, the extended fire risk category being a fire risk category that has not appeared in the parameter learning process of the fire risk prediction network; Performing fire risk description on the category knowledge data corresponding to the extended fire risk category to generate a category description vector sequence corresponding to the extended fire risk category; The method comprises: performing classification prediction on the forest state vector sequence and the category description vector sequence of each set fire risk category according to the classification prediction model in the fire risk prediction network, generating a fire risk classification prediction vector, and determining the fire risk category of the target forest state collection data according to the fire risk classification prediction vector, including: The classification prediction model in the fire risk prediction network is used to classify and predict the forest state vector sequence, the category description vector sequence of each set fire risk category, and the category description vector sequence of the extended fire risk category to generate a fire risk classification prediction vector, and the fire risk category of the target forest state collection data is determined based on the fire risk classification prediction vector.

[0012] On the other hand, an embodiment of the present application also provides an artificial intelligence-based forest fire monitoring and early warning system, including a processor and a machine-readable storage medium, the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0013] Based on the above aspects, the embodiment of the present application realizes efficient and accurate prediction of forest fire risks by constructing a forest state vector mapping model, a fire risk description model and a classification prediction model, and integrating them to form a fire risk prediction network, which significantly improves the real-time and accuracy of fire monitoring, and can issue early warnings in time before a fire occurs, effectively reducing the frequency and degree of harm of forest fires. Specifically, the complex forest state data is converted into a high-dimensional vector representation using deep learning technology, and combined with the fire risk description of the category knowledge data, the model can capture the fire risk characteristics more carefully, thereby improving the accuracy of fire risk prediction. By constructing and optimizing the classification prediction model, the classification prediction model can quickly complete the prediction of the fire risk category when receiving new forest state data, which provides the possibility for rapid response. The automatic generation and application of the fire risk prediction network realizes the intelligence of fire monitoring and early warning, reduces manual intervention, and improves the automation level and operation efficiency of the early warning system. As a result, the effectiveness of forest fire monitoring and early warning is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the execution flow of the forest fire monitoring and early warning method based on artificial intelligence provided in the embodiment of the present application.

[0015] Figure 2 This is a schematic diagram of the hardware architecture of the artificial intelligence-based forest fire monitoring and early warning system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of an artificial intelligence-based forest fire monitoring and early warning method provided in an embodiment of the present application. The artificial intelligence-based forest fire monitoring and early warning method is introduced in detail below.

[0017] Step S110, obtaining template forest status collection data and setting a fire risk category sequence.

[0018] In detail, the template forest state collection data refers to a part of real-time data extracted from the forest environment monitoring system, which is used to train the fire risk prediction model. These template forest state collection data contain various state parameters of the forest, such as temperature, humidity, wind force, wind direction, vegetation type, soil moisture, etc., which can fully reflect the current state of the forest. For example, assuming that the forest environment monitoring system uploads data every hour, the server selects data from these data within one day as template forest state collection data. These data include temperature 20°C, humidity 60%, wind speed 3m / s at 8 o'clock in the morning, temperature 25°C, humidity 50%, wind speed 5m / s at 12 o'clock in the afternoon, and temperature 15°C, humidity 70%, wind speed 2m / s at night, etc., and also include information such as vegetation type distribution and soil moisture at the corresponding time.

[0019] The set fire risk category sequence refers to a set of fire risk categories pre-defined based on historical fire data and expert experience, which are used to classify forest fire risks. These categories usually include different risk levels, such as "low risk", "medium risk", "high risk" and "extremely high risk". For example, the set fire risk category sequence may include "low risk" (corresponding to situations where the probability of fire is extremely low), "medium risk" (corresponding to situations where the probability of fire is average), "high risk" (corresponding to situations where the probability of fire is high) and "extremely high risk" (corresponding to situations where the probability of fire is extremely high and may spread rapidly).

[0020] That is, in this embodiment, the server is first connected to the forest environment monitoring system, which regularly collects and uploads forest status data. The server extracts a part of these real-time data as template forest status collection data, which will be used for subsequent training and model building.

[0021] At the same time, the server also obtains a predefined set fire risk category sequence, which includes all possible set fire risk categories. These set fire risk categories are pre-set based on historical fire data and expert experience.

[0022] Step S120: performing forest state vector representation on the template forest state collection data according to the forest state vector mapping model to generate a forest state vector sequence of the template forest state collection data.

[0023] In detail, the forest state vector mapping model is a deep learning network that maps the original forest state data into a high-dimensional vector space. This forest state vector mapping model is able to capture the complex characteristics of the forest state and convert it into a vector form that is more suitable for subsequent analysis and prediction. For example, suppose the original forest state data is a table containing multiple parameters such as temperature, humidity, wind speed, etc. The forest state vector mapping model will input this data into the network and output a high-dimensional vector through the network's multi-layer processing (such as convolutional layer, pooling layer, fully connected layer, etc.). This vector contains all the information in the original data, but in a more compact and easier to process form.

[0024] That is, in this embodiment, the server loads a pre-trained forest state vector mapping model, which is a deep learning network that can map the original forest state data into a high-dimensional vector space to better represent the complex state of the forest.

[0025] The server inputs the template forest state collection data into the forest state vector mapping model. The forest state vector mapping model processes each data point and outputs a corresponding forest state vector. In this way, the server obtains a sequence composed of forest state vectors corresponding to all template data points. This sequence will be used for subsequent classification tasks.

[0026] Step S130, for each set fire risk category corresponding to the category knowledge data, obtain the knowledge point description distribution corresponding to the category knowledge data, perform fire risk description on the knowledge point description distribution according to the fire risk description model, and generate a category description vector sequence of the set fire risk category.

[0027] In detail, the category knowledge data refers to a data set associated with each set fire risk category, which describes the definition, characteristics, historical cases, countermeasures and other information of the risk category. The category knowledge data is the input of the fire risk description model and is used to generate a category description vector. For example, for the "high risk" fire risk category, its category knowledge data may include data on multiple high-risk fire cases that have occurred in history, such as weather conditions when the fire occurred, vegetation types, fire spread speed, etc. These data are used to describe the characteristics of the "high risk" category.

[0028] The fire risk description model is a deep learning network that is used to generate a category description vector based on category knowledge data. This model can capture the key features in the category knowledge data and convert it into a vector form that can accurately describe the fire risk category. For example, the fire risk description model receives the category knowledge data of the "high risk" category as input, and outputs a vector describing the "high risk" category through network processing (such as embedding layer, attention mechanism, fully connected layer, etc.). This vector contains all the key feature information of the category.

[0029] That is, in this embodiment, for each set fire risk category, the server stores a set of category knowledge data related thereto, and these category knowledge data include the definition, characteristics, historical cases, etc. of the risk category. The server first analyzes these category knowledge data, extracts key knowledge points, and determines the distribution of these knowledge points in describing the risk category.

[0030] Next, the server uses the fire risk description model, which is also a deep learning network that can generate a vector that can accurately describe the fire risk category based on the knowledge point description distribution. The server performs this process for each set fire risk category, and finally obtains a sequence of description vectors of each category.

[0031] Step S140, based on the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each of the category description vector sequences, parameter learning is performed on the classification prediction model to generate a classification prediction model that meets the model convergence requirements.

[0032] In detail, the classification prediction model is a model for predicting the corresponding fire risk category based on the forest state vector. This classification prediction model evaluates the degree of association between the current state of the forest and each fire risk category by calculating the similarity or distance between the forest state vector and each category description vector, and generates a fire risk classification prediction vector. For example, the classification prediction model receives the forest state vector and each category description vector as input, and generates a fire risk classification prediction vector by calculating the similarity or distance between them (such as cosine similarity, Euclidean distance, etc.). Each element in this fire risk classification prediction vector represents the possibility that the current state of the forest belongs to the corresponding fire risk category. That is, the fire risk classification prediction vector is a probability distribution vector that represents the possibility that the current state of the forest belongs to each set fire risk category. Each element in this vector corresponds to a fire risk category, and its value represents the probability that the category is predicted to be a true category. For example, suppose the classification prediction model generates a fire risk classification prediction vector [0.1, 0.3, 0.5, 0.1], which indicates that the probability that the current state of the forest belongs to "low risk" is 0.1, the probability that it belongs to "medium risk" is 0.3, the probability that it belongs to "high risk" is 0.5 (highest), and the probability that it belongs to "extremely high risk" is 0.1. Therefore, the model predicts that the current state of the forest belongs to the "high risk" category.

[0033] That is, in this embodiment, the server now has two sets of vector sequences: one set is the forest state vector sequence, and the other set is the category description vector sequence. The server inputs these two sets of sequences into the classification prediction model, and the task of this classification prediction model is to predict the corresponding fire risk category based on the forest state vector.

[0034] The classification prediction model performs correlation analysis on each pair of forest state vectors and category description vectors to generate a fire risk classification prediction vector, which represents the possibility that the classification prediction model believes that the forest state belongs to each fire risk category.

[0035] Next, the server uses the fire risk category carried by the template forest state collection data as supervision data to learn the parameters of the classification prediction model. By continuously adjusting the weights and parameters of the classification prediction model, the prediction results of the model are closer and closer to the actual fire risk category. This process continues until the classification prediction model reaches the convergence requirement, that is, the prediction performance of the classification prediction model is no longer significantly improved.

[0036] Step S150, generating a fire risk prediction network based on the forest state vector mapping model, the fire risk description model and the classification prediction model that meets the model convergence requirements, wherein the fire risk prediction network is used to predict the fire risk category of the acquired target forest state collection data.

[0037] Finally, the server combines the trained forest state vector mapping model, fire risk description model and classification prediction model into a complete fire risk prediction network. The fire risk prediction network can receive new target forest state collection data, and through a series of processing and prediction, output the fire risk category corresponding to the target forest state collection data. In this way, the server completes the construction of an intelligent system that can predict forest fire risks in real time. This system can regularly receive new forest state data and output fire risk warnings in a timely manner, providing strong support for forest management and fire prevention and control.

[0038] Based on the above steps, the embodiment of the present application realizes efficient and accurate prediction of forest fire risks by constructing a forest state vector mapping model, a fire risk description model and a classification prediction model, and integrating them to form a fire risk prediction network, which significantly improves the real-time and accuracy of fire monitoring, and can issue early warnings in time before a fire occurs, effectively reducing the frequency and degree of harm of forest fires. Specifically, the complex forest state data is converted into a high-dimensional vector representation using deep learning technology, and combined with the fire risk description of the category knowledge data, the model can capture the fire risk characteristics more carefully, thereby improving the accuracy of fire risk prediction. By constructing and optimizing the classification prediction model, the classification prediction model can quickly complete the prediction of the fire risk category when receiving new forest state data, which provides the possibility for rapid response. The automatic generation and application of the fire risk prediction network realizes the intelligence of fire monitoring and early warning, reduces manual intervention, and improves the automation level and operation efficiency of the early warning system. As a result, the effectiveness of forest fire monitoring and early warning is significantly improved.

[0039] In a possible implementation manner, the forest state vector mapping model includes a plurality of feature space mapping units, and the template forest state collection data includes forest state collection data of a plurality of feature spaces.

[0040] Step S120 includes: Step S121, obtaining the plurality of feature space mapping units, each feature space mapping unit corresponding to a feature space.

[0041] Step S122, using the plurality of feature space mapping units, respectively perform state vector mapping on the forest state collection data of the feature spaces corresponding to the template forest state collection data, to generate a forest state vector sequence corresponding to the feature spaces of the template forest state collection data.

[0042] In this embodiment, the forest state vector mapping model has the ability to process multi-feature space data. Specifically, the forest state vector mapping model includes multiple feature space mapping units, each of which corresponds to a specific feature space, such as environmental state feature space, soil feature space, meteorological feature space, and biological feature space.

[0043] When the server receives the template forest state collection data, these template forest state collection data contain comprehensive information of multiple feature spaces. For example, temperature and humidity belong to the environmental state feature space, soil moisture belongs to the soil feature space, wind speed and wind direction belong to the meteorological feature space, and vegetation type may belong to the biological feature space.

[0044] In order to generate a forest state vector sequence, the server extracts all feature space mapping units from a pre-trained forest state vector mapping model. These feature space mapping units are part of the forest state vector mapping model and are specifically used to map data in different feature spaces into a high-dimensional vector space.

[0045] The server then uses these feature space mapping units to map the data in the corresponding feature spaces in the template forest state collection data. For example, the data in the environmental state feature space (such as temperature and humidity) will be sent to the corresponding environmental state feature space mapping unit to generate a vector representation of the environmental state. Similarly, the data in the soil feature space, meteorological feature space, and biological feature space will also be sent to their respective mapping units to generate corresponding vector representations.

[0046] After the above mapping process, the server obtains the forest state vectors of multiple feature spaces. These forest state vectors together constitute the forest state vector sequence of the template forest state collection data. This sequence is a high-dimensional vector set containing complex forest state information, which will be used for subsequent classification and prediction tasks.

[0047] In a possible implementation, step S121 includes: Step S1211, obtaining reference feature space mapping units corresponding to the test template data and each feature space, each reference feature space mapping unit corresponds to a feature space.

[0048] Step S1212, through the reference feature space mapping unit corresponding to each feature space, the forest state collection data of the corresponding feature space of the test template data is respectively mapped to the state vector to generate the forest state vector sequence of the independent feature space corresponding to the corresponding feature space of the test template data.

[0049] Step S1213, for each forest state vector sequence in an independent feature space, classification prediction is performed based on the forest state vector sequence in the independent feature space and the category description vector sequence of each set fire risk category to generate a corresponding fire risk classification prediction vector, and parameter learning is performed on the classification prediction model based on each of the fire risk classification prediction vectors to generate multiple classification prediction models.

[0050] Step S1214, generating multiple fire risk prediction networks based on the forest state vector mapping model, the fire risk description model and the classification prediction model.

[0051] Step S1215 , based on the prediction performance indicators of the multiple fire risk prediction networks, multiple feature space mapping units are selected from the reference feature space mapping units corresponding to the respective feature spaces.

[0052] In this embodiment, the server first reads the test template data from the database. These test template data contain real-time collected information of different feature spaces of the forest (such as temperature, humidity, wind speed, etc.). At the same time, the server loads predefined reference feature space mapping units corresponding to each feature space. These reference feature space mapping units are established based on historical data and expert knowledge. Each reference feature space mapping unit corresponds to a specific feature space and can convert the data of the feature space into a state vector that can be used for subsequent analysis.

[0053] The server uses the loaded reference feature space mapping unit to process the forest state collection data of each feature space in the test template data. For example, for the temperature feature space, the server will apply the corresponding temperature feature space mapping unit to convert the temperature data into a temperature state vector. This process will perform a similar conversion on each feature space in the test template data to generate a series of forest state vector sequences of independent feature spaces.

[0054] For each forest state vector sequence of an independent feature space, the server uses a preset algorithm (such as a machine learning algorithm) to perform classification predictions. The server compares these state vector sequences with the preset category description vector sequences of various fire risk categories to generate corresponding fire risk classification prediction vectors. Based on these classification prediction vectors, the server performs parameter learning on the classification prediction model and adjusts the model parameters to better reflect the relationship between the actual fire risk and the feature space state vector. This process generates a classification prediction model for each feature space.

[0055] The server integrates the forest state vector mapping model, fire risk description model and classification prediction model to form multiple fire risk prediction networks. Each fire risk prediction network contains a complete mapping and prediction logic from feature space state vector to fire risk classification.

[0056] The server evaluates the prediction performance indicators of each fire risk prediction network, such as accuracy, recall, etc. Based on these prediction performance indicators, the server selects multiple feature space mapping units with the best performance from the reference feature space mapping units corresponding to each feature space as the final feature space mapping unit set. These selected feature space mapping units will be used in subsequent forest fire risk prediction and monitoring tasks.

[0057] In a possible implementation, the method further includes: when the template forest status collection data includes forest status collection data of an environmental status feature space and does not include forest status collection data of other feature spaces, generating forest status collection data of other feature spaces based on the forest status collection data of the environmental status feature space, and generating forest status collection data of multiple feature spaces included in the template forest status collection data, wherein the forest status collection data of other feature spaces includes at least one of forest status collection data of a soil feature space, forest status collection data of a meteorological feature space and forest status collection data of a biological feature space.

[0058] In this embodiment, when processing the template forest state collection data, the server first checks whether the template forest state collection data contains all necessary feature space information. If it is found that the template forest state collection data only includes the forest state collection data of the environmental state feature space, but lacks at least one of the soil feature space, the meteorological feature space or the biological feature space, the server will start the data generation program.

[0059] The server uses the existing environmental state feature space data (such as temperature, humidity, etc.) combined with the preset association model and algorithm to estimate or simulate the missing feature space data. For example, if the soil feature space data is missing, the server may use the humidity and temperature data in the environmental state, combined with the known relationship between soil humidity and temperature, to generate the soil feature space data.

[0060] Through the above estimation or simulation process, the server generates corresponding forest state collection data for the missing feature space. These generated data, together with the original environmental state feature space data, constitute the complete template forest state collection data, including forest state collection information of multiple feature spaces. With the complete template forest state collection data, the server can continue to perform subsequent analysis and prediction tasks, such as state vector mapping, fire risk classification prediction, etc.

[0061] In a possible implementation, the method further includes: for each set fire risk category in the set fire risk category sequence, according to the correspondence between the previously defined fire risk category and the category knowledge data, based on the set fire risk category, retrieving the category knowledge data corresponding to the set fire risk category.

[0062] In this embodiment, the server first identifies a defined set fire risk category sequence, which includes all possible fire risk categories, such as "low risk", "medium risk" and "high risk".

[0063] The server accesses a pre-established database or knowledge base, which stores the correspondence between fire risk categories and category knowledge data. Category knowledge data may include the definition, characteristics, historical cases, countermeasures and other information of fire risks.

[0064] The server begins to traverse each set fire risk category in the set fire risk category sequence. For each set fire risk category in the set fire risk category sequence, the server retrieves the category knowledge data associated with the category according to the correspondence between the fire risk category and the category knowledge data. After the server retrieves the category knowledge data from the knowledge base, it extracts the category knowledge data and stores them in the server's memory or temporary database. The category knowledge data will be used for subsequent analysis, prediction or decision support tasks.

[0065] After having the category knowledge data corresponding to the set fire risk category, the server can continue to perform subsequent tasks, such as conducting more in-depth analysis based on this data, generating reports, or providing decision support for users.

[0066] In a possible implementation, the method further includes: Step A110 : for each set fire risk category in the set fire risk category sequence, generating guidance data for search query based on the set fire risk category.

[0067] Step A120, calling a neural network model based on the guidance data, and using the neural network model to generate category knowledge data corresponding to the set fire risk category.

[0068] In this embodiment, the server first identifies a defined set fire risk category sequence, which includes all possible fire risk categories, such as "low risk", "medium risk" and "high risk". For each set fire risk category in the set fire risk category sequence, the server generates a guide data related to the category. The guide data is a structured or formatted data that contains key information that can uniquely identify or describe the fire risk category, such as the name, number or feature description of the category.

[0069] The server has a pre-trained neural network model that can generate corresponding category knowledge data based on the input guidance data. The server uses the generated guidance data as input and calls the neural network model for inference or prediction. After receiving the guidance data, the neural network model starts internal calculation and processing. The neural network model converts the guidance data into corresponding category knowledge data through its complex network structure and algorithm. The generated category knowledge data can include detailed descriptions of fire risk categories, characteristics, historical cases, associated factors, countermeasures and other information.

[0070] After the neural network model generates category knowledge data, the server extracts these category knowledge data and stores them in the server's memory or database. These category knowledge data will be used for subsequent analysis, prediction, decision support or to provide detailed information display to users.

[0071] After having the category knowledge data corresponding to the set fire risk category, the server can continue to perform subsequent tasks, such as conducting more in-depth analysis based on this data, generating reports, providing decision support to users, or responding to user query requests.

[0072] In a possible implementation, step S140 includes: Step S141, taking the forest state vector sequence as a reference feature and each of the category description vector sequences as a retrieval feature, loading them into a classification prediction model based on pattern recognition in the fire risk prediction network, using the classification prediction model to perform a correlation analysis on the reference feature and each of the retrieval features, and generating a fire risk classification prediction vector corresponding to each set fire risk category.

[0073] Step S142, input the fire risk classification prediction vector corresponding to each set fire risk category into an independent fire risk category prediction task or a combined fire risk category prediction task, optimize the model weight information of the classification prediction model based on the prediction result, and generate a classification prediction model that meets the model convergence requirements.

[0074] In this embodiment, the server uses the previously generated forest state vector sequences as reference features. These vector sequences contain the state information of the forest in different feature spaces. At the same time, the server also loads the category description vector sequences corresponding to each set fire risk category into the model as retrieval features. These category description vector sequences describe the characteristics of different fire risk categories. The server loads these data into the classification prediction model based on pattern recognition and prepares for correlation analysis.

[0075] The classification prediction model starts to process the loaded data and performs correlation analysis on the forest state vector sequence and the category description vector sequence. The classification prediction model evaluates the degree of association between the current state of the forest and each fire risk category by calculating the similarity or distance between vectors. After the analysis is completed, the classification prediction model generates a fire risk classification prediction vector for each set fire risk category. This fire risk classification prediction vector reflects the possibility that the current state of the forest belongs to this category.

[0076] The server inputs the generated fire risk classification prediction vectors corresponding to each set fire risk category into an independent fire risk category prediction task or a combined fire risk category prediction task. In an independent prediction task, the prediction vectors of each category are processed separately to generate a prediction result for that category. In a combined prediction task, the prediction vectors of all categories are comprehensively considered to generate a comprehensive prediction result.

[0077] The server evaluates the performance of the classification prediction model based on the output of the prediction task. If the prediction result is inaccurate or the model does not meet the convergence requirements, the server will optimize the model weight information of the classification prediction model based on the prediction error. The optimization process can include adjusting weight parameters, changing the model structure, or adding training data, etc., in order to improve the prediction accuracy and stability of the model.

[0078] After multiple iterations and optimizations, the server finally generated a classification prediction model that met the model convergence requirements. This classification prediction model can accurately predict the corresponding fire risk category based on the forest state vector sequence, providing strong support for forest fire risk prediction and monitoring.

[0079] In a possible implementation, step S140 includes: Step S143, obtaining the fire risk category carried by the template forest state collection data, wherein the fire risk category carried by the template forest state collection data belongs to the set fire risk category sequence.

[0080] Step S144, based on the fire risk category carried by the template forest state collection data, determining the fire risk category supervision data corresponding to each set fire risk category of the template forest state collection data.

[0081] Step S145, calculating the category confidence of each set fire risk category corresponding to the template forest state collection data based on the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each category description vector sequence.

[0082] Step S146, constructing a training error parameter based on the fire risk category supervision data and the category confidence, and optimizing the model weight information of the classification prediction model with minimizing the training error parameter as the optimization direction, to generate a classification prediction model that meets the model convergence requirements.

[0083] In this embodiment, the server first accesses a database storing template forest state collection data. For each template forest state collection data, the server extracts the fire risk category label carried by it, which indicates the fire risk category to which the template forest state collection data belongs. The server confirms that these extracted fire risk categories belong to a predefined set fire risk category sequence, such as "low risk", "medium risk" and "high risk".

[0084] The server generates supervision data for each set fire risk category based on the fire risk category label of the template forest state collection data. The supervision data is a binary vector, where the position corresponding to the fire risk category of the template data is 1, and the other positions are 0. For example, if the fire risk category of the template data is "medium risk", then the supervision data corresponding to "low risk" is 0, the supervision data for "medium risk" is 1, and the supervision data for "high risk" is also 0.

[0085] The server uses the previously generated fire risk classification prediction vectors, which reflect the likelihood that the forest state belongs to each set fire risk category. For each template data, the server calculates the category confidence corresponding to each set fire risk category, that is, the value of the corresponding category in the prediction vector. The category confidence indicates the model's confidence in the prediction that the template data belongs to each set fire risk category.

[0086] The server constructs a training error parameter based on the fire risk category supervision data and category confidence. The training error parameter is the difference or loss between the supervision data and the category confidence, which is used to measure the accuracy of the model prediction. The server optimizes the model weight information of the classification prediction model using gradient descent or other optimization algorithms with the minimization of the training error parameter as the optimization direction. The optimization process includes adjusting the model weights to reduce the prediction error and improve the model's prediction accuracy for the fire risk category.

[0087] After multiple iterations and optimizations, the server generates a classification prediction model that meets the model convergence requirements. The model can accurately predict the corresponding fire risk category based on the forest state vector sequence and has stable prediction performance.

[0088] In one possible implementation, the method includes: Step S160, acquiring target forest status collection data and completing the trained fire risk prediction network.

[0089] Step S170, based on the forest state vector mapping model in the fire risk prediction network, the target forest state collection data is represented by a forest state vector to generate a forest state vector sequence.

[0090] Step S180, based on the fire risk description model in the fire risk prediction network, respectively describe the fire risk of the category knowledge data corresponding to each set fire risk category in the set fire risk category sequence, and generate a category description vector sequence for each set fire risk category.

[0091] Step S190, based on the classification prediction model in the fire risk prediction network, classify and predict the forest state vector sequence and the category description vector sequence of each set fire risk category, generate a fire risk classification prediction vector, and determine the fire risk category of the target forest state collection data based on the fire risk classification prediction vector.

[0092] In this embodiment, the server obtains real-time target forest state collection data from the forest state monitoring system. These target forest state collection data include environmental parameters such as forest temperature, humidity, wind speed, etc. At the same time, the server loads the trained fire risk prediction network, which contains multiple models for forest state vector representation, fire risk description and classification prediction.

[0093] The server uses the forest state vector mapping model in the fire risk prediction network to process the acquired target forest state data. The forest state vector mapping model converts the original forest state data into a forest state vector, which is a point in a high-dimensional space and can more accurately represent the current state of the forest. The server generates a vector sequence of the target forest state for subsequent fire risk prediction.

[0094] The server processes the predefined sequence of set fire risk categories according to the fire risk description model in the fire risk prediction network. The fire risk description model converts each set fire risk category (such as "low risk", "medium risk" and "high risk") into a corresponding category description vector, which is a point in a high-dimensional space that can accurately describe the characteristics of different fire risk categories. The server generates a sequence of category description vectors for each set fire risk category for subsequent classification prediction.

[0095] The server uses the classification prediction model in the fire risk prediction network to process the previously generated forest state vector sequence and each category description vector sequence. The classification prediction model calculates the similarity or distance between the forest state vector and each category description vector, and generates a fire risk classification prediction vector based on these calculation results. The fire risk classification prediction vector is a probability distribution that represents the probability that the target forest state belongs to each set fire risk category. The server determines the fire risk category of the target forest state collection data based on the maximum value or threshold in the fire risk classification prediction vector.

[0096] For example, if the predicted probability of the "high risk" category is the highest and exceeds a preset threshold, the server determines the fire risk category of the target forest state as "high risk".

[0097] In a possible implementation, the method further includes: Step A110, when an extended fire risk category appears in the set fire risk category sequence, category knowledge data corresponding to the extended fire risk category is obtained, and the extended fire risk category is a fire risk category that has not appeared in the parameter learning process of the fire risk prediction network.

[0098] Step A120, performing fire risk description on the category knowledge data corresponding to the extended fire risk category, and generating a category description vector sequence corresponding to the extended fire risk category.

[0099] Step S190 includes: using the classification prediction model in the fire risk prediction network to classify and predict the forest state vector sequence, the category description vector sequence of each set fire risk category, and the category description vector sequence of the extended fire risk category, generate a fire risk classification prediction vector, and determine the fire risk category of the target forest state collection data based on the fire risk classification prediction vector.

[0100] In this embodiment, during the parameter learning process of the fire risk prediction network, the server has learned the categories in the set fire risk category sequence. However, in actual applications, new, unlearned fire risk categories may appear, namely, extended fire risk categories. When the server detects that an extended fire risk category appears in the set fire risk category sequence, it will obtain the category knowledge data corresponding to the new category, which may come from an expert system, historical cases or other reliable data sources.

[0101] The server uses the fire risk description model in the fire risk prediction network to process the category knowledge data corresponding to the extended fire risk category. The fire risk description model converts this data into points in a high-dimensional space, namely, category description vectors, which can accurately describe the characteristics of the extended fire risk category. The server generates a sequence of category description vectors corresponding to the extended fire risk category for subsequent classification prediction.

[0102] When making classification predictions, the server not only considers the set fire risk categories that have been learned, but also considers the newly emerging extended fire risk categories. The server uses the classification prediction model in the fire risk prediction network to process the forest state vector sequence, the category description vector sequence of each set fire risk category, and the category description vector sequence of the extended fire risk category. The classification prediction model calculates the similarity or distance between the forest state vector and all category description vectors, and generates a fire risk classification prediction vector based on these calculation results. The fire risk classification prediction vector is a probability distribution that represents the probability that the target forest state belongs to each set fire risk category and the extended fire risk category.

[0103] The server determines the fire risk category of the target forest status collection data based on the maximum value or threshold in the fire risk classification prediction vector. The category can be a set fire risk category that has been learned or a newly emerging extended fire risk category.

[0104] Figure 2 The hardware structure of the artificial intelligence-based forest fire monitoring and early warning system 100 provided in the embodiment of the present application for implementing the above-mentioned artificial intelligence-based forest fire monitoring and early warning method is shown, as shown in FIG. Figure 2 As shown, the artificial intelligence-based forest fire monitoring and early warning system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0105] In one possible design, the artificial intelligence-based forest fire monitoring and early warning system 100 can be a single server or a server group. The server group can be centralized or distributed (for example, the artificial intelligence-based forest fire monitoring and early warning system 100 can be a distributed system). In some embodiments, the artificial intelligence-based forest fire monitoring and early warning system 100 can be local or remote. For example, the artificial intelligence-based forest fire monitoring and early warning system 100 can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the artificial intelligence-based forest fire monitoring and early warning system 100 can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the artificial intelligence-based forest fire monitoring and early warning system 100 can be implemented on an artificial intelligence-based forest fire monitoring and early warning system. By way of example only, the artificial intelligence-based forest fire monitoring and early warning system can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, or the like, or any combination thereof.

[0106] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the artificial intelligence-based forest fire monitoring and early warning system 100 uses to execute or use to complete the exemplary method described in this application.

[0107] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the artificial intelligence-based forest fire monitoring and early warning method of the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0108] The specific implementation process of the processor 110 can be found in the various method embodiments executed by the above-mentioned artificial intelligence-based forest fire monitoring and early warning system 100. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0109] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the above-mentioned forest fire monitoring and early warning method based on artificial intelligence is implemented.

[0110] It should be noted that in order to simplify the description disclosed in the present application and thereby facilitate the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, various features are sometimes grouped into one embodiment, drawing or description thereof.

Claims

1. A forest fire monitoring and early warning method based on artificial intelligence, characterized in that: The method comprises: Obtain template forest status collection data and set fire risk category sequence; According to the forest state vector mapping model, the template forest state collection data is represented by a forest state vector to generate a forest state vector sequence of the template forest state collection data; For each set fire risk category corresponding to the category knowledge data, obtain the knowledge point description distribution corresponding to the category knowledge data, perform fire risk description on the knowledge point description distribution according to the fire risk description model, and generate a category description vector sequence of the set fire risk category; According to the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each of the category description vector sequences, parameter learning is performed on the classification prediction model to generate a classification prediction model that meets the model convergence requirements; Based on the forest state vector mapping model, the fire risk description model and the classification prediction model that meets the model convergence requirements, a fire risk prediction network is generated, and the fire risk prediction network is used to predict the fire risk category of the acquired target forest state collection data.

2. The artificial intelligence-based forest fire monitoring and early warning method according to claim 1 is characterized in that: The forest state vector mapping model includes a plurality of feature space mapping units, and the template forest state collection data includes forest state collection data of a plurality of feature spaces; The step of performing forest state vector representation on the template forest state collection data according to the forest state vector mapping model to generate a forest state vector sequence of the template forest state collection data includes: Acquire the plurality of feature space mapping units, each feature space mapping unit corresponding to a feature space; Using the plurality of feature space mapping units, respectively perform state vector mapping on the forest state collection data of the feature space corresponding to the template forest state collection data, to generate a forest state vector sequence corresponding to the feature space of the template forest state collection data; Wherein, the obtaining of the plurality of feature space mapping units comprises: Acquire the reference feature space mapping units corresponding to the test template data and each feature space, each reference feature space mapping unit corresponds to a feature space; Through the reference feature space mapping units corresponding to the respective feature spaces, state vector mapping is performed on the forest state acquisition data of the corresponding feature space of the test template data to generate a forest state vector sequence of the independent feature space corresponding to the corresponding feature space of the test template data; For each forest state vector sequence in an independent feature space, classification prediction is performed based on the forest state vector sequence in the independent feature space and the category description vector sequence of each set fire risk category to generate a corresponding fire risk classification prediction vector, and parameter learning is performed on the classification prediction model based on each of the fire risk classification prediction vectors to generate multiple classification prediction models; Based on the forest state vector mapping model, the fire risk description model and the classification prediction model, generating multiple fire risk prediction networks; Based on the prediction performance indicators of the plurality of fire risk prediction networks, a plurality of feature space mapping units are selected from reference feature space mapping units corresponding to the respective feature spaces.

3. The artificial intelligence-based forest fire monitoring and early warning method according to claim 2 is characterized in that: The method further comprises: When the template forest status collection data includes the forest status collection data of the environment status feature space, but does not include the forest status collection data of other feature spaces, the forest status collection data of other feature spaces are generated based on the forest status collection data of the environment status feature space, and the forest status collection data of multiple feature spaces included in the template forest status collection data are generated, wherein the forest status collection data of other feature spaces include at least one of the forest status collection data of the soil feature space, the forest status collection data of the meteorological feature space and the forest status collection data of the biological feature space.

4. The artificial intelligence-based forest fire monitoring and early warning method according to claim 1 is characterized in that: The method further comprises: For each set fire risk category in the set fire risk category sequence, according to the correspondence between the fire risk category and the category knowledge data defined previously, based on the set fire risk category, the category knowledge data corresponding to the set fire risk category is retrieved.

5. The artificial intelligence-based forest fire monitoring and early warning method according to claim 1 is characterized in that: The method further comprises: For each set fire risk category in the set fire risk category sequence, generating guidance data for search query based on the set fire risk category; A neural network model is called based on the guidance data, and the neural network model is used to generate category knowledge data corresponding to the set fire risk category.

6. The artificial intelligence-based forest fire monitoring and early warning method according to claim 1 is characterized in that: The method of performing parameter learning on the classification prediction model based on the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each of the category description vector sequences to generate a classification prediction model that meets the model convergence requirements includes: The forest state vector sequence is used as a reference feature, and each of the category description vector sequences is used as a retrieval feature, and loaded into a classification prediction model based on pattern recognition in the fire risk prediction network, and the classification prediction model is used to perform a correlation analysis on the reference feature and each of the retrieval features to generate a fire risk classification prediction vector corresponding to each set fire risk category; The fire risk classification prediction vector corresponding to each set fire risk category is input into an independent fire risk category prediction task or a combined fire risk category prediction task, and the model weight information of the classification prediction model is optimized based on the prediction result to generate a classification prediction model that meets the model convergence requirements.

7. The artificial intelligence-based forest fire monitoring and early warning method according to claim 1 is characterized in that: The method of performing parameter learning on the classification prediction model based on the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each of the category description vector sequences to generate a classification prediction model that meets the model convergence requirements includes: Acquiring a fire risk category carried by the template forest state collection data, wherein the fire risk category carried by the template forest state collection data belongs to the set fire risk category sequence; Based on the fire risk category carried by the template forest state collection data, determining the fire risk category supervision data corresponding to each set fire risk category of the template forest state collection data; Calculating the category confidence of each set fire risk category corresponding to the template forest state collection data according to the fire risk classification prediction vector generated by classifying and predicting the forest state vector sequence and each category description vector sequence; Based on the fire risk category supervision data and the category confidence, a training error parameter is constructed, and the model weight information of the classification prediction model is optimized with minimizing the training error parameter as the optimization direction to generate a classification prediction model that meets the model convergence requirements.

8. The artificial intelligence-based forest fire monitoring and early warning method according to claim 1 is characterized in that: The method comprises: Obtain target forest status data and complete the trained fire risk prediction network; According to the forest state vector mapping model in the fire risk prediction network, the target forest state collection data is represented by a forest state vector to generate a forest state vector sequence; According to the fire risk description model in the fire risk prediction network, the category knowledge data corresponding to each set fire risk category in the set fire risk category sequence are respectively described for fire risk, and a category description vector sequence of each set fire risk category is generated; According to the classification prediction model in the fire risk prediction network, the forest state vector sequence and the category description vector sequence of each set fire risk category are classified and predicted to generate a fire risk classification prediction vector, and the fire risk category of the target forest state collection data is determined based on the fire risk classification prediction vector.

9. The artificial intelligence-based forest fire monitoring and early warning method according to claim 8 is characterized in that: The method further comprises: When an extended fire risk category appears in the set fire risk category sequence, obtaining category knowledge data corresponding to the extended fire risk category, the extended fire risk category being a fire risk category that has not appeared in the parameter learning process of the fire risk prediction network; Performing fire risk description on the category knowledge data corresponding to the extended fire risk category to generate a category description vector sequence corresponding to the extended fire risk category; The method comprises: performing classification prediction on the forest state vector sequence and the category description vector sequence of each set fire risk category according to the classification prediction model in the fire risk prediction network, generating a fire risk classification prediction vector, and determining the fire risk category of the target forest state collection data according to the fire risk classification prediction vector, including: The classification prediction model in the fire risk prediction network is used to classify and predict the forest state vector sequence, the category description vector sequence of each set fire risk category, and the category description vector sequence of the extended fire risk category to generate a fire risk classification prediction vector, and the fire risk category of the target forest state collection data is determined based on the fire risk classification prediction vector.

10. A forest fire monitoring and early warning system based on artificial intelligence, characterized in that: The artificial intelligence-based forest fire monitoring and early warning system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the artificial intelligence-based forest fire monitoring and early warning method described in any one of claims 1 to 9 above.

Images