Seismic operation integrated management method

Abstract

The application discloses a kind of seismic business integrated management method, utilize the powerful nonlinear fitting ability of deep learning, can deeply mine the complex internal relation between historical seismic business correlation data and the final casualty population quantity, through a large number of historical seismic business correlation data and its corresponding casualty population quantity trained casualty population prediction model, its prediction accuracy is much higher than traditional method, provide reliable basis for scientific decision-making, by associating the prediction result with original business data and storing, build the unified data asset with earthquake event as core, not only break information silos, more disaster assessment results are deeply integrated into the entire seismic business process, provide unified, authoritative data entry for emergency command, resource scheduling, information release, post-disaster assessment and the like subsequent work.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a seismic business integration management method. BACKGROUND

[0002] Earthquake is one of the most destructive natural disasters on earth, which can cause huge casualties and property losses in a very short time. Therefore, the rapid and accurate disaster assessment after an earthquake is the premise and key to effective emergency rescue, scientific decision-making and resource scheduling. The data related to the earthquake is scattered in different departments such as the Earthquake Administration, the Ministry of Civil Affairs, the Meteorological Bureau and the Ministry of Transport, and the data formats are different, the standards are different, and it is difficult to realize effective convergence and fusion. Emergency command personnel need to spend a lot of time collecting and sorting information from different channels, which seriously affects the response speed. Traditional casualty assessment mainly relies on historical experience formula, expert judgment or field investigation. The experience formula is often too simplified to consider complex factors such as terrain, building, population distribution, etc.; expert judgment is highly subjective; and field investigation is limited by traffic, communication and other conditions, which takes a long time and cannot provide overall assessment results within the golden rescue time. SUMMARY

[0003] The present application provides a seismic business integration management method, which aims to solve the problem of information island and low timeliness of casualty assessment in the prior art.

[0004] The present application provides a seismic business integration management method, which comprises:

[0005] Collecting historical seismic business correlation data corresponding to historical earthquakes and the number of casualties corresponding to the historical seismic business correlation data;

[0006] Using a deep learning algorithm to learn the data relationship between the historical seismic business correlation data and the number of casualties corresponding thereto, and forming a casualty population prediction model;

[0007] Collecting the latest target seismic business correlation data, identifying the target seismic business correlation data by using the casualty population prediction model, and determining the casualty population prediction result corresponding to the target seismic business correlation data;

[0008] Storing the target seismic business correlation data and the casualty population prediction result in association, and completely integrating the seismic business.

[0009] In a possible implementation, after collecting the historical seismic business correlation data corresponding to the historical earthquakes and the number of casualties corresponding to the historical seismic business correlation data, the method further comprises:

[0010] The random forest algorithm collects historical earthquake business correlation data corresponding to historical earthquakes, determines the historical earthquake business correlation data after screening, and uses a deep learning algorithm to learn the data relationship between the historical earthquake business correlation data and the corresponding casualty population in a subsequent process.

[0011] In a possible implementation, the deep learning algorithm is used to learn the data relationship between the historical earthquake business correlation data and the corresponding casualty population, and a casualty population prediction model is formed, including:

[0012] An initialization model is constructed using a deep learning algorithm;

[0013] The hyperparameters of the initialization model are initialized and encoded, and a plurality of different hyperparameter encodings are obtained;

[0014] Based on the historical earthquake business correlation data and the corresponding casualty population, a loss function value corresponding to each hyperparameter encoding is obtained;

[0015] According to the loss function value corresponding to the hyperparameter encoding, an optimal hyperparameter encoding is determined;

[0016] For any hyperparameter encoding, an adaptive position selection mechanism is used to perform initialization neighborhood search on the hyperparameter encoding, and a hyperparameter encoding after initialization neighborhood search is obtained;

[0017] For any hyperparameter encoding after initialization neighborhood search, according to the optimal hyperparameter encoding, a variable spiral search mechanism is used to perform variable spiral search on the hyperparameter encoding after initialization neighborhood search, and a hyperparameter encoding after variable spiral search is obtained;

[0018] For any hyperparameter encoding after variable spiral search, a fuzzy barycenter variable speed search mechanism is used to perform adaptive search on the hyperparameter encoding after variable spiral search, and a hyperparameter encoding after adaptive search is obtained;

[0019] For any hyperparameter encoding after adaptive search, a focused mutation search mechanism is used to perform global search on the hyperparameter encoding after adaptive search, and a hyperparameter encoding after global search is obtained;

[0020] It is determined whether a training end condition is met, if yes, a target hyperparameter encoding is determined according to the hyperparameter encoding after global search, and otherwise, the step of obtaining the loss function value is returned;

[0021] According to the target hyperparameter encoding and the initialization model, a casualty population prediction model is obtained.

[0022] In a possible implementation, the hyperparameters of the initialization model are initialized to obtain a plurality of different hyperparameter encodings, including:

[0023] For any one of the hyperparameters of the initialization model, the hyperparameters are randomly initialized between the upper limit of the hyperparameters and the lower limit of the hyperparameters, and the hyperparameters after initialization are encoded into a vector to obtain a hyperparameter encoding. The plurality of different hyperparameter encodings are obtained by repeating a plurality of times.

[0024] In a possible implementation, the loss function value corresponding to each hyperparameter encoding is obtained based on the historical earthquake business correlation data and the corresponding casualty population quantity, including:

[0025] For any one of the hyperparameter encodings, the hyperparameter encoding is applied to the initialization model, the historical earthquake business correlation data is taken as input data, the casualty population quantity corresponding to the historical earthquake business correlation data is taken as expected output data, and the root mean square loss function is used to obtain the loss function value corresponding to the hyperparameter encoding.

[0026] All the hyperparameter encodings are traversed to obtain the loss function value corresponding to each hyperparameter encoding.

[0027] In a possible implementation, for any one of the hyperparameter encodings, an adaptive position selection mechanism is used to perform initialization neighborhood search on the hyperparameter encoding to obtain the hyperparameter encoding after the initialization neighborhood search.

[0028]

[0029] wherein, represents the i th hyperparameter encoding in the j th training process, t represents the i th hyperparameter encoding in the j th training process, i represents the i th hyperparameter encoding after the j th initialization neighborhood search, represents the i th hyperparameter encoding after the j th initialization neighborhood search, i =1, 2,..., M, M represents the total number of hyperparameter encodings, i represents a natural constant, represents the maximum number of training, represents a sine function, represents a first random number between 0 and 1, represents a preset decision factor, represents a second random number between 0 and 1.

[0030] ​In one possible implementation, for any hyperparameter encoding after initializing the neighborhood search, a variable spiral search is performed on the hyperparameter encoding after initializing the neighborhood search based on the optimal hyperparameter encoding and using a variable spiral search mechanism, resulting in the hyperparameter encoding after the variable spiral search:

[0031]

[0032]

[0033] in, Indicates the first t During the training process, the first k Hyperparameter encoding after initial neighborhood search Indicates the first k Hyperparameter encoding following a variable spiral search k =1,2,...,M, where M represents the total number of hyperparameter codes. Indicates the first t The optimal hyperparameter encoding during the training process. Represents the natural constant. Represents the spiral shape factor. This represents a random spiral direction factor between [-1, 1]. Represents pi (π). Represents the cosine function. This represents the control coefficient for the spiral shape factor. This indicates the maximum number of training iterations.

[0034] In one possible implementation, for any hyperparameter encoding after a variable spiral search, an adaptive search is performed on the hyperparameter encoding after the variable spiral search using a fuzzy centroid variable speed search mechanism to obtain the hyperparameter encoding after the adaptive search, including:

[0035] The search progress parameter and the coding diversity parameter are obtained as follows:

[0036]

[0037]

[0038] in, This indicates the search progress parameter. Indicates the first m The coding diversity parameter corresponding to the hyperparameter encoding after a variable spiral search. m =1,2,...,M, where M represents the total number of hyperparameter codes. Indicates the first t The loss function value corresponding to the optimal hyperparameter encoding during this training process. Indicates the first t The loss function value corresponding to the optimal hyperparameter encoding during training iteration -1. Indicates the first t During the training process, the first m Hyperparameter encoding following a variable spiral search Indicates the first t During the training process, the first h Hyperparameter encoding following a variable spiral search h =1,2,...,M; express and The Euclidean distance between them;

[0039] Based on the search progress parameters, the first search factor, the second search factor, and the third search factor are determined as follows:

[0040]

[0041]

[0042]

[0043] in, Indicates the first search factor. Indicates the second search factor. Indicates the third search factor;

[0044] Based on the coded diversity parameters, the first diversity impact factor and the second diversity impact factor are obtained as follows:

[0045]

[0046]

[0047] in, This represents the first diversity influencing factor. This represents the second diversity influencing factor. This indicates a preset diversity threshold;

[0048] Based on the first search factor, the second search factor, the third search factor, the first diversity influence factor, and the second diversity influence factor, the fuzzy centroid weights are obtained as follows:

[0049]

[0050] in, Indicates the first m The fuzzy centroid weights corresponding to the hyperparameter encoding after a variable spiral search. This represents the minimum value corresponding to the fuzzy centroid weight. This represents the maximum value corresponding to the weight of the fuzzy centroid;

[0051] Based on the fuzzy centroid weights, an adaptive search is performed on the hyperparameter encoding after the variable spiral search, resulting in the following hyperparameter encoding after the adaptive search:

[0052]

[0053]

[0054] in, Indicates the first m Hyperparameter encoding after an adaptive search Indicates the first t During the training process, the first m The adaptive search quantity corresponding to the hyperparameter encoding after a variable spiral search. Indicates the first t +1 training session m The adaptive search quantity corresponding to the hyperparameter encoding after a variable spiral search. Indicates the first learning factor. Indicates the second learning factor. This represents a third random number between (0,1). This represents the fourth random number between (0,1). express The corresponding historical best value, Indicates the first t The optimal hyperparameter encoding during the training process.

[0055] In one possible implementation, for any hyperparameter encoding after adaptive search, a focused mutation search mechanism is used to perform a global search on the hyperparameter encoding after adaptive search, resulting in a hyperparameter encoding after global search, including:

[0056] The sampling factor is obtained as follows:

[0057]

[0058] in, Indicates the first t During the training process, the first n The hyperparameter encoding after the first adaptive search d dimensional hyperparameters, n =1,2,...,M, where M represents the total number of hyperparameter codes. d =1,2,...,D, where D represents the total number of hyperparameters in the hyperparameter encoding. Indicates the first t The optimal hyperparameter encoding during the training process is the first...d dimensional hyperparameter, denotes a cosine function, denotes a cosine function, denotes the first t dimensional hyperparameter after the first n adaptive search in the first training process, d dimensional hyperparameter corresponding to the sampling factor;

[0059] According to the sampling factor, the hyperparameter encoding after the adaptive search is globally searched to obtain the hyperparameter encoding after the global search as follows:

[0060]

[0061] wherein, denotes the first n dimensional hyperparameter after the first global search, d denotes a global search control factor randomly between (0, 1), denotes a Gaussian distribution random number with a mean of and a variance of

[0062] In a possible implementation, the determination of whether the training end condition is met includes: in a case where the number of training times is greater than or equal to a preset maximum number of training times, it is determined that the training end condition is met, otherwise it is determined that the training end condition is not met.

[0063] Beneficial effects:

[0064] The application provides a seismic business integrated management method, which utilizes the powerful nonlinear fitting capability of deep learning, can deeply mine the complex internal relationship between historical seismic business correlation data and the final number of casualties, and trains a casualty population prediction model through a large amount of historical seismic business correlation data and corresponding casualty population, the prediction accuracy of which is much higher than that of traditional methods, and provides a reliable basis for scientific decision-making. By correlating and storing the prediction results and original business data, a unified data asset is constructed with a seismic event as the core, which not only breaks the information island, but also deeply integrates disaster assessment results into the entire seismic business process, and provides a unified and authoritative data portal for subsequent work such as emergency command, resource scheduling, information release and post-disaster assessment. BRIEF DESCRIPTION OF DRAWINGS

[0065] ​​In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor under the premise of the drawings.

[0066] Figure 1 is a flowchart of a method for integrated management of earthquake services according to an embodiment of the present application. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present application will be clearly and completely described in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0068] As shown in Figure 1 , the present application provides a method for integrated management of earthquake services, comprising:

[0069] S101, collecting historical earthquake service correlation data corresponding to a historical earthquake and a casualty population corresponding to the historical earthquake service correlation data.

[0070] The historical earthquake service correlation data can include:

[0071] Seismic parameter data: such as magnitude (Richter magnitude), focal depth (km), epicenter latitude and longitude, and time of occurrence. These are basic parameters for describing the energy of the earthquake itself.

[0072] Geographical environment data: such as the altitude, slope, and slope direction (topography) of the epicenter area, whether it is located in an active fault zone (geological structure), the type of site soil (rock, hard soil, soft soil, etc.), and these data affect the propagation and amplification of seismic waves.

[0073] Disaster body data: such as the resident population density in administrative divisions or grids, the distribution ratio and quantity of different types of buildings (such as reinforced concrete structures, brick and concrete structures, and civil structures), and the distribution information of lifeline engineering and important facilities such as highways, railways, airports, hospitals, and schools, which are the main body of earthquake disasters.

[0074] Meteorological environment data: such as temperature, wind power, and precipitation forecast within 24 hours after the occurrence of the earthquake. Severe weather can seriously affect the development of rescue work and indirectly affect the casualty situation.

[0075] Date information: such as the date and time of the earthquake occurrence, and holiday information, which are factors that may affect the casualties in an earthquake.

[0076] It is worth noting that various historical earthquake business-related data in the embodiments of the present application should be numerically unified to ensure data normalization.

[0077] S102, using a deep learning algorithm to learn the data relationship between the historical earthquake business-related data and the corresponding number of casualties, and forming a casualty population prediction model.

[0078] Optionally, a CNN (Convolutional Neural Network) algorithm or other existing deep learning algorithms that can realize data relationship identification can be used to obtain the casualty population prediction model. The present application discards the simplifying assumptions of traditional empirical formulas, and uses the powerful nonlinear fitting capability of deep learning to deeply mine the complex internal relationship between magnitude, depth, population density, building structure, topography, and the final number of casualties. The casualty population prediction model trained by a large amount of historical data has a much higher prediction accuracy than traditional methods, providing a reliable basis for scientific decision-making.

[0079] S103, collecting the latest target earthquake business-related data, and calling the casualty population prediction model to identify the target earthquake business-related data, and determining the casualty population prediction result corresponding to the target earthquake business-related data.

[0080] Once an earthquake occurs, the scheme provided by the embodiments of the present application can automatically collect relevant data within a few minutes, and call the pre-trained casualty population prediction model for rapid prediction to generate a casualty population prediction result. Compared with the traditional manual collection and on-site investigation, the disaster assessment time is shortened from several hours or even several days to minutes, which wins valuable time for seizing the golden rescue time.

[0081] S104, storing the target earthquake business-related data and the casualty population prediction result in association, and completely integrating and managing the earthquake business.

[0082] The embodiments of the present application store the prediction results in association with the original service data, and construct a unified data asset with the earthquake event as the core. This not only breaks the information island, but also deeply integrates the disaster assessment results into the entire earthquake service process, provides a unified and authoritative data portal for subsequent work such as emergency command, resource scheduling, information release, and post-disaster assessment, and realizes the leap from data dispersion to business integration. With the continuous accumulation of new earthquake case data, the prediction model of the present application can be retrained and optimized regularly, and its prediction accuracy will continue to improve over time. At the same time, the system architecture design is flexible, and new data sources (such as social media data and unmanned aerial vehicle aerial data) or more advanced algorithm models can be easily accessed, and the system has good technical foresight and scalability.

[0083] In a possible implementation, after collecting the historical earthquake service associated data corresponding to the historical earthquake and the number of casualties corresponding to the historical earthquake service associated data, the method further includes:

[0084] The random forest algorithm is used to perform feature screening on the historical earthquake service associated data corresponding to the historical earthquake, the historical earthquake service associated data after screening is determined, and a deep learning algorithm is used to learn the data relationship between the historical earthquake service associated data and the number of casualties corresponding to the historical earthquake service associated data in a subsequent process.

[0085] The embodiments of the present application can reduce the influence of non-key features by screening the historical earthquake service associated data through the random forest algorithm, thereby improving the prediction accuracy. It is worth noting that after the historical earthquake service associated data after screening is determined, the target earthquake service associated data should have the same data structure as the historical earthquake service associated data after screening.

[0086] In a possible implementation, the deep learning algorithm is used to learn the data relationship between the historical earthquake service associated data and the number of casualties corresponding to the historical earthquake service associated data, and a casualty population prediction model is formed, including:

[0087] An initialization model is constructed using a deep learning algorithm. For example, a CNN algorithm is used to construct an initialization model.

[0088] The hyperparameters of the initialization model are initialized and coded to obtain a plurality of different hyperparameter encodings.

[0089] Based on the historical earthquake service associated data and the number of casualties corresponding to the historical earthquake service associated data, a loss function value corresponding to each hyperparameter encoding is obtained.

[0090] According to the loss function value corresponding to the hyperparameter encoding, the optimal hyperparameter encoding is determined.

[0091] For any one hyperparameter code, an adaptive position selection mechanism is used to initialize the neighborhood search of the hyperparameter code to obtain the hyperparameter code after the initialization neighborhood search.

[0092] For any one hyperparameter code after the initialization neighborhood search, a variable spiral search mechanism is used to perform variable spiral search on the hyperparameter code after the initialization neighborhood search according to the optimal hyperparameter code, to obtain the hyperparameter code after the variable spiral search.

[0093] For any one hyperparameter code after the variable spiral search, a fuzzy barycenter variable speed search mechanism is used to perform adaptive search on the hyperparameter code after the variable spiral search to obtain the hyperparameter code after the adaptive search.

[0094] For any one hyperparameter code after the adaptive search, a focused mutation search mechanism is used to perform global search on the hyperparameter code after the adaptive search to obtain the hyperparameter code after the global search.

[0095] It is judged whether the training end condition is met, if yes, the target hyperparameter code is determined according to the hyperparameter code after the global search, otherwise the step of obtaining the loss function value is returned.

[0096] According to the target hyperparameter code and the initialization model, a casualty population prediction model is obtained, that is, the hyperparameters in the target hyperparameter code are taken as the final hyperparameters of the initialization model, so as to obtain the casualty population prediction model.

[0097] In the prior art, the hyperparameter optimization of the deep learning model depends on artificial experience, grid search or simple heuristic algorithm. These methods often have low search efficiency, are easy to fall into local optimum, and are difficult to adapt to high-dimensional parameter space of complex models. The embodiment of the application innovatively introduces a multi-stage mixed intelligent search mechanism, combines adaptive position selection, variable spiral search, fuzzy barycenter variable speed search and focused mutation search, and constructs a systematic global optimization framework. Through the phased and adaptive search strategy, the framework gradually narrows down the optimal solution range, effectively balances the global exploration and local development capabilities, and greatly improves the scientificity and efficiency of hyperparameter optimization. Through the multi-stage search mechanism, the optimal hyperparameter combination is accurately located, so that the model can more fully learn the complex nonlinear relationship between historical earthquake data and casualty population. Compared with traditional methods, the prediction error is significantly reduced. The adaptive search strategy reduces invalid iterations and shortens the model convergence time, and is especially suitable for high-dimensional and large data volume of earthquake business correlation data scenarios. The global optimization mechanism such as focused mutation search effectively avoids overfitting and enhances the generalization ability, so that the model has more stable prediction performance in unknown earthquake events, and provides a more reliable basis for emergency decision-making.

[0098] Optionally, after each search of the hyperparameter encoding, out-of-bounds processing can be performed on the hyperparameter encoding to ensure its validity.

[0099] In one possible implementation, the hyperparameters of the initialized model are initialized and encoded to obtain multiple different hyperparameter codes, including:

[0100] For any hyperparameter of the initialization model, the hyperparameter is randomly initialized between the upper and lower bounds of the hyperparameter, and the initialized hyperparameter is encoded into a vector to obtain the hyperparameter encoding. This process is repeated multiple times to obtain multiple different hyperparameter encodings.

[0101] Alternatively, a chaotic mapping initialization method can be used to initialize the hyperparameter encoding, thereby making the distribution of the hyperparameter encoding in the solution space more uniform at the initial moment.

[0102] In one possible implementation, based on the historical earthquake operational data and its corresponding number of casualties, the loss function value corresponding to each hyperparameter code is obtained, including:

[0103] For any given hyperparameter encoding, the hyperparameter encoding is applied to the initialization model, using the historical earthquake operational data as input data and the number of casualties corresponding to the historical earthquake operational data as the expected output data. The root mean square loss function is used to obtain the loss function value corresponding to the hyperparameter encoding. All hyperparameter encodings are iterated over to obtain the loss function value corresponding to each hyperparameter encoding.

[0104] Optionally, the cross-entropy loss function can be used to obtain the loss function value corresponding to the hyperparameter encoding.

[0105] In one possible implementation, for any hyperparameter encoding, an adaptive position selection mechanism is used to perform an initial neighborhood search on the hyperparameter encoding, resulting in the hyperparameter encoding after the initial neighborhood search:

[0106]

[0107] in, Indicates the first t During the training process, the first i Hyperparameter encoding, Indicates the first i Hyperparameter encoding after initial neighborhood search i =1,2,...,M, where M represents the total number of hyperparameter codes. Represents the natural constant. Indicates the maximum number of training iterations. Represents the sine function. Represents the first random number between (0,1). This represents the pre-defined decision factors. This represents the second random number between (0,1).

[0108] The embodiments of this application employ an adaptive position selection mechanism to initialize the neighborhood search of the hyperparameter encoding. This allows the hyperparameter encoding to perform adaptive neighborhood search based on its own position in the solution space. Furthermore, as the algorithm progresses, the search range gradually decreases, thereby ensuring that the algorithm can converge.

[0109] In one possible implementation, for any hyperparameter encoding after initializing the neighborhood search, a variable spiral search is performed on the hyperparameter encoding after initializing the neighborhood search based on the optimal hyperparameter encoding and using a variable spiral search mechanism, resulting in the hyperparameter encoding after the variable spiral search:

[0110]

[0111]

[0112] in, Indicates the first t During the training process, the first k Hyperparameter encoding after initial neighborhood search Indicates the first k Hyperparameter encoding following a variable spiral search k =1,2,...,M, where M represents the total number of hyperparameter codes. Indicates the first t The optimal hyperparameter encoding during the training process. Represents the natural constant. Represents the spiral shape factor. This represents a random spiral direction factor between [-1, 1]. Represents pi (π). Represents the cosine function. The control coefficient representing the spiral shape factor can be set to 5; This indicates the maximum number of training iterations.

[0113] The embodiments of this application employ a variable spiral search mechanism to perform a variable spiral search on the hyperparameter encoding after the initial neighborhood search. This allows the hyperparameter encoding to perform a spiral search based on the optimal position found in the solution space, avoiding the problem of linear search easily getting trapped in local optima, and making it easier to find a better position.

[0114] In a possible implementation, for the hyperparameter encoding after any one variable spiral search, a fuzzy barycenter variable speed search mechanism is adopted to perform adaptive search on the hyperparameter encoding after the variable spiral search, to obtain the hyperparameter encoding after adaptive search, including:

[0115] The search progress parameter and the encoding diversity parameter are obtained as:

[0116]

[0117]

[0118] wherein, represents the search progress parameter, represents the encoding diversity parameter, m represents the encoding diversity parameter corresponding to the hyperparameter encoding after the i-th variable spiral search, m M represents the total number of hyperparameter encodings, represents the loss function value corresponding to the optimal hyperparameter encoding in the i-th training process, t represents the loss function value corresponding to the optimal hyperparameter encoding in the i-1-th training process, represents the hyperparameter encoding after the i-th variable spiral search in the i-th training process, t represents the hyperparameter encoding after the i-th variable spiral search in the i-th training process, represents the hyperparameter encoding after the i-th variable spiral search in the i-th training process, t represents the hyperparameter encoding after the i-th variable spiral search in the i-th training process, m represents the hyperparameter encoding after the i-th variable spiral search in the i-th training process, represents the hyperparameter encoding after the i-th variable spiral search in the i-th training process, t represents the hyperparameter encoding after the i-th variable spiral search in the i-th training process, h =1, 2,..., M. h represents the Euclidean distance between and. According to the search progress parameter, the first search factor, the second search factor and the third search factor are determined as:

[0119]

[0120]

[0121]

[0122]

[0123] wherein, represents the first search factor, represents the second search factor, represents the third search factor.

[0124] According to the encoding diversity parameter, the first diversity influence factor and the second diversity influence factor are obtained as:​​​

[0125]

[0126]

[0127] in, This represents the first diversity influencing factor. This represents the second diversity influencing factor. This represents the preset diversity threshold, which can be set to the mean of the encoded diversity parameters.

[0128] Based on the first search factor, the second search factor, the third search factor, the first diversity influence factor, and the second diversity influence factor, the fuzzy centroid weights are obtained as follows:

[0129]

[0130] in, Indicates the first m The fuzzy centroid weights corresponding to the hyperparameter encoding after a variable spiral search. This represents the minimum value corresponding to the weight of the fuzzy centroid, which can be set to 0.4; This represents the maximum value corresponding to the fuzzy centroid weight, which can be set to 0.9.

[0131] Based on the fuzzy centroid weights, an adaptive search is performed on the hyperparameter encoding after the variable spiral search, resulting in the following hyperparameter encoding after the adaptive search:

[0132]

[0133]

[0134] in, Indicates the first m Hyperparameter encoding after an adaptive search Indicates the first t During the training process, the first m The adaptive search quantity corresponding to the hyperparameter encoding after a variable spiral search. Indicates the first t +1 training session m The adaptive search quantity corresponding to the hyperparameter encoding after a variable spiral search. This represents the first learning factor, which can be set to 1.5; This represents the second learning factor, which can be set to 2. This represents a third random number between (0,1). This represents the fourth random number between (0,1). express corresponding historical optimal value, indicates the optimal hyperparameter encoding in the t training process.

[0135] The embodiment of the application adopts a fuzzy barycenter variable search mechanism to perform adaptive search on the hyperparameter encoding after the variable spiral search, considers two indexes of search progress parameters and encoding diversity parameters, so that the hyperparameter adjustment is more targeted and flexible, thereby avoiding search stagnation or premature convergence caused by improper parameter setting. This multi-index feedback can better capture subtle changes in the search process, so as to make different responses in the early, middle and late stages of the algorithm. This adaptability enables the algorithm to more effectively balance global exploration and local development when facing complex, multi-peak or dynamic problems.

[0136] In a possible implementation, for any hyperparameter encoding after adaptive search, a focused mutation search mechanism is adopted to perform global search on the hyperparameter encoding after adaptive search, to obtain the hyperparameter encoding after global search, including:

[0137] The sampling factor is obtained as:

[0138]

[0139] wherein, indicates the optimal hyperparameter encoding in the t training process. n dimensional hyperparameter of the hyperparameter encoding after the d dimensional hyperparameter of the hyperparameter encoding after the n =1, 2,..., M, M represents the total number of hyperparameter encodings, d =1, 2,..., D, D represents the total number of hyperparameters in the hyperparameter encoding, indicates the optimal hyperparameter encoding in the t training process. d dimensional hyperparameter of the optimal hyperparameter encoding in the indicates the ratio of the circumference of a circle to its diameter, indicates the cosine function, dimensional hyperparameter of the hyperparameter encoding after the t dimensional hyperparameter of the hyperparameter encoding after the n dimensional hyperparameter of the hyperparameter encoding after the d dimensional hyperparameter of the hyperparameter encoding after the

[0140] According to the sampling factor, the global search is performed on the hyperparameter encoding after adaptive search, and the hyperparameter encoding after global search is obtained as:

[0141]

[0142] wherein, indicates the optimal hyperparameter encoding in then a global search after the hyperparameter encoding of the first d dimensional hyperparameters, denotes a global search control factor that is random between (0, 1), denotes a Gaussian distribution random number with a mean of and a variance of .

[0143] The embodiment of the present application adopts a focused mutation search mechanism to perform global search on the hyperparameter encoding after adaptive search. Compared with the traditional random walk of uniform distribution, the focused mutation is performed on the optimal hyperparameter encoding, which can have a higher sampling density near the optimal hyperparameter encoding, and a series of mutation sequences around the population optimum value are generated. This mechanism makes the candidate solution generate statistically regular perturbations in the neighborhood of the global optimal solution, which not only maintains the population diversity, but also realizes fine search through the attenuation characteristics.

[0144] Through the mutual cooperation of the above-mentioned multiple mechanisms, the embodiment of the present application effectively avoids the algorithm falling into local optimum, so that the casualty population prediction model obtained by training can more accurately realize prediction.

[0145] In a possible implementation, the determination of whether the training end condition is met includes: in a case where the number of training times is greater than or equal to the preset maximum number of training times, it is determined that the training end condition is met, otherwise it is determined that the training end condition is not met.

[0146] Each of the embodiments in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts between the embodiments can be referred to each other.

[0147] The embodiments of the present application are described with reference to flowcharts and / or block diagrams of the methods, devices, electronic equipment and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing terminal equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal equipment realize the functions specified in the flowcharts and / or block diagrams. Figure 1 The apparatus that realizes the functions specified in one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0148] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or steps.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or steps.

[0150] Although preferred embodiments of the application have been described, those skilled in the art will recognize that additional modifications and changes can be made thereto without departing from the scope of the present application. Accordingly, the appended claims are intended to cover all such modifications and changes as fall within the scope of the application.

[0151] Finally, it should be noted that the terms "first", "second", and the like, herein do not denote any order, quantity, combination, or importance, but rather are used to distinguish one element from another, and are more especially used for the purpose of identification of conceptual breath. In addition, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0152] The principles and implementations of the present application have been described above with the specific examples. The above descriptions of the embodiments are only used to help understand the method and its core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation and application range can be changed, and the above descriptions of the embodiments should not be understood as limitations of the present application.

Claims

1. A method for integrated management of earthquake services, characterized in that, include: Collect historical earthquake operational data corresponding to historical earthquakes and the number of casualties corresponding to historical earthquake operational data; A deep learning algorithm is used to learn the data relationship between the historical earthquake operational data and the corresponding number of casualties to form a casualty prediction model. Collect the latest target earthquake operational data, and then use the casualty prediction model to identify the target earthquake operational data to determine the casualty prediction result corresponding to the target earthquake operational data. The target earthquake service-related data is associated with and stored in conjunction with the casualty population prediction results, enabling integrated management of all earthquake services. After collecting historical earthquake operational correlation data and the corresponding casualty figures, the following is also included: The random forest algorithm is used to filter the historical earthquake business-related data corresponding to historical earthquakes, and the filtered historical earthquake business-related data is determined. In the subsequent process, a deep learning algorithm is used to learn the data relationship between the historical earthquake business-related data and the corresponding number of casualties. A deep learning algorithm is used to learn the data relationship between the historical earthquake operational data and the corresponding number of casualties, forming a casualty prediction model, including: The initial model is constructed using deep learning algorithms; The hyperparameters of the initialization model are initialized and encoded to obtain multiple different hyperparameter codes; Based on the historical earthquake operational data and the corresponding number of casualties, the loss function value corresponding to each hyperparameter code is obtained; The optimal hyperparameter encoding is determined based on the loss function value corresponding to the hyperparameter encoding. For any hyperparameter encoding, an adaptive position selection mechanism is used to perform an initial neighborhood search on the hyperparameter encoding to obtain the hyperparameter encoding after the initial neighborhood search. For any hyperparameter encoding after initializing the neighborhood search, based on the optimal hyperparameter encoding, a variable spiral search mechanism is used to perform a variable spiral search on the hyperparameter encoding after initializing the neighborhood search, resulting in the hyperparameter encoding after the variable spiral search. For any hyperparameter encoding after a variable spiral search, a fuzzy centroid variable speed search mechanism is used to adaptively search the hyperparameter encoding after the variable spiral search, and the hyperparameter encoding after the adaptive search is obtained. For any hyperparameter encoding after adaptive search, a focused mutation search mechanism is used to perform a global search on the hyperparameter encoding after adaptive search, so as to obtain the hyperparameter encoding after global search. Determine whether the training termination condition is met. If so, determine the target hyperparameter encoding based on the hyperparameter encoding after global search. Otherwise, return to the step of obtaining the loss function value. Based on the target hyperparameter encoding and the initialization model, a casualty population prediction model is obtained.

2. The earthquake service integration management method according to claim 1, characterized in that, The hyperparameters of the initialized model are initialized and encoded to obtain multiple different hyperparameter codes, including: For any hyperparameter of the initialization model, the hyperparameter is randomly initialized between the upper and lower bounds of the hyperparameter, and the initialized hyperparameter is encoded into a vector to obtain the hyperparameter encoding. This process is repeated multiple times to obtain multiple different hyperparameter encodings.

3. The earthquake service integration management method according to claim 1, characterized in that, Based on the historical earthquake operational data and their corresponding casualty figures, the loss function value corresponding to each hyperparameter code is obtained, including: For any hyperparameter encoding, the hyperparameter encoding is applied to the initialization model, and the historical earthquake business association data is used as the input data, the number of casualties corresponding to the historical earthquake business association data is used as the expected output data, and the root mean square loss function is used to obtain the loss function value corresponding to the hyperparameter encoding. Iterate through all hyperparameter codes and obtain the loss function value corresponding to each hyperparameter code.

4. The earthquake service integration management method according to claim 1, characterized in that, For any hyperparameter encoding, an adaptive position selection mechanism is used to perform an initial neighborhood search on the hyperparameter encoding, resulting in the following hyperparameter encoding after the initial neighborhood search: in, Indicates the first t During the training process, the first i Hyperparameter encoding, Indicates the first i Hyperparameter encoding after initial neighborhood search i =1,2,...,M, where M represents the total number of hyperparameter codes. Represents the natural constant. Indicates the maximum number of training iterations. Represents the sine function. Represents the first random number between (0,1). This represents the pre-defined decision factors. This represents the second random number between (0,1).

5. The earthquake service integration management method according to claim 1, characterized in that, For any hyperparameter encoding after initializing the neighborhood search, based on the optimal hyperparameter encoding, and using a variable spiral search mechanism, a variable spiral search is performed on the hyperparameter encoding after the initializing neighborhood search, resulting in the following hyperparameter encoding after the variable spiral search: in, Indicates the first t During the training process, the first k Hyperparameter encoding after initial neighborhood search Indicates the first k Hyperparameter encoding following a variable spiral search k =1,2,...,M, where M represents the total number of hyperparameter codes. Indicates the first t The optimal hyperparameter encoding during the training process. Represents the natural constant. Represents the spiral shape factor. This represents a random spiral direction factor between [-1, 1]. Represents pi (π). Represents the cosine function. This represents the control coefficient for the spiral shape factor. This indicates the maximum number of training iterations.

6. The earthquake service integration management method according to claim 1, characterized in that, For any hyperparameter encoding resulting from a variable spiral search, a fuzzy centroid variable speed search mechanism is used to adaptively search the hyperparameter encoding resulting from the variable spiral search, resulting in the following hyperparameter encoding: The search progress parameter and the coding diversity parameter are obtained as follows: in, This indicates the search progress parameter. Indicates the first m The coding diversity parameter corresponding to the hyperparameter encoding after a variable spiral search. m =1,2,...,M, where M represents the total number of hyperparameter codes. Indicates the first t The loss function value corresponding to the optimal hyperparameter encoding during this training process. Indicates the first t The loss function value corresponding to the optimal hyperparameter encoding during training iteration -1. Indicates the first t During the training process, the first m Hyperparameter encoding following a variable spiral search Indicates the first t During the training process, the first h Hyperparameter encoding following a variable spiral search h =1,2,...,M; express and The Euclidean distance between them; Based on the search progress parameters, the first search factor, the second search factor, and the third search factor are determined as follows: in, Indicates the first search factor. Indicates the second search factor. Indicates the third search factor; Based on the coded diversity parameters, the first diversity impact factor and the second diversity impact factor are obtained as follows: in, This represents the first diversity influencing factor. This represents the second diversity influencing factor. This indicates a preset diversity threshold; Based on the first search factor, the second search factor, the third search factor, the first diversity influence factor, and the second diversity influence factor, the fuzzy centroid weights are obtained as follows: in, Indicates the first m The fuzzy centroid weights corresponding to the hyperparameter encoding after a variable spiral search. This represents the minimum value corresponding to the fuzzy centroid weight. This represents the maximum value corresponding to the weight of the fuzzy centroid; Based on the fuzzy centroid weights, an adaptive search is performed on the hyperparameter encoding after the variable spiral search, resulting in the following hyperparameter encoding after the adaptive search: in, Indicates the first m Hyperparameter encoding after an adaptive search Indicates the first t During the training process, the first m The adaptive search quantity corresponding to the hyperparameter encoding after a variable spiral search. Indicates the first t +1 training session m The adaptive search quantity corresponding to the hyperparameter encoding after a variable spiral search. Indicates the first learning factor. Indicates the second learning factor. This represents a third random number between (0,1). This represents the fourth random number between (0,1). express The corresponding historical best value, Indicates the first t The optimal hyperparameter encoding during the training process.

7. The earthquake service integration management method according to claim 1, characterized in that, For any hyperparameter encoding obtained after adaptive search, a focused mutation search mechanism is used to perform a global search on the hyperparameter encoding obtained after adaptive search, resulting in the following hyperparameter encoding: The sampling factor is obtained as follows: in, Indicates the first t During the training process, the first n The hyperparameter encoding after the first adaptive search d dimensional hyperparameters, n =1,2,...,M, where M represents the total number of hyperparameter codes. d =1,2,...,D, where D represents the total number of hyperparameters in the hyperparameter encoding. Indicates the first t The optimal hyperparameter encoding during the training process is the first... d dimensional hyperparameters, Represents pi (π). Represents the cosine function. Indicates the first t During the training process, the first n The hyperparameter encoding after the first adaptive search d The sampling factor corresponding to the dimensional hyperparameter; Based on the sampling factor, a global search is performed on the hyperparameter encoding after the adaptive search, resulting in the following hyperparameter encoding after the global search: in, Indicates the first n The hyperparameter encoding after the first global search d dimensional hyperparameters, This represents a global search control factor that is random between (0,1). Indicates As the mean, with is a Gaussian distributed random number with variance .

8. The earthquake service integration management method according to claim 1, characterized in that, Determine whether the training termination condition is met, including: if the number of training sessions is greater than or equal to the preset maximum number of training sessions, then the training termination condition is met; otherwise, the training termination condition is not met.

Images