Cross-regional rescue force distribution method based on dynamic resource game

Abstract

The invention discloses a cross-regional rescue force distribution method based on a dynamic resource game, and the method comprises the following steps: S1, collecting disaster data, carrying out the standardization processing, and extracting a feature set; s2, constructing a disaster scene graph based on the feature set, and generating a preliminary disaster scene model; s3, defining a game model, and generating a preliminary resource allocation model; s4, introducing an anti-factual reasoning algorithm, and optimizing the resource allocation model; s5, constructing a resource excitation module, and optimizing an excitation strategy; s6, predicting a resource demand based on the excitation strategy, and generating a scheduling scheme; and S7, in combination with the feedback information, generating a cross-regional rescue distribution scheme. Through the optimization algorithm based on the dynamic resource game, efficient, fair and dynamic resource allocation of cross-regional rescue force is realized, and the disaster rescue response speed and the resource utilization efficiency are remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of game theory application, and in particular to a cross-regional rescue force allocation method based on dynamic resource game. Background Art

[0002] With the frequent occurrence of natural disasters, cross-regional deployment of rescue forces and resource allocation have become particularly important. Post-disaster emergency rescue tasks involve multi-party coordination, and rescue resources (such as medical teams, supplies, equipment, etc.) need to be accurately and quickly deployed to the worst-hit areas. However, due to the diversity and complexity of disasters, traditional cross-regional resource scheduling methods have obvious deficiencies in efficiency, response time, and fairness. Existing technologies generally rely on static scheduling strategies and resource allocation based on predetermined rules. These methods are difficult to cope with the complex needs brought about by sudden disasters, and usually ignore the real-time deployment and dynamic changes of resources, resulting in limited optimization and fairness of resource allocation.

[0003] In the traditional rescue force allocation method, the allocation of resources in the disaster area mainly relies on manual instructions and experience. Especially when large-scale disasters occur, the shortage of resources and cross-regional coordination issues are more complicated, and often need to rely on national or local governments for dispatch. Although this type of method can cope with disasters of a certain scale, it often lacks flexibility and adaptability when faced with more sudden and rapidly changing complex disasters. In addition, due to the lag in information acquisition, decisions are usually based on pre-disaster resource planning and standardized processes, resulting in low efficiency and accuracy in resource dispatch. In addition, since the allocation of rescue resources is usually the responsibility of several major participants, the resource allocation process will be affected by the type of resources, the fairness of allocation, and the needs of the disaster area. Some areas may have excess resources while others may lack resources.

[0004] Another noteworthy shortcoming is that existing methods generally ignore the spatiotemporal correlation of resource demand between disaster areas. Disaster relief not only needs to consider the timeliness of rescue forces, but also needs to effectively evaluate the diffusion and propagation paths of resource demand between disaster areas. However, current resource allocation methods often use a single indicator to judge demand, lacking real-time tracking and prediction of dynamic changes in disasters. This makes it difficult for resource scheduling systems to respond quickly to sudden disasters, unable to adapt to the complexity and variability of disaster development, and resulting in lagging and imbalanced resource allocation.

[0005] In addition, current technologies lack effective incentive mechanisms and game theory models in resource allocation. In traditional rescue resource allocation methods, the interests of various participants are often not effectively adjusted, resulting in unclear priorities or conflicts of interest in the resource allocation process. Especially in cross-regional scheduling, due to the different resource demands, disaster severity and response speeds in various regions, how to promote the participation of all participants in resource allocation decisions through reasonable incentive mechanisms has become the key to solving the problems of fairness and efficiency in resource allocation. However, existing technologies have not fully introduced game theory optimization models and dynamic incentive strategies, resulting in resource allocation decisions failing to achieve an optimal balance between the various participants, often leading to serious imbalances in resource allocation in individual regions, affecting the efficiency of the overall rescue work.

[0006] Furthermore, most traditional resource allocation methods make decisions based on static algorithms or manual calculations, ignoring real-time data feedback and adjustments in dynamic environments. As the situation at the disaster site is constantly changing, the relationship between resource demand and supply may fluctuate at any time, which places higher demands on the real-time nature of resource scheduling. However, traditional methods mostly remain in the disaster warning stage or the post-disaster stage, lacking a scheduling model based on real-time dynamic information. This static method cannot make accurate resource allocation decisions quickly in the first time after a disaster occurs, resulting in inefficient resource scheduling and may even miss the best time for rescue.

[0007] Finally, another major shortcoming of existing technologies is the lack of a risk assessment mechanism for resource allocation. In the process of resource allocation, there may be an oversupply or undersupply of resources, making it difficult for rescue missions to proceed smoothly. Especially when dispatching across regions, the demand and transportation conditions in the disaster area may change dramatically, and most existing technologies do not introduce dynamic risk control modules. In this context, how to dynamically optimize resource allocation strategies based on feedback information from the disaster site, adjust the relationship between resource supply and demand in a timely manner, and ensure the effective use of rescue resources and risk avoidance has become an urgent problem to be solved.

[0008] Therefore, how to provide a cross-regional rescue force allocation method based on dynamic resource game is a problem that technicians in this field urgently need to solve. Summary of the invention

[0009] One purpose of the present invention is to propose a cross-regional rescue force allocation method based on dynamic resource game. The present invention makes full use of advanced algorithms such as game theory, dynamic resource scheduling, counterfactual reasoning, and reinforcement learning, and describes in detail how to dynamically optimize the cross-regional rescue resource allocation plan based on disaster site data and resource requirements of each region after a disaster occurs. By establishing a multi-level disaster scenario model and combining game optimization and incentive mechanisms, the present invention can effectively deploy and allocate rescue forces, improve resource utilization, rescue efficiency, and allocation fairness. Compared with the prior art, the present invention has the advantages of real-time dynamic adjustment, cross-regional fair allocation, and risk control, ensuring that rescue resources can be optimally allocated and used efficiently in complex and changeable disaster environments.

[0010] The cross-regional rescue force allocation method based on dynamic resource game according to an embodiment of the present invention comprises the following steps:

[0011] S1, collect disaster multi-source data, perform standardization processing and multi-dimensional feature extraction, and generate disaster scene feature sets;

[0012] S2. Based on the disaster scenario feature set, a dynamic multi-scale disaster scenario map is constructed to characterize the resource demand diffusion law and transportation path between disaster areas, and generate a preliminary disaster scenario model;

[0013] S3. Based on the disaster scenario model, define the dynamic resource game model in cross-regional rescue and generate a preliminary resource allocation model;

[0014] S4. Combining the disaster scenario feature set, disaster scenario model and preliminary resource allocation model, a multi-level counterfactual reasoning algorithm is introduced to conduct supply and demand analysis and generate an optimized resource allocation model;

[0015] S5. Use the optimized resource allocation model and supply and demand analysis results to build a resource adaptive incentive module and use the dynamic game algorithm to optimize the incentive strategy;

[0016] S6. Based on the optimized incentive strategy, combined with the dynamic response prediction network and multimodal sequence modeling algorithm, the time series changes of disaster propagation trends and resource demand are predicted to generate resource scheduling plans;

[0017] S7. Based on the resource scheduling plan, a dynamic risk control module is constructed. Combined with the feedback information from the disaster site, the resource allocation model is iteratively optimized to generate a cross-regional rescue allocation plan.

[0018] Optionally, the S3 specifically includes:

[0019] S31. Set the set of participants in the game model P = {p 1 ,p 2 ,…,p n}, where n represents the number of disaster areas, and each participant p i The corresponding resource demand vector is D i ={d i1 ,d i2 ,…,d im}, where m is the number of resource types, d ij represents the demand of the i-th region for the j-th type of resources;

[0020] S32, define a policy set S = {s 1 ,s 2 ,…,s n}, where s i ={s i1 ,s i2 ,…,s im} is the resource allocation strategy for the i-th region, satisfying the following constraints:

[0021]

[0022] Among them, R j is the total supply of the jth type of resource, s ij Assign to p i The amount of resources of the jth category;

[0023] S33. For each participant p i Define the payment function U i (s), used to quantify the benefits, taking into account rescue efficiency, time cost, resource consumption and distribution fairness:

[0024]

[0025] Among them, α, β, γ are weight coefficients, η ij is the response efficiency, τ i is the rescue time of the ith area, c ij is the unit cost of the jth type of resource, and m is the number of resource types;

[0026] S34, based on the payment function U i (s), define the global optimization objective function Φ(s), which is expressed as:

[0027]

[0028] in, is the variance of the allocation ratio, n represents the number of disaster areas, and λ is the weight parameter of allocation fairness;

[0029] S35. Use a method combining reinforcement learning and game optimization to solve the objective function Φ(s):

[0030]

[0031] Among them, s * is the final optimized strategy solution, s is the current strategy, argmax s∈S The corresponding strategy solution s when solving the function Φ(s) to maximize * , ρ is the strategy learning rate of reinforcement learning, is the gradient of the global optimization objective function Φ(s) with respect to the parameter θ, The weight parameters for game optimization, is the global average return;

[0032] S36. Strategy solution based on final optimization * and resource constraints to generate a preliminary resource allocation model.

[0033] Optionally, the S4 specifically includes:

[0034] S41, based on the disaster scene feature set D = {D 1 ,D 2 ,…,D n}, the preliminary resource allocation model S, combined with the dynamic adjustment function, calculates the supply and demand deviation matrix G = [δ ij ]:

[0035]

[0036] Among them, δ ij is the supply and demand deviation of the i-th region for the j-th resource, d ij is the demand in area i, s ij is the current allocation, ∈ is the regularization parameter;

[0037] S42. Define the counterfactual loss function L cf (h,G) Quantify the supply and demand balance of different resource allocation strategies:

[0038]

[0039] Among them, n is the number of disaster areas, m is the number of resource types, λ 1 is the time smoothing regularization weight, h is the resource allocation strategy at the current iteration moment;

[0040] S43. In the high-dimensional strategy space H, set the high-dimensional projection function P(h) to constrain the search range:

[0041]

[0042] Among them, h k ′ is the candidate strategy obtained by the k-th search, K is the number of searches for strategy optimization, η 1is the learning rate, is the gradient of the counterfactual loss function;

[0043] S44, the strategy set H'={h 1 ',h 2 ',…,h K '}, using dynamic constraint game optimization to solve the optimal strategy:

[0044]

[0045] Among them, h * is the optimized resource allocation strategy, h' is the candidate strategy in the current optimization process, Φ(h') is the joint objective function, and h ij is the number of resources of the jth type allocated to the i-th region under the current strategy h, is the global average return, U i (h') is the payoff function of the ith region, argmax h'∈H' To find the optimal resource allocation strategy h in the high-dimensional strategy space H' * , so that the objective function reaches the maximum value;

[0046] S45. Resource allocation strategy based on optimization * , calculate the supply and demand equilibrium G(h * ):

[0047]

[0048] If G(h * ) exceeds the threshold τ, then adjust the strategy:

[0049]

[0050] Among them, ρ o is the equilibrium regulation rate, δ ij (h * ) is the supply-demand deviation term, is the gradient;

[0051] S46, based on the optimized resource allocation strategy h * and supply-demand balance G(h * ), generate an optimized resource allocation model.

[0052] Optionally, the S5 specifically includes:

[0053] S51, based on the optimization of resource allocation model z = {z 1 ,z 2 ,…,z n} and the supply and demand analysis result G(z), define the resource incentive function Ω(z):

[0054]

[0055] Among them, μ 1 is the incentive weight coefficient, μ 2 is the balance adjustment coefficient, ε is the regularization parameter, n is the number of disaster areas, m is the number of resource categories, z ij The amount of resources of the jth type allocated to the i-th disaster area, d ij is the demand of the i-th disaster area for the j-th type of resources;

[0056] S52. Based on the resource incentive function, calculate the revenue adjustment amount Θ(z) of the resource supplier:

[0057]

[0058] Among them, λ k is the dynamic excitation adjustment coefficient;

[0059] S53. Based on the dynamic game optimization goal, a strengthening incentive strategy optimization model is constructed. By using the supply and demand analysis results and the game incentive mechanism, the resource supplier's benefits, equilibrium constraints and distribution stability are comprehensively considered to optimize the objective function:

[0060]

[0061] Among them, Φ * (z) is the objective function of strengthening the incentive strategy optimization, U i (z) is the revenue function of the i-th resource supplier, ξ is the smoothing incentive weight, is the amount of resources of the jth type allocated to the i-th area after the previous round of optimization, and ln is a logarithmic function;

[0062] S54. Use variational adaptive game optimization method to solve the equilibrium solution:

[0063]

[0064] Among them, z * is the resource allocation scheme, argmax z∈H To find a strategy in the strategy space H that can make the objective function Φ * (z) maximizes the optimal strategy z * , α o is the time series smoothing weight;

[0065] S55. Calculate the adaptability of the equilibrium solution:

[0066]

[0067] If Ψ(z *)>α p , then perform balance adjustment:

[0068]

[0069] Among them, α p To adjust the threshold for balance, Ψ(z * ) is the equilibrium adaptive loss function, is the optimal resource allocation, α d is the regularization parameter, is the gradient of the equilibrium loss function;

[0070] S56, based on resource allocation scheme z * and the impact of supply and demand incentives to generate optimized incentive strategies.

[0071] Optionally, the S6 specifically includes:

[0072] S61. Based on resource allocation scheme z * and optimized excitation strategy, defining the input data matrix X t :

[0073]

[0074] Among them, G(z * ) is the supply-demand equilibrium function, is the disaster situation data matrix, ξ k is the disaster propagation impact factor, Φ k (t) is the value of disaster characteristic k at time t;

[0075] S62. Using dynamic response prediction network f DRP , combined with the time dependency, predict the resource demand at the future time t+Δt:

[0076]

[0077] in, is the resource demand matrix, w j is the dynamic weight of resource category j under different disaster scenarios, θ k is the deep reinforcement learning parameter, m is the number of resource types;

[0078] S63, based on multimodal sequence modeling algorithm g MSM , calculate the resource scheduling weight at the future moment:

[0079]

[0080] in, is the resource demand matrix, δ 1 is the weight adjustment coefficient, Tij is the resource scheduling optimization matrix, ψ ij is the dynamic impact factor, ω t+Δt is the resource scheduling weight matrix, n represents the number of disaster areas, and m is the number of resource types;

[0081] S64. Combined with resource demand forecast results and resource scheduling weight ω t+Δt , define the resource scheduling optimization goal:

[0082]

[0083] Among them, Φ * (r) is the resource scheduling optimization objective function, r ij is the resource scheduling scheme, ε 1 is the scheduling stability weight parameter, is the resource allocation scheme, ε 2 is the scheduling time smoothness weight parameter, T is the time step of resource scheduling;

[0084] S65. Using variational reinforcement learning algorithm to solve resource scheduling optimization objectives:

[0085]

[0086] Among them, r * is the resource scheduling scheme, argmax r Φ * (r) is to find the optimal resource scheduling solution r * , is the resource scheduling solution after the previous round of optimization, ε 3 is the adjustment coefficient, ln is the logarithmic function;

[0087] S66, based on resource scheduling scheme r * , and input it into the dynamic resource scheduling system.

[0088] Optionally, the S7 specifically includes:

[0089] S71, based on resource scheduling scheme r * , define the dynamic risk metric for resource allocation:

[0090]

[0091] in, is the resource scheduling risk metric, α ij is the dynamic risk weight, is the allocation in the current resource scheduling scheme, d ij is the demand of disaster area i for resource j, ∈ tTo prevent small positive numbers with a denominator of zero, n is the number of disaster areas and m is the number of resource types;

[0092] S72. Define the standard deviation of distribution fairness σ f , which is used to measure the balance of the current allocation:

[0093]

[0094] When σ f >τ f , adjust the allocation scheme based on the fairness standard deviation:

[0095]

[0096] in, is the average resource allocation ratio of all disaster areas, τ f Threshold parameters optimized for fairness, is the adjusted resource allocation, ρ f To optimize the step size;

[0097] S73. Combine the disaster site feedback information F(t) to adjust the risk-adjusted resource allocation plan Perform optimization and correction to generate the final optimized resource allocation plan

[0098]

[0099] Among them, O f Optimize the weight for feedback, Q 1 and Q 2 They are demand modification weight and feedback response weight respectively;

[0100] S74. Introducing risk aversion function Redistribute and optimize high-risk allocation areas:

[0101]

[0102] Among them, ξ 3 is the risk control weight, ξ 4 is the time smoothing weight, is the resource allocation after the previous round of optimization, ln is the logarithmic function, and t is the time dimension;

[0103] S75. Integrate dynamic risk control, on-site feedback optimization and risk avoidance strategies to output cross-regional rescue allocation plans.

[0104] The beneficial effects of the present invention are:

[0105] The present invention overcomes the problems of delayed resource allocation, low scheduling efficiency, and poor allocation fairness in the prior art by introducing a cross-regional rescue force allocation method based on dynamic resource game. First, with the help of game theory models, the present invention can ensure that resource allocation between regions is more fair and reasonable through dynamic optimization strategies in a complex environment with multiple disaster areas and multiple resource demands. Unlike the traditional resource scheduling method that relies on static rules and manual experience, the present invention can obtain multi-source data at the disaster site in real time, and through multi-level disaster scenario modeling and dynamic game optimization, it achieves accurate matching of resource demand and supply, thereby improving the efficiency of rescue resource utilization.

[0106] In addition, the present invention makes full use of counterfactual reasoning algorithms and reinforcement learning technology to optimize the resource allocation decision-making process. In the face of the suddenness and complexity of disasters, the present invention can dynamically adjust the resource scheduling plan based on real-time data, respond to changes in the disaster area in a timely manner, and avoid the problem of resource mismatch or insufficient allocation caused by information lag in traditional methods. Through in-depth analysis of the needs of all parties in the rescue mission, an incentive mechanism is used to mobilize the enthusiasm of all participants, making the allocation of resources more efficient and timely.

[0107] The present invention also adds a dynamic risk control module, which continuously optimizes the resource allocation strategy through real-time feedback information from the disaster site, reducing the risks caused by excessive or insufficient resource allocation. Especially in the cross-regional scheduling process, considering the differences in demand in different regions and the complexity of resource supply, the present invention ensures the safety and stability of resource allocation through accurate risk assessment and dynamic adjustment. Ultimately, the present invention not only improves the utilization efficiency of rescue resources, but also reduces the response time of post-disaster rescue, ensuring that rescue work can be effectively carried out in the shortest time.

[0108] Therefore, the present invention has high real-time, accuracy and fairness. It can provide scientific decision-making support for cross-regional rescue force allocation through precise data analysis and intelligent optimization models within the first time after the disaster occurs, thereby significantly improving the efficiency and safety of rescue work. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0110] Figure 1 A flow chart of a cross-regional rescue force allocation method based on dynamic resource game proposed by the present invention;

[0111] Figure 2 This is a structural diagram of the cross-regional rescue force allocation method based on dynamic resource game proposed by the present invention. DETAILED DESCRIPTION

[0112] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0113] refer to Figure 1-2 ,The cross-regional rescue force allocation method based on dynamic resource game includes the following steps:

[0114] S1, collect disaster multi-source data, perform standardization processing and multi-dimensional feature extraction, and generate disaster scene feature sets;

[0115] S2. Based on the disaster scenario feature set, a dynamic multi-scale disaster scenario map is constructed to characterize the resource demand diffusion law and transportation path between disaster areas, and generate a preliminary disaster scenario model;

[0116] S3. Based on the disaster scenario model, define the dynamic resource game model in cross-regional rescue and generate a preliminary resource allocation model;

[0117] S4. Combining the disaster scenario feature set, disaster scenario model and preliminary resource allocation model, a multi-level counterfactual reasoning algorithm is introduced to conduct supply and demand analysis and generate an optimized resource allocation model;

[0118] S5. Use the optimized resource allocation model and supply and demand analysis results to build a resource adaptive incentive module and use the dynamic game algorithm to optimize the incentive strategy;

[0119] S6. Based on the optimized incentive strategy, combined with the dynamic response prediction network and multimodal sequence modeling algorithm, the time series changes of disaster propagation trends and resource demand are predicted to generate resource scheduling plans;

[0120] S7. Based on the resource scheduling plan, a dynamic risk control module is constructed. Combined with the feedback information from the disaster site, the resource allocation model is iteratively optimized to generate a cross-regional rescue allocation plan.

[0121] In this implementation, S3 specifically includes:

[0122] S31. Set the set of participants in the game model P = {p 1 ,p 2 ,…,p n}, where n represents the number of disaster areas, and each participant p i The corresponding resource demand vector is D i ={d i1 ,d i2 ,…,d im}, where m is the number of resource types, d ijrepresents the demand of the i-th region for the j-th type of resources;

[0123] S32, define a policy set S = {s 1 ,s 2 ,…,s n}, where s i ={s i1 ,s i2 ,…,s im} is the resource allocation strategy for the i-th region, satisfying the following constraints:

[0124]

[0125] Among them, R j is the total supply of the jth type of resource, s ij Assign to p i The amount of resources of the jth category;

[0126] S33. For each participant p i Define the payment function U i (s), used to quantify the benefits, taking into account rescue efficiency, time cost, resource consumption and distribution fairness:

[0127]

[0128] Among them, α, β, γ are weight coefficients, η ij is the response efficiency, τ i is the rescue time of the ith area, c ij is the unit cost of the jth type of resource, and m is the number of resource types;

[0129] S34. Based on the payment function U i (s), define the global optimization objective function Φ(s), which is expressed as:

[0130]

[0131] in, is the variance of the allocation ratio, n represents the number of disaster areas, and λ is the weight parameter of allocation fairness;

[0132] S35. Use a method combining reinforcement learning and game optimization to solve the objective function Φ(s):

[0133]

[0134] Among them, s * is the final optimized strategy solution, s is the current strategy, argmax s∈S The corresponding strategy solution s when solving the function Φ(s) to maximize *, ρ is the strategy learning rate of reinforcement learning, is the gradient of the global optimization objective function Φ(s) with respect to the parameter θ, The weight parameters for game optimization, is the global average return;

[0135] S36. Strategy solution based on final optimization * and resource constraints to generate a preliminary resource allocation model.

[0136] In this implementation, S4 specifically includes:

[0137] S41, based on the disaster scene feature set D = {D 1 ,D 2 ,…,D n}, the preliminary resource allocation model S, combined with the dynamic adjustment function, calculates the supply and demand deviation matrix G = [δ ij ]:

[0138]

[0139] Among them, δ ij is the supply and demand deviation of the i-th region for the j-th resource, d ij is the demand in area i, s ij is the current allocation, ∈ is the regularization parameter;

[0140] S42. Define the counterfactual loss function L cf (h,G) Quantify the supply and demand balance of different resource allocation strategies:

[0141]

[0142] Among them, n is the number of disaster areas, m is the number of resource types, λ 1 is the time smoothing regularization weight, h is the resource allocation strategy at the current iteration moment;

[0143] S43. In the high-dimensional strategy space H, set the high-dimensional projection function P(h) to constrain the search range:

[0144]

[0145] Among them, h k ′ is the candidate strategy obtained by the k-th search, K is the number of searches for strategy optimization, η 1 is the learning rate, is the gradient of the counterfactual loss function;

[0146] S44, the strategy set H'={h 1 ',h 2 ',…,h K'}, using dynamic constraint game optimization to solve the optimal strategy:

[0147]

[0148] Among them, h * is the optimized resource allocation strategy, h' is the candidate strategy in the current optimization process, Φ(h') is the joint objective function, and h ij is the number of resources of the jth type allocated to the i-th region under the current strategy h, is the global average return, U i (h') is the payoff function of the ith region, argmax h'∈H' To find the optimal resource allocation strategy h in the high-dimensional strategy space H' * , so that the objective function reaches the maximum value;

[0149] S45. Resource allocation strategy based on optimization * , calculate the supply and demand equilibrium G(h * ):

[0150]

[0151] If G(h * ) exceeds the threshold τ, then adjust the strategy:

[0152]

[0153] Among them, ρ o is the equilibrium regulation rate, δ ij (h * ) is the supply-demand deviation term, is the gradient;

[0154] S46, based on the optimized resource allocation strategy h * and supply-demand balance G(h * ), generate an optimized resource allocation model.

[0155] In this implementation manner, S5 specifically includes:

[0156] S51, based on the optimization of resource allocation model z = {z 1 ,z 2 ,…,z n} and the supply and demand analysis result G(z), define the resource incentive function Ω(z):

[0157]

[0158] Among them, μ 1 is the incentive weight coefficient, μ 2is the balance adjustment coefficient, ε is the regularization parameter, n is the number of disaster areas, m is the number of resource categories, z ij The amount of resources of the jth type allocated to the i-th disaster area, d ij is the demand of the i-th disaster area for the j-th type of resources;

[0159] S52. Based on the resource incentive function, calculate the revenue adjustment amount Θ(z) of the resource supplier:

[0160]

[0161] Among them, λ k is the dynamic excitation adjustment coefficient;

[0162] S53. Based on the dynamic game optimization goal, a strengthening incentive strategy optimization model is constructed. By using the supply and demand analysis results and the game incentive mechanism, the resource supplier's benefits, equilibrium constraints and distribution stability are comprehensively considered to optimize the objective function:

[0163]

[0164] Among them, Φ * (z) is the objective function of strengthening the incentive strategy optimization, U i (z) is the revenue function of the i-th resource supplier, ξ is the smoothing incentive weight, is the amount of resources of the jth type allocated to the i-th area after the previous round of optimization, and ln is a logarithmic function;

[0165] S54. Use variational adaptive game optimization method to solve the equilibrium solution:

[0166]

[0167] Among them, z * is the resource allocation scheme, argmax z∈H To find a strategy in the strategy space H that can make the objective function Φ * (z) maximizes the optimal strategy z * , α o is the time series smoothing weight;

[0168] S55. Calculate the adaptability of the equilibrium solution:

[0169]

[0170] If Ψ(z * )>α p , then perform balance adjustment:

[0171]

[0172] Among them, αp To adjust the threshold for balance, Ψ(z * ) is the equilibrium adaptive loss function, is the optimal resource allocation, α d is the regularization parameter, is the gradient of the equilibrium loss function;

[0173] S56, based on resource allocation scheme z * and the impact of supply and demand incentives to generate optimized incentive strategies.

[0174] In this implementation manner, S6 specifically includes:

[0175] S61. Based on resource allocation scheme z * and optimized excitation strategy, defining the input data matrix X t :

[0176]

[0177] Among them, G(z * ) is the supply and demand equilibrium function, is the disaster situation data matrix, ξ k is the disaster propagation impact factor, Φ k (t) is the value of disaster characteristic k at time t;

[0178] S62. Using dynamic response prediction network f DRP , combined with the time dependency, predict the resource demand at the future time t+Δt:

[0179]

[0180] in, is the resource demand matrix, w j is the dynamic weight of resource category j under different disaster scenarios, θ k is the deep reinforcement learning parameter, m is the number of resource types;

[0181] S63, based on multimodal sequence modeling algorithm g MSM , calculate the resource scheduling weight at future time:

[0182]

[0183] in, is the resource demand matrix, δ 1 is the weight adjustment coefficient, T ij is the resource scheduling optimization matrix, ψ ij is the dynamic impact factor, ω t+Δt is the resource scheduling weight matrix, n represents the number of disaster areas, and m is the number of resource types;

[0184] S64. Combined with resource demand forecast results and resource scheduling weight ω t+Δt , define the resource scheduling optimization goal:

[0185]

[0186] Among them, Φ * (r) is the resource scheduling optimization objective function, r ij is the resource scheduling scheme, ε 1 is the scheduling stability weight parameter, is the resource allocation scheme, ε 2 is the scheduling time smoothness weight parameter, T is the time step of resource scheduling;

[0187] S65. Using variational reinforcement learning algorithm to solve resource scheduling optimization objectives:

[0188]

[0189] Among them, r * is the resource scheduling scheme, argmax r Φ * (r) is to find the optimal resource scheduling solution r * , is the resource scheduling solution after the previous round of optimization, ε 3 is the adjustment coefficient, ln is the logarithmic function;

[0190] S66, based on resource scheduling scheme r * , and input it into the dynamic resource scheduling system.

[0191] In this implementation manner, the S7 specifically includes:

[0192] S71, based on resource scheduling scheme r * , define the dynamic risk metric for resource allocation:

[0193]

[0194] in, is the resource scheduling risk metric, α ij is the dynamic risk weight, is the allocation in the current resource scheduling scheme, d ij is the demand of disaster area i for resource j, ∈ t To prevent small positive numbers with a denominator of zero, n is the number of disaster areas and m is the number of resource types;

[0195] S72. Define the standard deviation of distribution fairness σ f, which is used to measure the balance of the current allocation:

[0196]

[0197] When σ f >τ f , adjust the allocation scheme based on the fairness standard deviation:

[0198]

[0199] in, is the average resource allocation ratio of all disaster areas, τ f Threshold parameters optimized for fairness, is the adjusted resource allocation, ρ f To optimize the step size;

[0200] S73. Combine the disaster site feedback information F(t) to adjust the risk-adjusted resource allocation plan Perform optimization and correction to generate the final optimized resource allocation plan

[0201]

[0202] Among them, O f Optimize the weight for feedback, Q 1 and Q 2 They are demand modification weight and feedback response weight respectively;

[0203] S74. Introducing risk aversion function Redistribute and optimize high-risk allocation areas:

[0204]

[0205] Among them, ξ 3 is the risk control weight, ξ 4 is the time smoothing weight, is the resource allocation after the previous round of optimization, ln is the logarithmic function, and t is the time dimension;

[0206] S75. Integrate dynamic risk control, on-site feedback optimization and risk avoidance strategies to output cross-regional rescue allocation plans.

[0207] Embodiment 1:

[0208] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a flood disaster in a certain province. In this flood disaster, the affected areas involve multiple counties and cities, the disaster is serious, the lives and safety of tens of thousands of people are threatened, and the infrastructure is severely damaged. After the disaster, the demand for resources such as medical care, daily necessities, and rescue teams has risen sharply, and rescue operations are in urgent need of coordination. However, due to the different resource demands in various disaster areas and the great difficulties in mutual allocation, the traditional resource allocation method has failed to fully play its role, resulting in insufficient resource supply in some areas and excess resources in other areas. In order to solve this problem, a cross-regional rescue force allocation method based on dynamic resource game is adopted.

[0209] In the emergency rescue of this flood disaster, first of all, the system collects multi-source data from multiple disaster areas, including disaster type, disaster area, population density, road traffic conditions, material needs, medical resources and other information, and standardizes and extracts features from these data to form a disaster scene feature set. This feature set reflects the basic situation of the disaster area and provides basic data for subsequent resource scheduling decisions.

[0210] Then, based on the disaster scenario feature set, the system constructed a dynamic multi-scale disaster scenario graph. This graph not only describes the geographical relationship between disaster areas, but also connects the resource flow between disaster areas through the resource demand diffusion law and transportation routes. Through this graph model, the system can accurately determine the resource demand of each disaster area and the optimal path for cross-regional transportation, and generate a preliminary disaster scenario model.

[0211] Next, the system constructed a dynamic resource game model to simulate the competition and cooperation relationship between disaster areas (as game participants) in resource allocation. The resource demand of each disaster area and the demand equation for rescue resources were input into the game model. In order to ensure the fairness of rescue resources, the system used a payment function to quantify the benefits of each area in resource allocation, and obtained a preliminary resource allocation model through game optimization calculation.

[0212] Then, the system analyzed the supply and demand relationship in each disaster area and optimized resource allocation by introducing a multi-level counterfactual reasoning algorithm. Through this optimization, the system can maximize the fairness and efficiency of resource allocation while taking into account the needs of all parties. At the same time, based on the optimized resource allocation model, the system built a resource adaptive incentive module, further optimized the incentive strategy, and ensured that all participants could actively respond to resource allocation decisions.

[0213] On this basis, the system combines the dynamic response prediction network with the multimodal sequence modeling algorithm to predict the spread trend of disasters and changes in resource demand, and generates a resource scheduling plan. Through this prediction, the system can determine in advance which areas will have a sharp increase in resource demand in the next few hours, and adjust the resource scheduling strategy in a timely manner to ensure that resources can be accurately and efficiently allocated to where they are most needed.

[0214] Finally, the system iteratively optimized the resource allocation model based on the feedback from the disaster site and generated the final cross-regional rescue allocation plan. This plan ensured that the resource allocation in each disaster area was balanced to the greatest extent in terms of efficiency and fairness, and responded to changes in the disaster situation in a timely manner.

[0215] Table 1 Comparison of rescue resource allocation based on traditional method and the method of the present invention

[0216]

[0217] Table 2 Comparison of resource arrival time based on traditional method and the method of the present invention

[0218]

[0219]

[0220] As can be seen from Table 1, the differences between the traditional method and the present invention in terms of resource arrival time and resource gap. Under the traditional method, the resource arrival time of disaster area A is 12 hours, the total amount of resources is 3,000, and there is a resource gap of 1,000; the resource arrival time of disaster area B is 14 hours, and the resource gap is 500; the resource arrival time of disaster area C is 16 hours, and there is a resource gap of 1,500; the resource arrival time of disaster area D is 10 hours, but the resources have met the demand. Using the method of the present invention, the resource arrival time of all disaster areas is greatly shortened, 4 hours for disaster area A, 5 hours for disaster area B, 6 hours for disaster area C, and 3 hours for disaster area D. In addition, the resource needs of all disaster areas are fully met, and the resource gap is 0. This shows that the present invention can accurately allocate rescue resources to where they are needed in a shorter time, eliminating the resource gap problem in the traditional method.

[0221] From the data in Table 2, we can see the obvious difference in time saving. In disaster area A, the resource arrival time of the traditional method is 12 hours, while the method of the present invention is 4 hours, saving 8 hours; disaster area B saves 9 hours; disaster area C saves 10 hours; disaster area D saves 7 hours. From these data, we can see that the present invention greatly improves the timeliness of resource allocation and significantly reduces the time of rescue response.

[0222] In general, the present invention effectively improves the dispatching efficiency of rescue resources, shortens the arrival time of resources, and ensures that the needs of each disaster area are fully met, avoiding the imbalance in resource allocation, through advanced algorithms such as dynamic resource game optimization, counterfactual reasoning, and reinforcement learning. Compared with the traditional static and empirical resource dispatching method, the present invention can allocate resources more scientifically and accurately, improve rescue efficiency, and alleviate the plight of the people in the disaster area to the greatest extent, reflecting significant social benefits and practical application value.

[0223] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A cross-regional rescue force allocation method based on dynamic resource game, characterized in that: The steps include: S1, collect disaster multi-source data, perform standardization processing and multi-dimensional feature extraction, and generate disaster scene feature sets; S2. Based on the disaster scenario feature set, a dynamic multi-scale disaster scenario map is constructed to characterize the resource demand diffusion law and transportation path between disaster areas, and generate a preliminary disaster scenario model; S3. Based on the disaster scenario model, define the dynamic resource game model in cross-regional rescue and generate a preliminary resource allocation model; S4. Combining the disaster scenario feature set, disaster scenario model and preliminary resource allocation model, a multi-level counterfactual reasoning algorithm is introduced to conduct supply and demand analysis and generate an optimized resource allocation model; S5. Use the optimized resource allocation model and supply and demand analysis results to build a resource adaptive incentive module and use the dynamic game algorithm to optimize the incentive strategy; S6. Based on the optimized incentive strategy, combined with the dynamic response prediction network and multimodal sequence modeling algorithm, the time series changes of disaster propagation trends and resource demand are predicted to generate resource scheduling plans; S7. Based on the resource scheduling plan, a dynamic risk control module is constructed. Combined with the feedback information from the disaster site, the resource allocation model is iteratively optimized to generate a cross-regional rescue allocation plan.

2. The cross-regional rescue force allocation method based on dynamic resource game according to claim 1 is characterized in that: The S3 specifically includes: S31. Set the set of participants in the game model P = {p1, p2, ..., p n }, where n represents the number of disaster areas, and each participant p i The corresponding resource demand vector is D i ={d i1 ,d i2 ,…,d im }, where m is the number of resource types, d ij represents the demand of the i-th region for the j-th type of resources; S32, define a strategy set S = {s1, s2, ..., s n }, where s i ={s i1 ,s i2 ,…,s im } is the resource allocation strategy for the i-th region, satisfying the following constraints: Among them, R j is the total supply of the jth type of resource, s ij Assign to p i The j-th type of resources; S33. For each participant p i Define the payment function U i (s), used to quantify the benefits, taking into account rescue efficiency, time cost, resource consumption and distribution fairness: Among them, α, β, γ are weight coefficients, η ij is the response efficiency, τ i is the rescue time of the ith area, c ij is the unit cost of the jth type of resource, and m is the number of resource types; S34. Based on the payment function U i (s), define the global optimization objective function Φ(s), which is expressed as: in, is the variance of the allocation ratio, n represents the number of disaster areas, and λ is the weight parameter of allocation fairness; S35. Use a method combining reinforcement learning and game optimization to solve the objective function Φ(s): Among them, s * is the final optimized strategy solution, s is the current strategy, argmax s∈S The corresponding strategy solution s when solving the function Φ(s) to maximize * , ρ is the strategy learning rate of reinforcement learning, is the gradient of the global optimization objective function Φ(s) with respect to the parameter θ, The weight parameters for game optimization, is the global average return; S36. Strategy solution based on final optimization * and resource constraints to generate a preliminary resource allocation model.

3. The cross-regional rescue force allocation method based on dynamic resource game according to claim 1 is characterized in that: The S4 specifically includes: S41, based on the disaster scene feature set D = {D1, D2, ..., D n }, the preliminary resource allocation model S, combined with the dynamic adjustment function, calculates the supply and demand deviation matrix G = [δ ij ]: Among them, δ ij is the supply and demand deviation of the i-th region for the j-th resource, d ij is the demand in area i, s ij is the current allocation, ∈ is the regularization parameter; S42. Define the counterfactual loss function L cf (h,G) Quantify the supply and demand balance of different resource allocation strategies: Among them, n is the number of disaster areas, m is the number of resource types, λ1 is the time smoothing regularization weight, and h is the resource allocation strategy at the current iteration time; S43. In the high-dimensional strategy space H, set the high-dimensional projection function P(h) to constrain the search range: Among them, h k ′ is the candidate strategy obtained by the k-th search, K is the number of searches for strategy optimization, η1 is the learning rate, is the gradient of the counterfactual loss function; S44, the strategy set H'={h1',h2',…,h K '}, using dynamic constraint game optimization to solve the optimal strategy: Among them, h * is the optimized resource allocation strategy, h' is the candidate strategy in the current optimization process, Φ(h') is the joint objective function, and h ij is the number of resources of the jth type allocated to the i-th region under the current strategy h, is the global average return, U i (h') is the payoff function of the ith region, argmax h'∈H' To find the optimal resource allocation strategy h in the high-dimensional strategy space H' * , so that the objective function reaches the maximum value; S45. Resource allocation strategy based on optimization * , calculate the supply and demand equilibrium G(h * ): If G(h * ) exceeds the threshold τ, then adjust the strategy: Among them, ρ o is the equilibrium regulation rate, δ ij (h * ) is the supply-demand deviation term, is the gradient; S46, based on the optimized resource allocation strategy h * and supply-demand balance G(h * ), generate an optimized resource allocation model.

4. The cross-regional rescue force allocation method based on dynamic resource game according to claim 1 is characterized in that: The S5 specifically includes: S51, based on the optimization resource allocation model z = {z1, z2, ..., z n } and the supply and demand analysis result G(z), define the resource incentive function Ω(z): Among them, μ1 is the incentive weight coefficient, μ2 is the balance adjustment coefficient, ε is the regularization parameter, n is the number of disaster areas, m is the number of resource categories, z ij The amount of resources of the jth type allocated to the i-th disaster area, d ij is the demand of the i-th disaster area for the j-th type of resources; S52. Based on the resource incentive function, calculate the revenue adjustment amount Θ(z) of the resource supplier: Among them, λ k is the dynamic excitation adjustment coefficient; S53. Based on the dynamic game optimization goal, a strengthening incentive strategy optimization model is constructed. By using the supply and demand analysis results and the game incentive mechanism, the resource supplier's benefits, equilibrium constraints and distribution stability are comprehensively considered to optimize the objective function: Among them, Φ * (z) is the objective function of strengthening the incentive strategy optimization, U i (z) is the revenue function of the i-th resource supplier, ξ is the smoothing incentive weight, is the amount of resources of the jth type allocated to the i-th area after the previous round of optimization, and ln is a logarithmic function; S54. Use variational adaptive game optimization method to solve the equilibrium solution: Among them, z * is the resource allocation scheme, argmax z∈H To find a strategy in the strategy space H that can make the objective function Φ * (z) maximizes the optimal strategy z * , α o is the time series smoothing weight; S55. Calculate the adaptability of the equilibrium solution: If Ψ(z * )>α p , then perform balance adjustment: Among them, α p To adjust the threshold for balance, Ψ(z * ) is the equilibrium adaptive loss function, is the optimal resource allocation, α d is the regularization parameter, is the gradient of the equilibrium loss function; S56, based on resource allocation scheme z * and the impact of supply and demand incentives to generate optimized incentive strategies.

5. The cross-regional rescue force allocation method based on dynamic resource game according to claim 1 is characterized in that: The S6 specifically includes: S61. Based on resource allocation scheme z * and optimized excitation strategy, defining the input data matrix X t : Among them, G(z * ) is the supply and demand equilibrium function, is the disaster situation data matrix, ξ k is the disaster propagation impact factor, Φ k (t) is the value of disaster characteristic k at time t; S62. Using dynamic response prediction network f DRP , combined with the time dependency, predict the resource demand at the future time t+Δt: in, is the resource demand matrix, w j is the dynamic weight of resource category j under different disaster scenarios, θ k is the deep reinforcement learning parameter, m is the number of resource types; S63, based on multimodal sequence modeling algorithm g MSM , calculate the resource scheduling weight at future time: in, is the resource demand matrix, δ1 is the weight adjustment coefficient, T ij is the resource scheduling optimization matrix, ψ ij is the dynamic impact factor, ω t+Δt is the resource scheduling weight matrix, n represents the number of disaster areas, and m is the number of resource types; S64. Combined with resource demand forecast results and resource scheduling weight ω t+Δt , define the resource scheduling optimization goal: Among them, Φ * (r) is the resource scheduling optimization objective function, r ij is the resource scheduling scheme, ε1 is the scheduling stability weight parameter, is the resource allocation scheme, ε2 is the scheduling time smoothness weight parameter, and T is the time step of resource scheduling; S65. Using variational reinforcement learning algorithm to solve resource scheduling optimization objectives: Among them, r * is the resource scheduling scheme, argmax r Φ * (r) is to find the optimal resource scheduling solution r * , is the resource scheduling plan after the previous round of optimization, ε3 is the adjustment coefficient, and ln is the logarithmic function; S66, based on resource scheduling scheme r * , and input it into the dynamic resource scheduling system.

6. The cross-regional rescue force allocation method based on dynamic resource game according to claim 1 is characterized in that: The S7 specifically includes: S71, based on resource scheduling scheme r * , define the dynamic risk metric for resource allocation: in, is the resource scheduling risk metric, α ij is the dynamic risk weight, is the allocation in the current resource scheduling scheme, d ij is the demand of disaster area i for resource j, ∈ t To prevent small positive numbers with a denominator of zero, n is the number of disaster areas and m is the number of resource types; S72. Define the standard deviation of distribution fairness σ f , which is used to measure the balance of the current allocation: When σ f >τ f , adjust the allocation scheme based on the fairness standard deviation: in, is the average resource allocation ratio of all disaster areas, τ f Threshold parameters optimized for fairness, is the adjusted resource allocation, ρ f To optimize the step size; S73. Combine the disaster site feedback information F(t) to adjust the risk-adjusted resource allocation plan Perform optimization and correction to generate the final optimized resource allocation plan Among them, O f is the feedback optimization weight, Q1 and Q2 are the demand modification weight and feedback response weight respectively; S74. Introducing risk aversion function Redistribute and optimize high-risk allocation areas: Among them, ξ3 is the risk control weight, ξ4 is the time smoothing weight, is the resource allocation after the previous round of optimization, ln is the logarithmic function, and t is the time dimension; S75. Integrate dynamic risk control, on-site feedback optimization and risk avoidance strategies to output cross-regional rescue allocation plans.

Images