Emergency Decision-making Effect Evaluation Method for Dam-break Emergency Incidents with Multi-factor Coupling

Abstract

The present invention discloses a method for evaluating the emergency decision-making effect of dam-break emergencies with multi-factor coupling. This method characterizes the mapping and evolution mechanism between the emergency decision-making elements of dam-break emergencies and the disaster-causing consequences through the coupling of organization-personnel-technology. A method for evaluating the emergency decision-making effect after a disaster based on a mathematical model is proposed, which depicts the emergency decision-making process of dam-break emergencies driven by data and mechanism, and quantifies the emergency decision-making effect. The present invention can reasonably and reliably evaluate the emergency decision-making effect of dam-break emergencies, reduce the disaster-causing consequences, and provide a scientific basis for timely and effectively implementing emergency disposal decisions.

Description

Technical Field

[0001] The present invention relates to a method for evaluating the effect of emergency decision-making for dam-break emergencies with multi-factor coupling, specifically referring to forming a scientific and quantitative basis for judging the effect of disaster emergency decision-making by establishing the multi-factor coupling relationship of organization-personnel-technology. Background Art

[0002] Emergency decision-making for dam-break emergencies is a key means to reduce disaster-causing consequences. It refers to the dynamic process in which, after the dam-break of a reservoir dam occurs, the decision-making subject collects information and data from all parties in the current subjective and objective environment, analyzes and judges the situation relying on prior knowledge, technology, and experience, and formulates an emergency response plan within a short time to control the development of the disaster situation until the event is eliminated. With the development of the economy and society, the protected objects downstream of reservoir dams in China have undergone profound changes (such as an increase in the population, an increase in infrastructure, and an expansion of urban scale, etc.). The dam-break of a reservoir dam will face unprecedented losses of life and property. Effective emergency decision-making for dam-break emergencies is a key means to reduce disaster-causing consequences, especially to ensure "no casualties".

[0003] Emergency decision-making for dam-break emergencies involves the coupling of multiple dynamic factors of people (emergency organizations, downstream masses), water (dam-break flood), and materials (rescue and emergency materials), and has the characteristics of urgency, spread, and comprehensive complexity. Existing emergency decision-making methods only consider the influence of one or several factors, lack the consideration of comprehensive elements of emergency decision-making, have an unclear understanding of the coupling relationship between elements, a fuzzy definition of the critical early warning decision-making time, and an immeasurable emergency decision-making effect. It is difficult to form a scientific and timely emergency response plan and cannot be effectively applied during the emergency process. In dam-break emergencies, a wrong decision-making plan may even lead to a multiple increase in disaster consequences and losses. Summary of the Invention

[0004] Object of the Invention: Aiming at the problems and deficiencies in the prior art, the present invention provides a method for evaluating the effect of emergency decision-making for dam-break emergencies with multi-factor coupling, which can extract and learn the quantitative relationship between decision-making factors and decision-making effects, and simulate and predict the results of emergency decision-making for dam-break emergencies through machine learning.

[0005] Technical Solution: A method for evaluating the effect of emergency decision-making for dam-break emergencies with multi-factor coupling, by establishing the multi-factor coupling relationship of organization-personnel-technology, modeling and representing the effect of emergency decision-making; combining the dam engineering elements, quantifying the risk perception level of the emergency organization, the behavior of dispatching emergency materials, the follow-up rate of rescue materials, the number of at-risk people, the evacuation rate, the average inundation degree of the disaster area, and the number of people losing their lives, to form a basis for judging the effect of disaster emergency decision-making; establishing a neural network model - a learning model for the effect of emergency decision-making for dam-break emergencies based on a shared learning architecture, estimating the dam risk level and the number of people losing their lives, so as to quantify the effect of post-disaster emergency decision-making.

[0006] Build a learning model for the emergency decision-making effect of dam-break emergencies based on a shared learning architecture; the input of this model is the emergency decision-making elements of dam-break emergencies, including: (1) the risk level perceived by the emergency organization, (2) the behavior of allocating emergency rescue materials, (3) the follow-up rate of rescue materials, (4) the number of at-risk people, (5) the evacuation rate, and (6) the average inundation degree of the disaster area; the output of the model is the emergency decision-making effect after the disaster, including: (1) the risk level of the dam, and (2) the number of people who lost their lives; optimize the learning model for the emergency decision-making effect of dam-break emergencies through historical data-driven; input the emergency decision-making elements of the dam-break emergency at the current moment and output the emergency decision-making effect at the current moment.

[0007] The input of the risk level perceived by the emergency organization, the behavior of allocating emergency rescue materials, the follow-up rate of rescue materials, the number of at-risk people, the evacuation rate, and the average inundation degree of the disaster area is expressed as follows:

[0008] X = {a, b, c, d, e, f}

[0009] In the formula, a is the risk level perceived by the emergency organization, b is the behavior of allocating emergency rescue materials, c is the follow-up rate of rescue materials, d is the evacuation rate, e is the number of at-risk people, and f is the average inundation degree of the disaster area; among them, a and b are auxiliary variables, c and d are speed variables, and e and f are state variables.

[0010] The output of the learning model for the emergency decision-making effect of dam-break emergencies based on the shared learning architecture to obtain the risk level of the dam and the number of people who lost their lives is expressed as follows:

[0011] y = {θ, λ}

[0012] In the formula, θ represents the risk level of the dam; λ represents the number of people who lost their lives.

[0013] The model architecture is as follows:

[0014] The first layer is the input layer: the attributes of the input data divide the input into three groups, corresponding to the auxiliary variable, the speed variable, and the state variable respectively, comprehensively describing the organizational, personnel, and technical elements;

[0015] The second layer is the organizational-personnel-technical coupling relationship calculation layer, which is used to calculate the quantitative value of the coupling relationship among the three;

[0016] The third layer is the coupling weighting module layer, which uses the coupling quantitative value as the weight to perform weighted calculation and splicing on the vectors composed of the auxiliary variable, the speed variable, and the state variable respectively, forming three-channel features of the auxiliary variable, the speed variable, and the state variable;

[0017] The fourth layer is the attention weighting module layer, which performs attention weight weighting on the three-channel features;

[0018] The fifth layer is the splicing layer, which splices the attention-weighted features to form multi-factor comprehensive features;

[0019] The sixth to eighth layers are the dual-stream identification layer, including dual-stream convolutional layers, pooling layers and Softmax layers, which are used to output the dam risk level and the number of people's lives lost, representing the emergency decision-making effect at the current moment.

[0020] Calculate the coupling relationship among the three factors of organization, personnel and technology;

[0021] Among them, the organization, personnel and technology factors are:

[0022] x1 = [a, b], x2 = [c, d], x3 = [e, f]

[0023] Calculate the coupling relationship weights among the three factors of organization-personnel-technology:

[0024] Q 11 = <x1, x1>, Q 12 = <x1, x2>, Q 13 = <x1, x3>

[0025] Q 21 = <x2, x1>, Q 22 = <x2, x2>, Q 23 = <x2, x3>

[0026] Q 31 = <x3, x1>, Q 32 = <x 31 , x2>, Q 33 = <x3, x3>

[0027] Among them, <> is the dot product operation. For <x1, x1>,

[0028] Based on the quantization results of the factor coupling relationship, weight the vectors composed of auxiliary variables, velocity variables and state variables respectively:

[0029] F1 = concat(Q 11 x1 T , Q 12 x1 T , Q 13 x1 T )

[0030] F2 = concat(Q 21 x2 T , Q 22 x2 T , Q 23 x2T )

[0031] F3 = concat(Q 31 x3 T , Q 32 x3 T , Q 33 x3 T )

[0032] Among them, concat() is the splicing calculation, and the comprehensive feature is obtained by weighted splicing of the attention weights of F1, F2, and F3:

[0033] F = con(W1F1, W2F2, W3F3)

[0034] Among them, W1, W2, and W3 are the attention weights corresponding to the organizational, personnel, and technical element features;

[0035] The spliced comprehensive feature F is input into the pooling layer and the softmax layer for calculation to obtain the dam risk level θ and the number of people λ of mass life losses.

[0036] Optimize the emergency decision-making effect learning model for the dam break emergency through historical data-driven;

[0037] First, construct a loss function for optimizing the dam risk level and the number of people of mass life losses, and the specific expression is as follows:

[0038] L = ɑ1L1 + α2L2

[0039] In the formula, L1 represents the loss function for optimizing the dam risk level, and L2 represents the loss function for optimizing the number of people of mass life losses.

[0040] Beneficial effects: The emergency decision-making effect evaluation method for dam break emergencies with multi-factor coupling provided by the present invention takes into account the coupling between the organizational, personnel, and technical dynamic factors involved in the emergency decision-making of dam break emergencies. It breaks through the error of a single factor in predicting the emergency decision-making effect, accurately evaluates the emergency decision-making effect, and improves the effectiveness of emergency decision-making for disaster prevention and mitigation. Brief Description of the Drawings

[0041] Figure 1 It is the model structure diagram of the embodiment of the present invention. Detailed Embodiments

[0042] The following further clarifies the present invention in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent forms of modification by those skilled in the art fall within the scope defined by the appended claims of this application.

[0043] A method for evaluating the emergency decision-making effect of dam-break emergencies with multi-factor coupling, which establishes the multi-factor coupling relationship of organization-personnel-technology and models to represent the emergency decision-making effect; combines the dam engineering elements to quantify the emergency organization's perceived risk level, the behavior of emergency rescue material allocation, the follow-up rate of rescue materials, the number of at-risk people, the evacuation rate, the average inundation degree of the disaster area, and the number of people's lives lost, forming the basis for judging the emergency decision-making effect of disasters; establishes a neural network model - a learning model for the emergency decision-making effect of dam-break emergencies based on a shared learning architecture to estimate the dam risk level and the number of people's lives lost, so as to quantify the post-disaster emergency decision-making effect.

[0044] Construct a learning model for the emergency decision-making effect of dam-break emergencies based on a shared learning architecture; the inputs of the model are the emergency decision-making elements of dam-break emergencies, including: (1) the emergency organization's perceived risk level, (2) the behavior of emergency rescue material allocation, (3) the follow-up rate of rescue materials, (4) the number of at-risk people, (5) the evacuation rate, (6) the average inundation degree of the disaster area; the outputs of the model are the post-disaster emergency decision-making effects, including: (1) the dam risk level, (2) the number of people's lives lost; optimize the learning model for the emergency decision-making effect of dam-break emergencies through historical data-driven; input the emergency decision-making elements of the dam-break emergency at the current moment and output the emergency decision-making effect at the current moment.

[0045] The input expression of the model is as follows:

[0046] X = {a, b, c, d, e, f}

[0047] In the formula, a is the emergency organization's perceived risk level, b is the behavior of emergency rescue material allocation, c is the follow-up rate of rescue materials, d is the evacuation rate, e is the number of at-risk people, f is the average inundation degree of the disaster area; among them, a and b are auxiliary variables, c and d are speed variables, and e and f are state variables;

[0048] According to the attributes of the input data, the input is divided into three groups,

[0049] x1 = [a, b], x2 = [c, d], x3 = [e, f]

[0050] So far, the input of the learning model for the emergency decision-making of dam-break emergencies based on a shared learning architecture is completed.

[0051] The learning module consists of 2 parallel streams, and each parallel stream receives 3 groups of inputs at the same time and inputs them into the shared layer. The specific expression is as follows:

[0052]

[0053] In the formula, For the shared weights, x′1, x′2, and x′3 are the eigenvalues after shared learning respectively. They are input into a convolutional layer with a convolutional kernel size of 1×5 for convolutional calculation, and the outputs are x″1, x″2, and x″3, which are then input into the concat(function) layer for concatenation. The specific expression is as follows:

[0054] X″ = [x″1 x″2 x″3]

[0055] The concatenated feature X″ is input into the pooling layer and the softmax(soft maximum) layer for calculation to obtain the dam risk level θ and the number of people λ of mass casualties. The feature sequence obtained through shared learning is continuously input into a convolutional layer with a convolutional kernel size of 1×5 for convolutional calculation, and then further input into the concat layer for concatenation. The concatenated feature is input into the pooling layer and the softmax layer for calculation to obtain the calculation results of the dam risk level and the number of people of mass casualties.

[0056] Thus, the output of the learning model for emergency decision-making effects of dam-break emergencies based on the shared learning architecture is completed.

[0057] During the optimization process of the model, historical database-driven model optimization is adopted:

[0058] A loss function for optimizing the dam risk level and the number of people of mass casualties is constructed. The specific expression is as follows:

[0059] L = ɑ1L1 + ɑ2L2

[0060] In the formula, L1 represents the loss function for optimizing the dam risk level, and L2 represents the loss function for optimizing the number of people of mass casualties;

[0061] Adopting historical database-driven, the parameters of the shared layer are optimized through the backpropagation algorithm. The specific expression is as follows:

[0062]

[0063] Then the weight parameters are updated through backpropagation. The specific method is as follows:

[0064]

[0065] The following expression is obtained through the update of the weight parameters:

[0066]

[0067] In the formula, ε represents the learning rate of multi-task learning.

[0068] Thus, the optimization of the learning model for emergency decision-making effects of dam-break emergencies based on the shared learning architecture is completed.

[0069] The output expression of the emergency decision-making effect learning model for dam-break emergencies based on a shared learning architecture is as follows:

[0070] y = {θ, λ}

[0071] Where θ represents the dam risk level; λ represents the number of people's lives lost.

[0072] As Figure 1 shown, the model architecture is:

[0073] The first layer is the input layer: The attributes of the input data are divided into three groups, corresponding to the auxiliary variables, velocity variables, and state variables respectively, comprehensively describing the organizational, personnel, and technical elements;

[0074] The second layer is the organization-personnel-technology coupling relationship calculation layer, which is used to calculate the quantitative value of the coupling relationship among the three;

[0075] The third layer is the coupling weighting module layer, which uses the coupling quantitative value as the weight to perform weighted calculation and splicing on the vectors composed of the auxiliary variables, velocity variables, and state variables respectively, forming three-channel features of the auxiliary variables, velocity variables, and state variables;

[0076] The fourth layer is the attention weighting module layer, which performs attention weight weighting on the three-channel features;

[0077] The fifth layer is the splicing layer, which splices the features after attention weighting to form multi-element comprehensive features;

[0078] The sixth to eighth layers are the two-stream identification layer, including the two-stream convolutional layer, pooling layer, and Softmax layer, which are used to output the dam risk level and the number of people's lives lost, representing the emergency decision-making effect at the current moment.

[0079] Calculate the coupling relationship weights among the three elements of organization-personnel-technology:

[0080] Q 11 =<x1, x1>, Q 12 =<x1, x2>, Q 13 =<x1, x3>

[0081] Q 21 =<x2, x1>, Q 22 =<x2, x2>, Q 23 =<x2, x3>

[0082] Q 31 =<x3, x1>, Q 32 =<x 31 , x2>, Q 33 =<x3, x3>

[0083] Among them, <> represents the dot product operation. Taking <x1, x1> as an example,

[0084] In the quantization result based on the factor coupling relationship, weights are respectively assigned to the vectors composed of the auxiliary variable, the velocity variable, and the state variable:

[0085] F1 = concat(Q 11 x1 T , Q 12 x1 T , Q 13 x1 T )

[0086] F2 = concat(Q 21 x2 T , Q 22 x2 T , Q 23 x2 T )

[0087] F3 = concat(Q 31 x3 T , Q 32 x3 T , Q 33 x3 T )

[0088] Among them, concat() is the concatenation calculation. The comprehensive feature is obtained by performing attention weight weighting and concatenation on F1, F2, and F3:

[0089] F = concat(W1F1, W2F2, W3F3)

[0090] Among them, W1, W2, and W3 are the attention weights corresponding to the organizational, personnel, and technical element features;

[0091] The concatenated comprehensive feature F is input into the pooling layer and the softmax layer for calculation to obtain the dam risk level θ and the number of people λ of mass life losses.

Claims

1. A method for evaluating the emergency decision-making effect of dam-break emergencies with multi-factor coupling, characterized in that, By establishing the multi-factor coupling relationship of organization-personnel-technology, the emergency decision-making effect is modeled and characterized; combined with the dam project elements, the risk perception level of the emergency organization, the behavior of emergency rescue material allocation, the follow-up rate of rescue materials, the number of at-risk people, the evacuation rate, the average inundation degree of the disaster area, and the number of people's lives lost are quantified to form the judgment basis for the disaster emergency decision-making effect; a neural network model is established to estimate the dam risk level and the number of people's lives lost to quantify the post-disaster emergency decision-making effect. Construct an emergency decision-making effect learning model for dam-break emergencies based on a shared learning architecture. The model architecture is as follows: The first layer is the input layer: The attributes of the input data are divided into three groups, corresponding to auxiliary variables, velocity variables, and state variables respectively, comprehensively describing the organization, personnel, and technology elements. The second layer is the organization-personnel-technology coupling relationship calculation layer, which is used to calculate the quantified value of the coupling relationship among the three. The third layer is the coupling weighting module layer. Taking the coupling quantified value as the weight, the vectors composed of auxiliary variables, velocity variables, and state variables are weighted and calculated and spliced respectively to form three-channel features of auxiliary variables, velocity variables, and state variables. The fourth layer is the attention weighting module layer, which performs attention weight weighting on the three-channel features. The fifth layer is the splicing layer, which splices the attention-weighted features to form multi-factor comprehensive features. The sixth to eighth layers are the two-stream identification layer, including two-stream convolutional layer, pooling layer, and Softmax layer, which are used to output the dam risk level and the number of people's lives lost to characterize the emergency decision-making effect at the current moment.

2. The emergency decision-making effect evaluation method for dam-break emergencies with multi-factor coupling according to claim 1, characterized in that Construct an emergency decision-making effect learning model for dam-break emergencies based on a shared learning architecture. The input of this model is the emergency decision-making elements of dam-break emergencies, including: (1) the risk perception level of the emergency organization, (2) the behavior of emergency rescue material allocation, (3) the follow-up rate of rescue materials, (4) the number of at-risk people, (5) the evacuation rate, (6) the average inundation degree of the disaster area; the output of the model is the post-disaster emergency decision-making effect, including: (1) the dam risk level, (2) the number of people's lives lost; the emergency decision-making effect learning model for dam-break emergencies is optimized through historical data-driven; the emergency decision-making elements of the dam-break emergency at the current moment are input, and the emergency decision-making effect at the current moment is output.

3. The multi-factor coupling emergency decision-making effect evaluation method for dam-break emergencies according to claim 2, characterized in that, The input of the risk perception level of the emergency organization, the behavior of emergency rescue material allocation, the follow-up rate of rescue materials, the number of at-risk people, the evacuation rate, and the average inundation degree of the disaster area is expressed as follows: X = {a, b, c, d, e, f} In the formula, a is the risk perception level of the emergency organization, b is the behavior of emergency rescue material allocation, c is the follow-up rate of rescue materials, d is the evacuation rate, e is the number of at-risk people, f is the average inundation degree of the disaster area; among them, a and b are auxiliary variables, c and d are velocity variables, and e and f are state variables. The output of the emergency decision-making effect learning model for dam-break emergencies based on the shared learning architecture to obtain the dam risk level and the number of people's lives lost is expressed as follows: y = {θ, λ} In the formula, θ represents the dam risk level; λ represents the number of people's lives lost.

4. The method for evaluating the emergency decision-making effect of dam-break emergencies with multi-factor coupling according to claim 3, wherein Calculate the coupling relationship among the three elements of organization, personnel, and technical elements; Among them, the organization, personnel, and technical elements are: x1 = [a, b], x2 = [c, d], x3 = [e, f] Calculate the coupling relationship weights among the three elements of organization-personnel-technology: Q 11 = <x1, x1>, Q 12 = <x1, x2>, Q 13 = <x1, x3> Q 21 = <x2, x1>, Q 22 = <x2, x2>, Q 23 = <x2, x3> Q 31 = <x3, x1>, Q 32 = <x 31 , x2>, Q 33 = <x3, x3> where, <> is the dot product operation, for 5. The method for evaluating the emergency decision-making effect of dam-break emergencies with multi-factor coupling according to claim 3, characterized in that, Based on the quantization results of the element coupling relationship, weight the vectors composed of auxiliary variables, speed variables, and state variables respectively: F1 = concat(Q 11 x1 T , Q 12 x1 T , Q 13 x1 T ) F2 = concat(Q 21 x2 T , Q 22 x2 T , Q 23 x2 T ) F3 = concat(Q 31 x3 T , Q 32 x3 T , Q 33 x3 T ) Among them, concat() is the splicing calculation, and the attention weight weighted splicing of F1, F2, and F3 is performed to obtain the comprehensive feature: F = concat(W1F1, W2F2, W3F3) Among them, W1, W2, and W3 are the attention weights corresponding to the organization, personnel, and technical element features; Input the spliced comprehensive feature F into the pooling layer and the softmax layer for calculation to obtain the dam risk level θ and the number of people λ of mass life losses.

6. The method for evaluating the emergency decision-making effect of dam-break emergencies with multi-factor coupling according to claim 3, wherein, Optimize the emergency decision-making effect learning model for dam break emergencies through historical data driving; First, construct a loss function for optimizing the dam risk level and the number of people of mass life losses, and the specific expression is as follows: L = α1L1 + α2L2 In the formula, L1 represents the loss function for optimizing the dam risk level, and L2 represents the loss function for optimizing the number of people of mass life losses.

Images