Regional risk determination method and device, computer program product and electronic equipment

Abstract

The invention discloses a regional risk determination method and device, a computer program product and electronic equipment. The method relates to the field of public security or other related fields, and comprises the following steps: obtaining a risk prediction demand of a target area, and determining a target data type according to the risk prediction demand, the target data type being used for representing a data type required for predicting the risk of the target area; initial data conforming to the target data type are obtained from different data sources, M initial data are obtained, data fusion is conducted on the M initial data, N target fusion data are obtained, M is larger than N, and M and N are positive integers; and inputting the N target fusion data into the risk prediction model, outputting a risk prediction result, and determining an early warning strategy based on the risk prediction result after the risk prediction result is obtained. According to the method and the device, the problem of insufficient accuracy and timeliness of identification and early warning of regional risks in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of public security or other related fields, and in particular, to a method, device, computer program product and electronic device for determining regional risks. Background Art

[0002] With the acceleration of global urbanization, the continuous expansion of urban scale and the rapid gathering of population have significantly increased the complexity of urban operation systems. In order to ensure the normal operation of cities and the safety of residents, urban risk perception and early warning have become the key to ensuring the normal operation of cities and the safety of residents. Urban risks include natural disasters and public safety incidents. These risks not only affect the stability of urban infrastructure, but also directly affect the safety of life and property of residents. At present, urban risk perception systems mainly rely on the analysis of single data sources, such as meteorological data, video surveillance, social media data, etc. Although these technologies can effectively identify and warn some risks in specific scenarios, their limitations cannot be ignored.

[0003] First, analysis of a single data source cannot capture the full picture of urban risks and lacks a comprehensive reflection of the complexity and diversity of risks, which can easily lead to incomplete risk identification and increase the risk of false alarms and omissions. Since information between different data sources is often isolated, it can easily lead to fragmented risk perception and make it difficult to form a comprehensive urban risk cognition framework, thus affecting the accuracy and timeliness of risk warnings.

[0004] In addition, with the increase in urban size and the diversification of risk types, the demand for risk perception systems is increasing. However, related technologies have obvious deficiencies in data processing and model selection, and it is difficult to meet residents' urgent needs for risk perception, especially when dealing with large-scale, sudden risk events.

[0005] Currently, no effective solution has been proposed to address the problem of insufficient accuracy and timeliness in identifying and warning of regional risks in relevant technologies. Summary of the invention

[0006] The main purpose of this application is to provide a method, device, computer program product and electronic device for determining regional risks to solve the problem of insufficient accuracy and timeliness in identifying and warning of regional risks in related technologies.

[0007] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for determining regional risk is provided. The method comprises: obtaining the risk prediction requirements of the target region, determining the target data type according to the risk prediction requirements, wherein the target data type is used to characterize the data type required when predicting the risk of the target region; obtaining initial data that conforms to the target data type from different data sources to obtain M initial data, and fusing the M initial data to obtain N target fused data, wherein M is greater than N, and M and N are positive integers; inputting the N target fused data into the risk prediction model, and outputting the risk prediction result, wherein after obtaining the risk prediction result, determining the early warning strategy based on the risk prediction result.

[0008] Furthermore, initial data that conforms to the target data type is obtained from different data sources to obtain M initial data, including: determining multiple data sources according to the target data type, wherein each data source refers to a device that collects M initial data in the target area; determining screening conditions according to risk prediction requirements, extracting K to-be-processed data from the data source according to the screening conditions, and pre-processing the K to-be-processed data to obtain M initial data, wherein K is greater than or equal to M and K is a positive integer.

[0009] Furthermore, the M initial data are subjected to data fusion to obtain N target fused data, including: performing modal classification on the M initial data according to the scene features of each initial data to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; performing vector conversion on the initial features associated with the initial data in each group of initial modal data in the N groups of initial modal data to obtain N groups of feature vectors; determining the learning weights of the initial features of each initial data in the M initial data to obtain M learning weights; for a group of initial modal data, performing weighted calculation on a group of feature vectors associated with a group of initial modal data and the learning weights of the initial features associated with a group of initial modal data to obtain a target fused data.

[0010] Furthermore, the M initial data are fused to obtain N target fused data, including: performing modal classification on the M initial data according to the scene characteristics of each initial data to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; obtaining Y scene categories, determining the prediction probability that each group of initial modal data belongs to the Y scene categories, and obtaining N groups of prediction probability sets, wherein each group of prediction probability sets includes Y prediction probabilities, and Y is a positive integer; determining the prediction weight of each group of prediction probability sets to obtain N prediction weights, and performing weighted calculation on each prediction weight and each group of prediction probability sets to obtain N target fused data.

[0011] Furthermore, the M initial data are fused to obtain N target fused data, including: performing vector conversion on the initial features of the M initial data to obtain M feature vectors; for one feature vector, using a probability distribution function to calculate the normalization factors of the feature vector and M-1 feature vectors respectively to obtain M-1 normalization factors, and using the M-1 normalization factors to determine the offset data of the feature vector; for one feature vector, determining the coincidence data of the feature vector and M-1 feature vectors respectively to obtain M-1 coincidence data; constructing a feature matrix according to the M coincidence data and the M offset data, and determining N target fused data based on the feature matrix.

[0012] Furthermore, the early warning strategy is determined in the following way: extracting risk features from risk prediction results, wherein the risk features are used to characterize the degree of association of each target fusion data with P risk categories, where P is a positive integer; determining the levels of the P risk categories based on the risk features, obtaining P risk levels, and generating an early warning strategy based on the P risk levels.

[0013] Furthermore, the risk prediction model is trained in the following manner: obtaining Q historical data of a historical time period and obtaining an initial risk prediction model, where Q is a positive integer; using the Q historical data to perform unsupervised training on the initial risk prediction model to obtain a first prediction model; obtaining P risk categories, using the P risk categories to label the Q historical data to obtain Q historical category data, and using the Q historical feature data to perform supervised training on the first prediction model to obtain a second prediction model; performing knowledge distillation processing on the second prediction model to obtain a risk prediction model.

[0014] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a device for determining regional risk is provided. The device includes: a first acquisition unit, which is used to obtain the risk prediction requirements of the target region, and determine the target data type according to the risk prediction requirements, wherein the target data type is used to characterize the data type required when predicting the risk of the target region; a second acquisition unit, which is used to obtain initial data that conforms to the target data type from different data sources, obtain M initial data, and perform data fusion on the M initial data to obtain N target fused data, wherein M is greater than N, and M and N are positive integers; an input unit, which is used to input the N target fused data into a risk prediction model and output a risk prediction result, wherein after the risk prediction result is obtained, an early warning strategy is determined based on the risk prediction result.

[0015] According to another aspect of an embodiment of the present invention, a computer storage medium is provided. The computer storage medium is used to store a program. When the program is executed, the device where the computer storage medium is located is controlled to execute a method for determining regional risks.

[0016] According to another aspect of an embodiment of the present invention, there is also provided an electronic device comprising one or more processors and a memory; the memory stores computer-readable instructions, and the processor is used to run the computer-readable instructions, wherein the computer-readable instructions execute a method for determining regional risks when they are run.

[0017] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, a method for determining regional risk is performed.

[0018] Through the present application, the following steps are adopted: obtaining the risk prediction needs of the target area, determining the target data type according to the risk prediction needs, wherein the target data type is used to characterize the data type required for predicting the risk of the target area; obtaining initial data that meets the target data type from different data sources to obtain M initial data, and fusing the M initial data to obtain N target fused data, wherein M is greater than N, and M and N are positive integers; inputting the N target fused data into a risk prediction model, and outputting a risk prediction result, wherein after obtaining the risk prediction result, determining an early warning strategy based on the risk prediction result, solving the problem of insufficient accuracy and timeliness in identifying and warning of regional risks in related technologies, identifying the risk prediction needs of the target area, obtaining the target data type, obtaining the initial data according to the data type, fusing the initial data to obtain the target fused data, and finally using the risk prediction model to process the target fused data to obtain the risk prediction result, thereby achieving the effect of improving the accuracy and efficiency of identifying and warning of regional risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 is a flow chart of a method for determining regional risk according to an embodiment of the present application;

[0021] Figure 2 is a schematic diagram of a risk prediction model training method provided according to an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of an urban risk perception platform provided according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of an urban risk perception system provided according to an embodiment of the present application;

[0024] Figure 5is a schematic diagram of a device for determining regional risk according to an embodiment of the present application;

[0025] Figure 6 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0027] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set between this system and the relevant user or organization. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0030] It should be noted that the collected information used in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse to use it.

[0031] The present invention is described below in conjunction with preferred implementation steps. Figure 1 is a flow chart of a method for determining regional risk according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0032] Step S101, obtaining risk prediction requirements for a target area, and determining a target data type according to the risk prediction requirements, wherein the target data type is used to characterize the data type required for predicting the risk of the target area.

[0033] Specifically, the risk prediction needs may be regional high temperature prediction needs, traffic status prediction needs, and heavy rain prediction needs, etc. The target area may refer to an area where risk prediction is required (such as a city). In order to improve the accuracy and timeliness of risk prediction, after obtaining the risk prediction needs of the target area, the risk prediction needs are first identified, and then the environment and urban infrastructure of the target area are determined based on the needs. For example, when the target area is a coastal city, the risk prediction needs of the area may be prediction needs for risks such as heavy rain, sea level rise, and drainage system efficiency; when the target area is an area where earthquakes occur frequently, the risk prediction needs may be prediction needs for geological changes.

[0034] Furthermore, after extracting relevant information from the above-mentioned risk prediction needs, the target data type required for predicting regional risks can be determined based on the extracted information. For example, after identifying the risk prediction needs of coastal cities, the determined target data type may be meteorological data (such as real-time temperature and humidity, wind speed and direction, air pressure, precipitation and other meteorological parameters, as well as long-term meteorological trends and seasonal change data); after identifying the risk prediction needs generated by cities with frequent earthquakes, the determined target data type may be geographic spatial data (such as topography, river distribution, building density, population distribution and other information).

[0035] Step S102, obtaining initial data that conforms to the target data type from different data sources to obtain M initial data, and performing data fusion on the M initial data to obtain N target fused data, wherein M is greater than N, and M and N are positive integers.

[0036] Specifically, the data source can be sensor devices such as earthquake monitoring sensors, water level sensors, smoke sensors, air quality monitors, or meteorological stations or satellite remote sensing systems. After determining the target data type, initial data can be extracted from multiple data sources based on these data types. For example, when the target data type is a meteorological data type, real-time and historical meteorological data such as temperature, humidity, precipitation, wind speed and direction, etc. can be obtained from a meteorological station or a satellite remote sensing system; when the target data type is a video data type, multiple video data can be obtained through camera devices in the city, and then the video analysis algorithm can be used to identify the objects, behaviors, and environmental changes therein, thereby determining public safety events such as traffic congestion.

[0037] In addition, the data source can also be a database stored in historical time periods, which may include historical disaster records, emergency response cases, urban operation data, etc. Through these data, the risk prediction model can be trained to learn and master risk patterns and development trends.

[0038] It should be noted that after obtaining the initial data stored in the data source, these data can also be subjected to multimodal fusion processing to obtain a more comprehensive, accurate and valuable information set. Among them, when fusing the data, feature-level fusion, decision-level fusion or hybrid fusion can be used for processing, which can not only improve the comprehensive utilization rate of the data, but also provide a more comprehensive and in-depth information basis for subsequent risk identification and assessment.

[0039] Step S103, inputting N target fusion data into the risk prediction model, and outputting the risk prediction result, wherein after obtaining the risk prediction result, determining the early warning strategy based on the risk prediction result.

[0040] Specifically, the risk prediction model can be a deep learning model that can process, fuse and understand different types of data (such as text, images, audio, video, etc.), and learn their common semantics from various forms of information to achieve interaction between different modalities. The risk prediction model can be a convolutional neural network model, a long short-term memory network model, etc.

[0041] Since the processed target fusion data contains a multi-angle description of urban risks, such as meteorological parameters, image features and other information, these data can be used as input to the risk prediction model to ensure that the model can fully understand the complexity and diversity of potential risks. After the risk prediction model obtains the above target fusion data, it can analyze the target fusion data to predict possible future risk events and their impact, and then output the risk prediction results, where the risk prediction results, as the probability distribution of one or more risk events, can include information such as event type, probability of occurrence, and expected impact range.

[0042] Furthermore, the above risk prediction results can be analyzed to evaluate the urgency and severity of risk events, and then determine different early warning strategies. For example, when the risk prediction results indicate that there is a high probability of flooding in a certain area within the next 24 hours, different early warning strategies can be formulated according to the possible flooding depth and impact range, and the early warning information can be quickly released to residents in the affected areas through multiple channels such as SMS platforms, social media, radio and television.

[0043] The method for determining regional risks provided in the embodiment of the present application obtains the risk prediction requirements of the target region, determines the target data type according to the risk prediction requirements, wherein the target data type is used to characterize the data type required for predicting the risk of the target region; obtains initial data that conforms to the target data type from different data sources to obtain M initial data, and fuses the M initial data to obtain N target fused data, wherein M is greater than N, and M and N are positive integers; inputs the N target fused data into a risk prediction model, and outputs a risk prediction result, wherein after obtaining the risk prediction result, determines an early warning strategy based on the risk prediction result, solves the problem of insufficient accuracy and timeliness in identifying and warning of regional risks in related technologies, obtains the target data type by identifying the risk prediction requirements of the target region, obtains the initial data according to the data type, fuses the initial data to obtain the target fused data, and finally uses the risk prediction model to process the target fused data to obtain the risk prediction result, thereby achieving the effect of improving the accuracy and efficiency of identifying and warning of regional risks.

[0044] In order to improve the accuracy and efficiency of risk identification and early warning, optionally, in the method for determining regional risks provided in an embodiment of the present application, initial data that conforms to the target data type is obtained from different data sources to obtain M initial data, including: determining multiple data sources according to the target data type, wherein each data source refers to a device that collects M initial data in the target area; determining filtering conditions according to risk prediction requirements, extracting K data to be processed from the data source according to the filtering conditions, and pre-processing the K data to be processed to obtain M initial data, wherein K is greater than or equal to M and K is a positive integer.

[0045] Specifically, when using the target data type to collect initial data, it is first necessary to determine the corresponding data source based on the determined target data type. For example, when the target data type is geographic environment data, the data source can be environmental data sensors distributed in different locations in the region, such as earthquake monitoring sensors, water level sensors, smoke sensors, air quality monitors and other Internet of Things devices used to warn of structural safety, environmental changes, etc.; when the target data type is a meteorological data type, the data source can be a meteorological data source, such as a weather station, satellite remote sensing system, etc. that can collect meteorological parameters such as temperature, humidity, precipitation, wind speed, and atmospheric pressure; when the target data type is a video data type, the data source can be a surveillance camera network in the city for analyzing objects, behaviors, and environmental changes; the data source can also be a historical data source associated with a historical database, which can store meteorological records, disaster cases, emergency response records, etc. for historical time periods, and can realize model training and risk pattern recognition.

[0046] Furthermore, in order to improve the pertinence and efficiency of data processing, after determining the data source, the filtering conditions can be determined according to specific risk prediction requirements, such as risk type, time range, geographical area, etc., and then the initial data related to risk prediction can be accurately extracted through the filtering conditions. For example, when predicting typhoon risks, the filtering conditions can be to filter meteorological data within a specific time range; when monitoring public safety events, the filtering conditions can be to filter the content of specific geographical locations and time information. After extracting multiple data to be processed from the data source using the filtering conditions, these data can also be pre-processed such as data cleaning to obtain the initial data. This embodiment can collect information directly related to risk prediction by extracting initial data from the data source using the filtering conditions, covering multiple dimensions of risk events, and thus can significantly improve the depth and breadth of risk prediction, reduce false positives and false negatives that may be caused by a single data source, and improve the accuracy of model prediction.

[0047] Optionally, in the method for determining regional risks provided in an embodiment of the present application, M initial data are fused to obtain N target fused data, including: modally classifying the M initial data according to the scene features of each initial data to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; performing vector conversion on the initial features associated with the initial data in each group of initial modal data in the N groups of initial modal data to obtain N groups of feature vectors; determining the learning weights of the initial features of each initial data in the M initial data to obtain M learning weights; for a group of initial modal data, performing weighted calculation on a group of feature vectors associated with a group of initial modal data and the learning weights of the initial features associated with a group of initial modal data to obtain a target fused data.

[0048] Specifically, after the initial data is determined using the screening conditions, in order to help the risk prediction model output a more comprehensive risk prediction result, the initial data can be fused and processed. First, the scene features of each initial data can be identified. For example, the scene features corresponding to meteorological data are weather scenes, the scene features corresponding to video surveillance data are public safety event scenes, and the scene features corresponding to social media data are social change scenes, etc. Then, the initial data can be classified into scenes based on the scene features, so as to obtain multiple groups of initial modal data corresponding to different scene types, such as initial modal data for natural disaster scenes, initial modal data for public safety scenes, etc.

[0049] Furthermore, after obtaining the above initial modal data, feature extraction can be performed on each initial data in each group of initial modal data, and the extracted initial features can be converted into feature vectors, and the learning weight of each feature vector can be determined. For example, in an earthquake early warning scenario, since the geographic feature vectors of earthquake monitoring sensors can directly reflect the physical characteristics of earthquake activity, these vectors can be associated with higher learning weights. Finally, all feature vectors and corresponding learning weights are weighted, that is, the target fusion data corresponding to each group of initial modal data is calculated by the following formula: Among them, α i Represents the feature vector The learning weight of i represents the i-th eigenvector. This embodiment determines the learning weight of the eigenvector of each initial data, and then determines the target fusion data by means of feature-level fusion, so as to effectively combine them according to the scene characteristics, effectively avoiding the one-sidedness and limitations of information when analyzing a single data source, and helping the model to more accurately capture the features related to the risk, thereby improving the accuracy and timeliness of the early warning.

[0050] There are many data fusion methods. Optionally, in the method for determining regional risks provided in an embodiment of the present application, M initial data are fused to obtain N target fused data, including: modally classifying the M initial data according to the scene characteristics of each initial data to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; obtaining Y scene categories, determining the prediction probability that each group of initial modal data belongs to the Y scene categories, and obtaining N groups of prediction probability sets, wherein each group of prediction probability sets includes Y prediction probabilities, and Y is a positive integer; determining the prediction weight of each group of prediction probability sets to obtain N prediction weights, and performing weighted calculation on each prediction weight and each group of prediction probability sets to obtain N target fused data.

[0051] Specifically, in order to improve the prediction accuracy of the risk prediction model, the decision-level fusion method can also be used to realize data processing. First, after classifying based on the scene features of each initial data and obtaining multiple groups of initial modal data, the probability of each group of initial modal data belonging to each scene category can be calculated according to the scene type of the initial modal data, and then a corresponding set of prediction probability sets can be obtained, that is, the scene type and each scene category are similarly calculated to determine the prediction probability in this way. After further determining the prediction weight based on the scene category, the target fusion data can be calculated using the prediction weight and the prediction probability set. At this time, the target fusion data P can be calculated by the following formula fused (y): Among them, P i (y) represents the predicted probability that the i-th group of initial modal data belongs to the y-th scene category, β i Represents the prediction weight of the prediction probability set corresponding to each set of initial modal data. This embodiment can cover a wide range of risk scenario types by calculating the target fusion data using decision-level fusion. At the same time, it provides quantitative risk assessment through the calculation of prediction probability, significantly improving the comprehensiveness and accuracy of risk identification.

[0052] In addition to using feature-level fusion and decision-level fusion methods for data processing, data fusion can also be achieved based on a multi-fusion method. Optionally, in the method for determining regional risk provided in an embodiment of the present application, M initial data are fused to obtain N target fusion data, including: performing vector conversion on the initial features of the M initial data to obtain M feature vectors; for one feature vector, using a probability distribution function to calculate the normalization factors of the feature vector and M-1 feature vectors respectively, to obtain M-1 normalization factors, and using the M-1 normalization factors to determine the offset data of the feature vector; for one feature vector, determining the overlap data of the feature vector and M-1 feature vectors respectively, to obtain M-1 overlap data; constructing a feature matrix based on the M overlap data and the M offset data, and determining N target fusion data based on the feature matrix.

[0053] Specifically, when using the multi-fusion method for data processing, first, feature extraction can be performed on all the acquired initial data to obtain the corresponding initial features, and then these initial features can be vectorized to obtain the corresponding feature vectors. For example, the initial features can be visual features such as objects and behaviors extracted from image data, or semantic features such as keywords and topics extracted from text data, or statistical features such as numerical changes and trends obtained from sensor data.

[0054] Furthermore, the probability distribution function (also known as mass function) in the DS (Dempster-Shafer) theory can be used to calculate the normalization factor for each eigenvector, that is, the normalization factor of the i-th eigenvector and the j-th eigenvector can be calculated using the above mass function, where the DS theory can handle uncertainty and incomplete information, and the mass function can represent the degree of trust in a specific proposition.

[0055] Then, the degree of conflict or inconsistency between the information represented by each two eigenvectors is calculated based on the above normalization factor, that is, the offset data of each eigenvector is calculated by the following formula: Conf(i,j)=-log2(1-C), where C represents the normalization factor, and i and j represent the i-th and j-th eigenvectors respectively.

[0056] Then, the offset data of each two eigenvectors is used to calculate the offset data of a single eigenvector and all other eigenvectors: Among them, t represents the total number of data standard types, and Conf(i,k) represents the offset data of the i-th eigenvector and the k-th eigenvector. Then, based on the probability distribution function, a coincidence function representing the degree of similarity or consistency is constructed, that is, the coincidence data of the i-th eigenvector relative to all other eigenvectors is determined according to the following formula: Among them, τ represents the boundary value for determining whether the feature vectors overlap, d ik Represents the distance between two eigenvectors determined by the mass function.

[0057] Furthermore, normalization is performed based on the coincidence data and offset data of each eigenvector to obtain the importance function: Among them, ε0 represents the normalized threshold. According to the above formula, the feature matrix {a mn}, where the element a in the matrix mn Indicates the importance of data standard m relative to data standard n:

[0058] Finally, the above matrix is ​​used to determine the target fusion data S m : This embodiment calculates the target fusion data through a multi-fusion method, which can quantitatively evaluate the correlation and complementarity between different data sources, effectively improve the fusion effect of multimodal data, enhance the ability to understand complex risk scenarios, avoid false positives and omissions that may be caused by single data source analysis, and significantly improve the accuracy and reliability of risk identification.

[0059] Optionally, in the method for determining regional risks provided in an embodiment of the present application, the early warning strategy is determined in the following manner: extracting risk features from risk prediction results, wherein the risk features are used to characterize the degree of association of each target fusion data with P risk categories, where P is a positive integer; determining the levels of P risk categories based on the risk features, obtaining P risk levels, and generating an early warning strategy based on the P risk levels.

[0060] After the risk prediction model outputs the risk prediction results, the corresponding early warning strategy can be determined based on the risk prediction results. Specifically, first, multiple preset risk categories, such as natural disasters, etc., can be obtained, and then risk characteristics related to the above risk categories can be extracted from the risk prediction results. For example, risk characteristics may include the frequency, intensity, impact range, time pattern, etc. of risk events that can characterize the degree of association with different risk categories.

[0061] Further, the risk level can be determined according to the risk characteristics, and a corresponding early warning strategy can be generated based on the risk category and risk level, wherein the early warning strategy can include the content of the early warning information, the early warning method, the early warning object, the early warning time, etc., to ensure that the early warning information is received by the right people at the right time. This embodiment uses the risk prediction results to determine the early warning strategy, which helps to optimize resource allocation and emergency response strategies, improve the decision-making support capabilities of the city when facing various risks, and ensure that when risk events occur, they can respond quickly and deal with them effectively.

[0062] Optionally, in the method for determining regional risks provided in an embodiment of the present application, the risk prediction model is trained in the following manner: obtaining Q historical data of a historical time period and obtaining an initial risk prediction model, where Q is a positive integer; using the Q historical data to perform unsupervised training on the initial risk prediction model to obtain a first prediction model; obtaining P risk categories, using the P risk categories to label the Q historical data to obtain Q historical category data, and using the Q historical feature data to perform supervised training on the first prediction model to obtain a second prediction model; performing knowledge distillation processing on the second prediction model to obtain a risk prediction model.

[0063] Specifically, in order to improve the accuracy of risk prediction, it is necessary to optimize the risk prediction model. Figure 2 is a schematic diagram of a risk prediction model training method provided in an embodiment of the present application, such as Figure 2 As shown, first, multiple historical data in the past historical time period can be collected and integrated, where the historical data can be historical meteorological records, public safety event records, etc. Then, these historical data are subjected to data fusion processing such as large model processing, feature level fusion, decision level fusion or hybrid fusion strategy to obtain multiple historical fusion data.

[0064] Furthermore, after obtaining the initial risk prediction model, the above historical fusion data can be used to perform unsupervised training on the initial risk prediction model to obtain a usable pre-trained model (i.e., the first prediction model) with preliminary risk identification and prediction capabilities. This model can reduce dependence on large amounts of data and lay a solid foundation for subsequent supervised training of the model. After obtaining the risk category, the above historical fusion data can be labeled based on the risk category to obtain historical category data that characterizes different risk scenarios. The historical category data is then used to perform supervised training on the first prediction model to obtain an application adaptation model (i.e., the second prediction model). It should be noted that during this supervised training stage, the model can learn specific features that are closely related to the risk category and improve the accuracy of risk prediction.

[0065] Furthermore, in order to improve the prediction accuracy, the second prediction model can be further optimized through project polishing to obtain a real-time risk prediction model, and finally the real-time risk prediction model is subjected to knowledge distillation, that is, by transferring the decision knowledge of the large model to the small model, and finally obtaining a risk prediction model with better computational efficiency and generalization ability (that is, a high-precision risk prediction small model).

[0066] It should be noted that when optimizing the above model, the model parameters can be adjusted through loss function, learning rate, regularization, etc. to ensure that the model can fully capture the complex relationship between data. When it comes to the learning rate, first you need to determine the learning rate that can control the distance that the agent moves in the negative direction of the gradient at each step. The learning rate can be calculated using the basic gradient descent formula: Among them, θ t is the current parameter, θ t+1 is the updated parameter, J(θ t ) is the loss function value under the current parameters, is the gradient of the loss function with respect to the parameters, and η represents the learning rate.

[0067] Since the model needs to minimize the objective function value L(γ) to complete the optimization, the minimized objective function value L(γ) can be expressed as: minL(γ),γ={γ1,γ2,…,γ n}, and according to Lagrange's theorem and projection method, the update strategy is: γ l+1 ←γ l +ρ t D t , where ρ t is the step size determined by linear search, D t is the vector of the gradient descent direction.

[0068] In addition, the regularization method can reduce model overfitting by adding a penalty term to the loss function. The principle of L1 regularization is as follows: Among them, J(θ) is the original loss function, λ is the regularization coefficient used to control the strength of the regularization term, and n is the number of parameters. L1 regularization helps to produce a sparse weight matrix, that is, many weights become zero; the loss function can be expressed as: This embodiment optimizes the training of the risk prediction model through unsupervised pre-training and supervised fine-tuning, which can help the model learn the broad characteristics and specific patterns of risk events, significantly improve the accuracy and stability of risk prediction, and at the same time significantly reduce the computing resource requirements of the model, thereby improving the efficiency of real-time data processing and risk warning.

[0069] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0070] The present application also provides a system for determining regional risk. Figure 3 is a schematic diagram of an urban risk perception platform provided according to an embodiment of the present application, Figure 4 is a schematic diagram of an urban risk perception system provided according to an embodiment of the present application, such as Figure 3 , Figure 4 As shown, the platform includes: a city risk perception system, a business application module and a base support module, wherein the city risk perception system includes a data acquisition layer, a data processing layer, a model training layer and an application layer; the base support module includes a hardware device sub-module and an algorithm capability sub-module, the hardware device sub-module includes a CPU, a storage server, an algorithm GPU and a cloud server, and the algorithm capability sub-module includes a multimodal large model; the business application module includes a platform application layer, an algorithm application layer and an auxiliary application layer.

[0071] It should be noted that the platform application layer can realize manual review, intelligent grading, algorithm feedback, risk classification, alarm streaming and warning release, that is, it can identify potential public safety events based on fusion data, evaluate their urgency and scope of impact, generate targeted warning information, and release warnings through multiple channels. The algorithm application layer can realize water level rise prediction, rainstorm and flood prediction, air quality prediction, earthquake disaster prediction, fire warning prediction, crowd gathering prediction, smart traffic prediction, knife and weapon prediction and fight prediction. The auxiliary application layer can realize department management, job authority management, process management, video sharing management, data control processing and parameter setting, that is, it provides management functions for different emergency departments, including resource allocation, personnel scheduling, etc., to ensure efficient and orderly emergency response; process management can define the standard process of warning information generation, review and release to ensure the accuracy and reliability of information; job authority management can set different platform access rights according to user roles and responsibilities to ensure data security and privacy; parameter setting can allow users to adjust model parameters and warning strategies to meet the public safety needs of specific areas.

[0072] The CPU and GPU in the hardware submodule can provide computing resources to support the training of deep learning models and real-time data processing; the storage server can store historical data and model training results to ensure data integrity and accessibility; the cloud server can provide elastic computing and storage services to ensure the high availability and scalability of the platform. The multimodal large model in the algorithm capability submodule can achieve efficient integration and analysis of cross-modal data.

[0073] When predicting regional risks, we first need to determine the target data type required for predicting regional risks. At this time, the data acquisition layer can collect multiple initial data that meet the target data type, and use the data processing layer to clean, label, extract features, and fuse the collected data to obtain the target fused data.

[0074] Furthermore, after the risk prediction model is trained using the model training layer in the above system, the target fusion data is input into the trained risk prediction model, and the risk prediction model outputs the risk prediction result, and the application layer evaluates the risk prediction result, that is, the algorithm application layer processes the risk prediction result through risk identification, risk assessment, alarm generation, and algorithm feedback, and then obtains the alarm strategy. Finally, the platform application layer publishes the alarm strategy, that is, the above strategy is published to urban residents through alarm processing, alarm push, and warning release.

[0075] This embodiment uses the urban risk perception platform to predict regional risks, which can provide a more comprehensive and in-depth understanding of the city's operating conditions and potential risks. When a risk event occurs, it can quickly provide accurate risk information, speed up emergency response, reduce losses, and improve the effectiveness of emergency handling.

[0076] The embodiment of the present application also provides a device for determining regional risk. It should be noted that the device for determining regional risk in the embodiment of the present application can be used to execute the method for determining regional risk provided in the embodiment of the present application. The device for determining regional risk provided in the embodiment of the present application is introduced below.

[0077] Figure 5 is a schematic diagram of a device for determining regional risk according to an embodiment of the present application, such as Figure 5 As shown, the device includes: a first acquisition unit 50, a second acquisition unit 51, and an input unit 52.

[0078] A first acquisition unit 50 is used to acquire risk prediction requirements of a target area and determine a target data type according to the risk prediction requirements, wherein the target data type is used to characterize a data type required for predicting the risk of the target area;

[0079] The second acquisition unit 51 is used to acquire initial data that conforms to the target data type from different data sources to obtain M initial data, and perform data fusion on the M initial data to obtain N target fusion data, where M is greater than N, and M and N are positive integers;

[0080] The input unit 52 is used to input the N target fusion data into the risk prediction model and output the risk prediction result. After the risk prediction result is obtained, the early warning strategy is determined based on the risk prediction result.

[0081] The device for determining regional risk provided in the embodiment of the present application obtains the risk prediction requirements of the target region through the first acquisition unit 50, and determines the target data type according to the risk prediction requirements, wherein the target data type is used to characterize the data type required for predicting the risk of the target region; the second acquisition unit 51 obtains initial data that conforms to the target data type from different data sources to obtain M initial data, and fuses the M initial data to obtain N target fused data, wherein M is greater than N, and M and N are positive integers; the input unit 52 inputs the N target fused data into the risk prediction model, and outputs the risk prediction result, wherein after the risk prediction result is obtained, the early warning strategy is determined based on the risk prediction result, thereby solving the problem of insufficient accuracy and timeliness in identifying and warning of regional risks in related technologies, and obtaining the target data type by identifying the risk prediction requirements of the target region, and fusing the initial data after obtaining the initial data according to the data type to obtain the target fused data, and finally using the risk prediction model to process the target fused data to obtain the risk prediction result, thereby achieving the effect of improving the accuracy and efficiency of identifying and warning of regional risks.

[0082] Optionally, in the regional risk determination device provided in the embodiment of the present application, the second acquisition unit 51 includes: a first determination module, used to determine multiple data sources according to the target data type, wherein each data source refers to a device that collects M initial data in the target area; a second determination module, used to determine filtering conditions according to risk prediction requirements, extract K to-be-processed data from the data source according to the filtering conditions, and pre-process the K to-be-processed data to obtain M initial data, wherein K is greater than or equal to M and K is a positive integer.

[0083] Optionally, in the regional risk determination device provided in the embodiment of the present application, the second acquisition unit 51 includes: a first classification module, used to perform modal classification on M initial data according to the scene features of each initial data, to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; a first conversion module, used to perform vector conversion on the initial features associated with the initial data in each group of initial modal data in the N groups of initial modal data, to obtain N groups of feature vectors; a third determination module, used to determine the learning weights of the initial features of each initial data in the M initial data, to obtain M learning weights; a first calculation module, used to perform weighted calculation on a group of feature vectors associated with a group of initial modal data and the learning weights of the initial features associated with a group of initial modal data, for a group of initial modal data, to obtain a target fusion data.

[0084] Optionally, in the regional risk determination device provided in the embodiment of the present application, the second acquisition unit 51 includes: a second classification module, used to perform modal classification on M initial data according to the scene characteristics of each initial data, to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; a first acquisition module, used to obtain Y scene categories, determine the prediction probability that each group of initial modal data belongs to Y scene categories, and obtain N groups of prediction probability sets, wherein each group of prediction probability sets includes Y prediction probabilities, and Y is a positive integer; a fourth determination module, used to determine the prediction weight of each group of prediction probability sets, to obtain N prediction weights, and perform weighted calculation on each prediction weight and each group of prediction probability sets to obtain N target fusion data.

[0085] Optionally, in the regional risk determination device provided in the embodiment of the present application, the second acquisition unit 51 includes: a second conversion module, used to perform vector conversion on the initial features of M initial data to obtain M feature vectors; a second calculation module, used to calculate the normalization factors of the feature vector and M-1 feature vectors respectively using the probability distribution function for one feature vector, to obtain M-1 normalization factors, and use the M-1 normalization factors to determine the offset data of the feature vector; a fifth determination module, used to determine the overlap data of the feature vector and M-1 feature vectors respectively for one feature vector, to obtain M-1 overlap data; a construction module, used to construct a feature matrix based on the M overlap data and the M offset data, and determine N target fusion data based on the feature matrix.

[0086] Optionally, in the regional risk determination device provided in the embodiment of the present application, the input unit 52 includes: an extraction module, used to extract risk features from the risk prediction results, wherein the risk features are used to characterize the degree of association of each target fusion data with P risk categories, where P is a positive integer; a sixth determination module, used to determine the levels of P risk categories based on the risk features, obtain P risk levels, and generate an early warning strategy based on the P risk levels.

[0087] Optionally, in the regional risk determination device provided in the embodiment of the present application, the input unit 52 includes: a second acquisition module, used to obtain Q historical data of a historical time period and obtain an initial risk prediction model, where Q is a positive integer; a training module, used to use the Q historical data to perform unsupervised training on the initial risk prediction model to obtain a first prediction model; a third acquisition module, used to obtain P risk categories, use the P risk categories to label the Q historical data, obtain Q historical category data, and use the Q historical feature data to perform supervised training on the first prediction model to obtain a second prediction model; a processing module, used to perform knowledge distillation processing on the second prediction model to obtain a risk prediction model.

[0088] The above-mentioned regional risk determination device includes a processor and a memory. The above-mentioned first acquisition unit 50, second acquisition unit 51, input unit 52, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0089] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set to solve the problem of insufficient accuracy and timeliness in identifying and warning of regional risks in related technologies by adjusting kernel parameters.

[0090] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0091] An embodiment of the present invention provides a computer storage medium, which is used to store a program. When the program is running, the device where the computer storage medium is located is controlled to execute a method for determining regional risks.

[0092] Figure 6 is a schematic diagram of an electronic device provided according to an embodiment of the present application, such as Figure 6 As shown, an embodiment of the present invention provides an electronic device, the electronic device 60 includes a processor, a memory, and a program stored in the memory and executable on the processor, the processor is used to execute computer-readable instructions, wherein the computer-readable instructions execute a method for determining regional risk when executed. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0093] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of a method for determining regional risk in each embodiment of the present application.

[0094] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0095] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0096] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0097] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0098] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0099] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0100] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0101] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0102] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for determining regional risk, characterized in that: include: Obtaining risk prediction requirements for a target area, and determining a target data type according to the risk prediction requirements, wherein the target data type is used to characterize the data type required for predicting the risk of the target area; Acquire initial data that conforms to the target data type from different data sources to obtain M initial data, and perform data fusion on the M initial data to obtain N target fusion data, where M is greater than N, and M and N are positive integers; The N target fusion data are input into a risk prediction model, and a risk prediction result is output. After the risk prediction result is obtained, an early warning strategy is determined based on the risk prediction result.

2. The method according to claim 1, characterized in that Acquire initial data that conforms to the target data type from different data sources, and obtain M initial data including: Determine multiple data sources according to the target data type, wherein each data source refers to a device in the target area that collects the M initial data; Determine a screening condition according to the risk prediction requirement, extract K unprocessed data from the data source according to the screening condition, and pre-process the K unprocessed data to obtain the M initial data, wherein K is greater than or equal to M and K is a positive integer.

3. The method according to claim 1, characterized in that The M initial data are subjected to data fusion to obtain N target fusion data including: Performing modality classification on the M initial data according to the scene features of each initial data to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; Performing vector conversion on initial features associated with initial data in each group of initial modal data in the N groups of initial modal data, respectively, to obtain N groups of feature vectors; Determine the learning weight of the initial feature of each of the M initial data respectively to obtain M learning weights; For a set of initial modal data, a set of feature vectors associated with the set of initial modal data and learning weights of initial features associated with the set of initial modal data are weightedly calculated to obtain a target fusion data.

4. The method according to claim 1, characterized in that: The M initial data are subjected to data fusion to obtain N target fusion data including: Performing modality classification on the M initial data according to the scene features of each initial data to obtain N groups of initial modal data, wherein each group of initial modal data is associated with a scene type; Obtain Y scene categories, determine the prediction probability that each group of initial modal data belongs to the Y scene categories, and obtain N groups of prediction probability sets, where each group of prediction probability sets includes Y prediction probabilities, and Y is a positive integer; The prediction weight of each set of prediction probability sets is determined to obtain N prediction weights, and each prediction weight is weightedly calculated with each set of prediction probability to obtain the N target fusion data.

5. The method according to claim 1, characterized in that The M initial data are subjected to data fusion to obtain N target fusion data including: Performing vector conversion on the initial features of the M initial data to obtain M feature vectors; For one eigenvector, a probability distribution function is used to calculate normalization factors of the eigenvector and M-1 eigenvectors respectively, to obtain M-1 normalization factors, and the offset data of the eigenvector is determined using the M-1 normalization factors; For one feature vector, determine the coincidence data of the feature vector and M-1 feature vectors respectively, and obtain M-1 coincidence data; A feature matrix is ​​constructed according to the M overlap degree data and the M offset degree data, and the N target fusion data are determined based on the feature matrix.

6. The method according to claim 1, characterized in that The early warning strategy is determined by: Extracting risk features from the risk prediction results, wherein the risk features are used to characterize the degree of association of each target fusion data with P risk categories, where P is a positive integer; The levels of the P risk categories are determined according to the risk characteristics to obtain P risk levels, and the early warning strategy is generated based on the P risk levels.

7. The method according to claim 1, characterized in that The risk prediction model is trained in the following way: Obtain Q historical data of the historical time period and obtain the initial risk prediction model, where Q is a positive integer; Using the Q historical data to perform unsupervised training on the initial risk prediction model to obtain a first prediction model; Obtaining P risk categories, using the P risk categories to perform category labeling on the Q historical data to obtain Q historical category data, and using the Q historical feature data to perform supervised training on the first prediction model to obtain a second prediction model; The second prediction model is subjected to knowledge distillation processing to obtain the risk prediction model.

8. A device for determining regional risk, characterized in that: include: A first acquisition unit is used to acquire risk prediction requirements of a target area, and determine a target data type according to the risk prediction requirements, wherein the target data type is used to characterize a data type required for predicting the risk of the target area; A second acquisition unit is used to acquire initial data that conforms to the target data type from different data sources to obtain M initial data, and perform data fusion on the M initial data to obtain N target fusion data, wherein M is greater than N, and M and N are positive integers; An input unit is used to input the N target fusion data into a risk prediction model and output a risk prediction result, wherein after obtaining the risk prediction result, an early warning strategy is determined based on the risk prediction result.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for determining regional risk according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining regional risks as described in any one of claims 1 to 7.

Images