Emergency planning optimization method and device based on action parallelization and graph conversion, equipment, medium and product

Abstract

The application discloses an emergency planning optimization method and device based on action parallelization and graph conversion, equipment, medium and product, relates to the field of artificial intelligence and robots, and the method comprises the following steps: selecting an emergency planner according to a task corresponding to a user demand, and constructing an emergency planning tree; constructing an observation action branch structure based on each observation result of the observation action; extracting an effect action and a belief state between two observation action nodes in the emergency planning tree; determining parallel effect action pairs in a linear effect action sequence based on the belief state; representing the parallel effect action pairs as a hierarchical graph based on a planning graph model; effect actions at the same level in the hierarchical graph can be executed in parallel; all hierarchical graphs and observation action branch structures are recursively combined to construct an optimized global emergency planning tree, and the task corresponding to the user demand is executed according to the global emergency planning tree, and the execution efficiency of the emergency planning is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and robots, and in particular to an emergency planning optimization method and device based on action parallelization and graph conversion, equipment, medium and product. BACKGROUND

[0002] In the field of artificial intelligence and robots, the decision-making ability of an agent is an important indicator of its intelligence level. Task planning is an important way to achieve effective decision-making of an agent. According to the current environmental state information, a series of action sequences are generated to guide the agent to change the current state to the target state and complete the task requirements.

[0003] In the real world, the uncertainty of the environment often brings many challenges to the task planning of the agent. Therefore, emergency planning, as an important branch of traditional task planning, focuses on planning problems with uncertain initial states. The emergency planner generates a tree-shaped emergency plan by searching for available effect actions and observation actions. The output emergency plan consists of effect actions and observation actions. Its execution can be regarded as traversing the entire planning tree from the root node to the leaf node, and selecting the correct branch according to the observation results obtained by the observation actions. In the forward state search process, the planner selects the action with the highest utility value in the current state and applies its action effect to convert the current state until the target state is satisfied.

[0004] However, in the forward state search process, the emergency planner only considers the action with the highest utility value in each search state and uses it as the successor action in that state. Therefore, the resulting emergency planning result only allows sequential execution, which will result in low execution efficiency of emergency planning in long-cycle and parallelizable scenarios. For complex robot tasks in the real world, this limitation becomes more prominent, further hindering the application of emergency planning methods in production practice. SUMMARY

[0005] The purpose of the present application is to provide an emergency planning method, device, equipment, medium and product based on action parallelization and graph conversion to solve the problem of low execution efficiency of emergency planning.

[0006] To achieve the above-mentioned purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides an emergency planning method based on action parallelization and graph conversion, comprising:

[0008] According to a task corresponding to a user demand, an emergency planner is selected to construct an emergency planning tree; the emergency planning tree comprises a plurality of nodes; the nodes comprise observation action nodes and effect action nodes, and edges between every two nodes on the emergency planning tree have belief states; a plurality of effect actions exist between observation actions corresponding to two observation action nodes; the observation actions are used to acquire environment state information; and the effect actions are used to change environment states;

[0009] An observation action branch structure is constructed based on each observation result of the observation actions; each observation action branch structure comprises a linear effect action sequence, and the linear effect action sequence comprises a plurality of effect actions;

[0010] An extraction module is configured to extract effect actions and belief states between two observation action nodes in the emergency planning tree;

[0011] Based on the belief states, pairs of effect actions that can be parallel in the linear effect action sequence are determined;

[0012] Based on a planning graph model, the pairs of effect actions that can be parallel are represented as hierarchical graphs; effect actions in the same level in the hierarchical graphs can be executed in parallel;

[0013] All hierarchical graphs and the observation action branch structure are recursively combined to construct an optimized global emergency planning tree, and a task corresponding to a user demand is executed according to the global emergency planning tree.

[0014] In a second aspect, the application provides an emergency planning device based on action parallelization and graph conversion, comprising:

[0015] An emergency planning tree construction module is configured to select an emergency planner to construct an emergency planning tree according to a task corresponding to a user demand; the emergency planning tree comprises a plurality of nodes; the nodes comprise observation action nodes and effect action nodes, and edges between every two nodes on the emergency planning tree have belief states; a plurality of effect actions exist between observation actions corresponding to two observation action nodes; the observation actions are used to acquire environment state information; and the effect actions are used to change environment states;

[0016] An observation action branch structure construction module is configured to construct observation action branch structures based on each observation result of the observation actions; each observation action branch structure comprises a linear effect action sequence, and the linear effect action sequence comprises a plurality of effect actions;

[0017] An extraction module is configured to extract effect actions and belief states between two observation action nodes in the emergency planning tree;

[0018] a parallelizable effect action pair determination module configured to determine, based on the belief state, a parallelizable effect action pair in the linear effect action sequence;

[0019] a hierarchical graph construction module configured to represent the parallelizable effect action pair as a hierarchical graph based on a planning graph model, wherein effect actions at a same level in the hierarchical graph can be executed in parallel;

[0020] an optimized global emergency planning tree construction module configured to combine all the hierarchical graphs and the observation action branch structure recursively to construct an optimized global emergency planning tree, and perform a task corresponding to the user demand according to the global emergency planning tree.

[0021] In a third aspect, the present application provides a computer device, comprising a memory, a processor, a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the emergency planning method based on action parallelization and graph conversion according to any one of the above.

[0022] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the emergency planning method based on action parallelization and graph conversion according to any one of the above.

[0023] In a fifth aspect, the present application provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the emergency planning method based on action parallelization and graph conversion according to any one of the above.

[0024] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0025] First, according to the task corresponding to the user demand, an emergency planning tree is constructed, an observation action branch structure is constructed, effect actions and belief states between two observation action nodes in the emergency planning tree are extracted, hidden parallel relationships between actions in the emergency planning tree are identified, and a parallelizable effect action pair in a linear effect action sequence is obtained; second, the parallelizable effect action pair obtained above is converted into a hierarchical graph based on a planning graph model, wherein effect actions at a same level in the hierarchical graph can be executed in parallel. Unlike the actions in the original planning tree which can only be sequentially ordered and executed in the conventional technical solution, the present application determines a parallelizable effect action pair and converts it into a hierarchical graph, realizes the parallel execution of effect actions at the same level, and improves the execution efficiency of the emergency planning.

[0026] In addition, the application converts the parallelizable effect actions into a single hierarchical graph based on a planning graph model by identifying the parallelizable effect actions, and then recursively merges the hierarchical graph into an emergency planning tree step by step, so that the final result can support parallel execution of the effect actions and handle uncertain observation results through observation actions. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0028] Figure 1 A flowchart of an emergency planning method based on action parallelization and graph conversion in an embodiment of the present application;

[0029] Figure 2 An architecture diagram of an emergency planning method based on action parallelization and graph conversion provided in an embodiment of the present application;

[0030] Figure 3 A schematic diagram of a planning result of a traditional emergency planner;

[0031] Figure 4 A schematic diagram of a planning result of an optimized global emergency planning tree constructed by the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all 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.

[0033] The above purposes, features and advantages of the present application can be more obvious and easy to understand. The present application will be further described in detail below with reference to the drawings and specific embodiments.

[0034] An emergency planning method based on action parallelization and graph conversion is provided in the embodiments of the present application. The method is executed by a computer device, which can be executed by a terminal or a server alone or by a terminal and a server together. In the embodiments of the present application, as shown in the drawings, the method includes the following steps. Figure 1

[0035] ​S1: selecting an emergency planner according to a task corresponding to a user demand, and constructing an emergency planning tree; the emergency planning tree comprises a plurality of nodes; the nodes comprise observation action nodes and effect action nodes, and edges between every two nodes on the emergency planning tree have belief states; a plurality of effect actions exist between observation actions corresponding to two observation action nodes; the observation actions are used to acquire environment state information; and the effect actions are used to change environment states.

[0036] S2: constructing observation action branch structures based on branching of each observation result of the observation actions; each observation action branch structure comprises a linear effect action sequence, and the linear effect action sequence comprises a plurality of effect actions.

[0037] S3: extracting effect actions and belief states between two observation action nodes in the emergency planning tree.

[0038] S4: determining pairs of effect actions that can be parallel in the linear effect action sequence based on the belief states.

[0039] S5: representing the pairs of effect actions that can be parallel as hierarchical graphs based on a planning graph model; effect actions in the same level in the hierarchical graphs can be executed in parallel.

[0040] S6: recursively merging all hierarchical graphs and the observation action branch structures to construct an optimized global emergency planning tree, and executing a task corresponding to a user demand according to the global emergency planning tree.

[0041] In an exemplary embodiment, S4 specifically comprises: filtering and identifying pairs of effect actions that can be parallel in the linear effect action sequence according to a screening rule based on the belief states; the screening rule comprises that the actions are all applicable in the same belief state; execution of one action cannot invalidate the remaining actions; the actions cannot cause inconsistency of a new belief state generated; observation actions are processed separately; and the actions are observation actions or effect actions.

[0042] Further, in the emergency planning, a planning tree obtained by an emergency planner comprises two types of actions: effect actions that change environment states and observation actions that acquire environment state information, and branching is performed according to each possible observation result of the observation actions. In each branch, a sequence of actions has a large number of potential parallel actions. In order to identify such parallelizable actions, the present application lays a foundation for subsequent parallelization based on a set of rules. The rules are as follows:

[0043] (1) The actions should all be applicable in the same belief state. A prerequisite for parallelizing the actions is that their prerequisites are satisfied in the same belief.

[0044] (2) The execution of one action should not invalidate other actions. Action applicability does not necessarily mean that they can be executed in parallel. During execution, the effects of one action can conflict with the preconditions of other actions, thus invalidating the other actions.

[0045] (3) Actions should not result in inconsistent new beliefs. If the applicable actions in the same belief result in inconsistent effects, the resulting new belief will be uncertain, thus causing conflicts in the subsequent optimization.

[0046] (4) Observation actions should be handled separately. For a world predicate that is initially unknown, applying an observation action a helps the robot observe the true value p or of the world predicate, thus resulting in two subsequent beliefs b ′+ and b ′- , both of which are valid but inconsistent with each other.

[0047] In an exemplary embodiment, S5 specifically comprises: the planning graph model comprises two kinds of nodes and edges; the two kinds of nodes respectively occupy a proposition layer and an action layer, and the proposition layer and the action layer are alternated with each other; the proposition nodes in the proposition layer are instantiated propositions; the action nodes in the action layer correspond to effect actions or maintenance actions; the maintenance action is used to convert a proposition node from a previous level to a next level, and does not change the belief state; the edges before the action nodes represent precondition connections, and the edges after the action nodes represent effect connections; the precondition and the effect of the maintenance action are the same; the level is the proposition layer; taking the initial belief, the target belief and the linear effect action sequence between the initial belief and the target belief in the emergency planning tree as inputs of the planning graph model, calling a local level graph conversion algorithm, converting the parallel relationship between the effect actions into a level relationship through an expansion manner until all propositions in the target belief are in the last proposition layer in the planning graph model, determining a level graph; wherein the initial belief is a belief state after a previous observation action node; the target state is a belief state before a subsequent observation action node; the first level of the planning graph model comprises all propositions in the initial belief.

[0048] In an exemplary embodiment, during the expansion process, if the preconditions of any effect action in the linear effect action sequence are satisfied in the previous proposition layer, the effect action is inserted into the next action layer, and the effect of the effect action is added to the next proposition layer.

[0049] In an exemplary embodiment, during the expansion process, a mutually exclusive relationship between nodes in the planning graph model is constructed; three types of mutually exclusive constraints are included between two actions on the same level; the mutually exclusive constraints include demand competition constraints, inconsistent effect constraints, and execution interference constraints; the demand competition constraint is that the two actions have mutually exclusive prerequisites in the previous proposition layer; the inconsistent effect constraint is that the effect of one action is the negation of the effect of another action; the execution interference constraint is that the effect of one action negates the prerequisite of another action; if the inconsistent support condition is met, the two propositions at the same level are mutually exclusive; the inconsistent support condition is that the two propositions negate each other, or, all actions that implement the proposition in the previous action layer are combined into pairs and are mutually exclusive.

[0050] Furthermore, the parallel relationship between actions can be represented by a graph structure. This application uses graph-based transformation technology to convert the linear effect action sequence in the original planning tree into a set of hierarchical graph structures, which makes it possible to explicitly represent the parallel relationship between non-conflicting, parallelizable actions in the original planning results and provides a basis for the subsequent parallel execution of these actions.

[0051] In this application, the hierarchical graph can be represented as a tuple γ =<N,L> , where N is a set of action nodes and L is a function that maps each node in γ to a level in a set of levels.

[0052] For two nodes n1 and n2 in N, L(n1)=L(n2) means that operations n1(a) and n2(a) can be executed in parallel.

[0053] This hierarchical graph transformation utilizes a planning graph model. In this model, there are two types of nodes and edges: the two nodes occupy the proposition layer Ip and the action layer Ia, respectively, alternating between them. Nodes in the proposition layer consist of instantiated propositions. In the action layer, nodes represent effect actions or maintenance actions. Maintenance actions have the same prerequisites and effects and are designed to transition proposition nodes from the previous level to the next without changing any state. Edges before action nodes in the planning graph represent prerequisite connections, while edges after them represent effect connections.

[0054] The hierarchical graph transformation process is described as follows: It takes as input the initial beliefs, target beliefs, and a set of effective actions extracted from the contingency planning tree. It then searches for solutions by expanding and extracting the planning graph. The first level of the planning graph consists of the propositions in the initial beliefs. Starting from this first level, the planning graph is continuously expanded until all target belief propositions appear in the final level.

[0055] During expansion, if any action in the input effect-action set has its preconditions satisfied in the previous proposition level, they are inserted into the next action level. Then, the effects of these actions are added to the next new proposition level.

[0056] Based on the above planning graph structure, parallelism between actions can now be characterized by the hierarchical relationship. However, during expansion, due to the relaxation of the search space, there can be mutual exclusions between actions and propositions. It is crucial to avoid potential conflicts in the hierarchical graph due to mutual exclusions. Therefore, while expanding the planning graph, the mutual exclusion relationship between nodes also needs to be constructed. Generally, there are three types of mutual exclusion constraints between two actions in the same level, due to the following reasons:

[0057] (1) demand competition constraint: two actions have mutually exclusive preconditions in the previous proposition level.

[0058] (2) inconsistent effect constraint: the effect of one action is the negation of the effect of the other action.

[0059] (3) execution interference constraint: the effect of one action negates the precondition of the other action.

[0060] In addition, two propositions in the same level will also be mutually exclusive when the following inconsistent support condition is met:

[0061] Inconsistent support condition: two propositions negate each other, or all combinations that achieve the proposition in the previous action level are pairwise mutually exclusive.

[0062] The above expansion process is repeated until all propositions in the goal belief are satisfied in the last proposition level.

[0063] After ensuring that all pairs of target propositions are not mutually exclusive, the extraction of the solution is immediately started from the back to the front.

[0064] The extraction process of the solution is as follows: for each target proposition in the last proposition level, search the previous action level to find the action whose effect appears. If any action achieves the goal, it will be marked. Then store the preconditions of the marked action as new goals in the previous proposition level. When extracting an action level, if all the marked actions in this level are pairwise mutually exclusive, the extraction process fails, and the planning graph will be expanded by one level. Then the extraction of the solution will be retried from the last level again. Once the initial proposition level is extracted, the marked actions in each level will be returned as a feasible solution. The returned hierarchical graph consists of several levels of effect actions, where actions in the same level can be executed in parallel without any conflict.

[0065] In an exemplary embodiment, S6 specifically comprises: based on the task corresponding to the user demand, traversing from the root node of the emergency planning tree to a next node, and recording an intermediate effect action set as a linear effect action sequence; the next node is an observation action node or a leaf node; taking the initial belief, a target belief related to the next node traversed, the linear effect action sequence, and the next node as inputs, calling a local hierarchical graph conversion algorithm to determine an initial hierarchical graph, constructing an optimized global emergency planning tree; the initial hierarchical graph currently only contains effect actions; if the next node traversed is a leaf node, returning to the last hierarchical graph; if the next node traversed is an observation action node, taking an observation action as a new level in the hierarchical graph, appending it to the original hierarchical graph, and processing two observation action branch structures corresponding to the observation action node respectively; traversing from the initial hierarchical layer to the last hierarchical graph in turn, and executing the actions in the nodes in the hierarchical graph level by level.

[0066] Further, based on the previously converted hierarchical graph, the parallel relationship between the effect actions in each branch can be intuitively represented. Then, the entire emergency planning tree is optimized and executed by recursively merging these hierarchical graphs with observation action branch structures, so as to realize plan parallelization and solve the uncertainty in the overall planning scale.

[0067] The application formalizes the optimized overall emergency planning as Π=<Γ,E>, where Γ is a set of hierarchical graph nodes, and E is a set of edges representing observation connections between nodes. Each node γ∈Γ is composed of a converted hierarchical graph.

[0068] The specific process of optimizing the emergency planning is as follows:

[0069] (1) First, an initial emergency planning tree τ is obtained by using a selected emergency planner to solve.

[0070] (2) From the root node of τ, traverse to a next observation action node or leaf node n G . Record the intermediate effect action set in A * .

[0071] (3) Recursively construct an optimized tree. Take the initial belief b I , the target belief related to n G , the extracted A * , and n G as inputs, call a local hierarchical graph conversion algorithm to obtain an initial hierarchical graph, which currently only contains effect actions.

[0072] (4) If the node n G traversed is a leaf node, return to the hierarchical graph; if n GObserve the action node, add the observation action as a new level in the hierarchy graph to the original hierarchy graph, and process the two subsequent branches of the node separately:

[0073] (4.1) Process the branch where the observation result of the observation action n(a) is true. First, in the belief b G Apply n(a) to get the new belief b + Then continue from the next node n.next in this branch + Start traversing until the next observation action node or leaf node n + G , record the intermediate effect action A in this branch *+ Then repeat steps (2)-(4) to recursively construct the local hierarchical graph in the branch and connect the hierarchical graph to the current hierarchical graph node γ.

[0074] (4.2) Processing the branch where the observation result of the observation action n(a) is false, the process is similar to that of (4.1).

[0075] The execution of the optimized plan can be viewed as a sequential traversal of the global hierarchical graph from the first to the last level, executing the actions in each node of the hierarchy graph level by level. Effect actions at the same level of the hierarchy graph can be executed in parallel. When an observation action is executed in the graph, execution continues with the next graph node based on the actual observed value, continuing to the end of the tree.

[0076] like Figure 2 As shown, this application selects a suitable emergency planner to solve and obtain the initial emergency planning tree based on user needs. Based on the above planning tree, a series of effect actions between two observation action nodes and the relevant belief states in the planning tree are first extracted. Then, the extracted effect action and other information are filtered and identified through a number of rules to obtain all hidden parallel effect action pairs. Then, based on the planning graph model, the linear effect action set is converted into a set of hierarchical graphs, and the effect actions at the same level in the hierarchical graph can be executed in parallel. After completing the above-mentioned local parallelization conversion, the hierarchical graph is recursively merged with the observation action branch structure to obtain the optimized global emergency planning tree, thereby achieving parallel optimization of the overall planning, and at the same time having the ability to process uncertain observation results through observation actions.

[0077] Take the scenario of a service robot delivering coffee as an example. Figure 3 As shown in the figure, in the initial state, there is a robot at position PS and a cup of coffee at position P1 or P2, which needs to be confirmed by observation. The actions that the robot can perform are:

[0078] (1) Effect action: 1. GoTo(a,b) - run from position a to position b; 2. PickUp(p) - pick up the coffee at position p; 3. Stir() - stir the coffee in your hand; 4. Place(p) - put the coffee down at position p.

[0079] (2) Observation action: SenseCoffee(p) - observe whether the coffee is at position p.

[0080] The robot's task is to plan a series of actions starting from the PS, determine the location of the coffee by executing the actions, pick up the coffee and deliver it to the guest's position PG. The coffee needs to be stirred before being handed to the guest.

[0081] The results directly obtained by using traditional emergency planners are as follows Figure 3 As shown in the figure, the robot starts from PS and moves to P2, where it checks to see if there is coffee. If there is, it picks up the coffee, moves to PG, stirs the coffee, and finally places the coffee in PG. If there is no coffee, it means the coffee is at P1. The robot moves from P2 to P1, picks up the coffee at P2, moves to PG, stirs the coffee, and finally places the coffee in PG.

[0082] The results of parallel optimization of the above plan using this application are as follows: First, starting from the root node, traverse the entire emergency plan, encounter the observation action node 2, extract the effect action between the initial state and the observation action node, there is only one effect action 1, so it cannot be further parallelized; continue to traverse from the observation action node 2, the node generates two branches, corresponding to two belief states St and Sf respectively: start traversing from the branch with the true result until the leaf node 6, the belief state after 6 is set to Stg, extract the effect actions 3, 4, 5, 6 between 2 and 6 and put them into the set A*={PickUp(P2), Stir(), GoTo(P2,PG), Place(PG)}. Using the extracted St, Stg, A* as input, the planning graph model and conversion algorithm in this application are used to parallelize the actions in this branch. After this optimization, the actions Stir() and GoTo(P2,PG) are parallelized, and both can be executed at the same time, which reduces the time taken for the entire task to a certain extent. Similarly, the same extraction and parallelization steps are applied to the other branch of observation action 2, and the final result is as follows Figure 4 shown.

[0083] Based on the same inventive concept, the application further provides an action parallelization and graph conversion based emergency planning device for implementing the above-mentioned action parallelization and graph conversion based emergency planning method. The device provides a similar solution to the implementation scheme described in the above-mentioned method, and therefore the specific limitations in one or more embodiments of the action parallelization and graph conversion based emergency planning device provided below can refer to the limitations of the action parallelization and graph conversion based emergency planning method described above, which will not be repeated here.

[0084] In one exemplary embodiment, an action parallelization and graph conversion based emergency planning device is provided, comprising:

[0085] An emergency planning tree construction module is configured to select an emergency planner and construct an emergency planning tree according to a task corresponding to a user demand; the emergency planning tree comprises a plurality of nodes; the nodes comprise observation action nodes and effect action nodes, and edges between every two nodes on the emergency planning tree have belief states; a plurality of effect actions exist between observation actions corresponding to two observation action nodes; the observation actions are configured to obtain environmental state information; and the effect actions are configured to change the environmental state.

[0086] An observation action branch structure construction module is configured to construct observation action branch structures based on branching of each observation result of the observation actions; each observation action branch structure comprises a linear effect action sequence, and the linear effect action sequence comprises a plurality of effect actions.

[0087] An extraction module is configured to extract effect actions and belief states between two observation action nodes in the emergency planning tree.

[0088] A parallelizable effect action pair determination module is configured to determine parallelizable effect action pairs in the linear effect action sequence based on the belief states.

[0089] A hierarchical graph construction module is configured to represent the parallelizable effect action pairs as hierarchical graphs based on a planning graph model; effect actions at the same level in the hierarchical graphs can be executed in parallel.

[0090] An optimized global emergency planning tree construction module is configured to recursively combine all hierarchical graphs and the observation action branch structures to construct an optimized global emergency planning tree, and execute a task corresponding to the user demand according to the global emergency planning tree.

[0091] The application has the following advantages:

[0092] 1. The execution efficiency of the planning result is greatly improved. The planning tree generated by the existing emergency planning method is executed in a way of traversing from the root node to the leaf node one by one. When the tree branch contains actions that can be executed in parallel, the method cannot well support the occurrence of such behavior. The present application adopts a new hierarchical graph structure to represent the original planning result, which contains more information, and the parallel relationship between actions is represented by the hierarchical relationship. The original process of traversing a single action node is now changed to traversing multiple nodes in one layer at a time, and the nodes traversed at the same time can be executed in parallel. This execution strategy greatly reduces the total time required to complete the entire planning, which is of great significance in practical applications.

[0093] 2. Highly compatible with existing emergency planners. The existing emergency planners focus on improving the speed of single solving, and for different problem domains, the solving efficiency is very different due to the different heuristic functions adopted. The present application integrates the above planners in a modular way, and makes secondary optimization based on the results of the above planners, so it has good compatibility and can cope with various emergency planning scenarios.

[0094] In an exemplary embodiment, a computer device, which can be a server or a terminal, is provided. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store emergency planning data based on action parallelization and graph conversion. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement an emergency planning method based on action parallelization and graph conversion.

[0095] In an exemplary embodiment, a computer device is provided, including a memory and a processor, and the memory stores a computer program. The processor executes the computer program to implement the above method.

[0096] In an exemplary embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the above method.

[0097] In one example embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the above method.

[0098] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to a memory, database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile and volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0099] In the present application, all actions of obtaining signals, information or data are performed in compliance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization of the owner of the corresponding device.

[0100] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0101] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, it should be understood that the application encompasses all possible combinations of the technical features unless such a combination is not technically possible.

[0102] The principles and implementation manners of the present application are described herein by using specific examples, and the above embodiments are only used to help understand the method of the present application and its core idea; meanwhile, according to the idea of the present application, the specific implementation manners and application scopes will be changed by those skilled in the art. In conclusion, the content of the present specification should not be understood as a limitation of the present application.

Claims

1. A contingency planning method based on action parallelization and graph transformation, characterized in that: The emergency planning method based on action parallelization and graph transformation includes: According to the task corresponding to the user's needs, a contingency planner is selected to construct a contingency planning tree; the contingency planning tree includes multiple nodes; the nodes include observation action nodes and effect action nodes, and the edge between each two nodes in the contingency planning tree has a belief state; there are multiple effect actions between the observation actions corresponding to two observation action nodes; the observation actions are used to obtain environmental state information; and the effect actions are used to change the environmental state; Based on each observation result of the observation action, a branch structure of observation action is constructed; each of the observation action branch structures includes a linear effect action sequence, and the linear effect action sequence includes multiple effect actions; Extracting the effect action and belief state between two observation action nodes in the emergency planning tree; Based on the belief state, determining parallelizable effect-action pairs in the linear effect-action sequence; Based on the planning graph model, the parallelizable effect-action pairs are represented as a hierarchical graph, specifically including: The planning graph model includes two types of nodes and edges; the two types of nodes occupy the proposition layer and the action layer respectively, and the proposition layer and the action layer alternate with each other; the proposition nodes in the proposition layer are instantiated propositions; the actions corresponding to the action nodes in the action layer are effect actions or maintenance actions; the maintenance actions are used to convert proposition nodes from the previous level to the next level without changing the belief state; the edges before the action nodes represent prerequisite connections, and the edges after the action nodes represent effect connections; the prerequisites and effects of the maintenance actions are the same; the level is the proposition layer; The initial belief, target belief, and linear effect-action sequence between the initial belief and the target belief in the emergency planning tree are used as inputs of the planning graph model, and a local hierarchical graph conversion algorithm is called. By extension, all propositions in the target belief are in the last proposition layer in the planning graph model, and the parallel relationship between the effect-actions is converted into a hierarchical relationship to determine a hierarchical graph; wherein the initial belief is the belief state after the previous observation action node; the target state is the belief state before the next observation action node; the first level of the planning graph model includes all propositions in the initial belief; and the effect-actions at the same level in the hierarchical graph can be executed in parallel; All hierarchical graphs and the observation action branch structures are recursively merged to construct an optimized global emergency planning tree, and tasks corresponding to user needs are executed according to the global emergency planning tree.

2. The emergency planning method based on action parallelization and graph transformation according to claim 1, characterized in that: Based on the belief state, determining effect-action pairs that can be run in parallel in the linear effect-action sequence specifically includes: Based on the belief state, the parallel effect-action pairs in the linear effect-action sequence are filtered and identified according to screening rules; the screening rules include that the actions are applicable in the same belief state; the execution of one action cannot invalidate the remaining actions; the action cannot cause inconsistency in the new belief state generated; observation actions are processed separately; wherein, the action is an observation action or an effect action.

3. The emergency planning method based on action parallelization and graph transformation according to claim 1, characterized in that: During the expansion process, if the prerequisite of any effect action in the linear effect action sequence is satisfied in the previous proposition layer, the effect action is inserted into the next action layer, and the effect of the effect action is added to the next proposition layer.

4. The emergency planning method based on action parallelization and graph transformation according to claim 1, characterized in that: During the expansion process, mutually exclusive relationships between nodes in the planning graph model are constructed; three types of mutually exclusive constraints are included between two actions at the same level; the mutually exclusive constraints include demand competition constraints, inconsistent effect constraints, and execution interference constraints; the demand competition constraint means that the two actions have mutually exclusive prerequisites in the previous proposition layer; the inconsistent effect constraint means that the effect of one action is the negation of the effect of another action; and the execution interference constraint means that the effect of one action negates the prerequisite of another action. If the inconsistent support condition is met, two propositions at the same level are mutually exclusive; the inconsistent support condition is that the two propositions negate each other, or all actions that realize the proposition in the previous action layer are combined into pairs that are mutually exclusive.

5. The emergency planning method based on action parallelization and graph transformation according to claim 1, characterized in that: All hierarchical graphs and the observation action branch structures are recursively merged to construct an optimized global emergency planning tree, and tasks corresponding to user needs are executed according to the global emergency planning tree, specifically including: Based on the task corresponding to the user demand, traverse from the root node of the emergency planning tree to the next node, and record the intermediate effect action set as a linear effect action sequence; the next node is an observation action node or a leaf node; Taking the initial belief, the target belief related to the next node traversed, the linear effect action sequence, and the next node as input, calling the local hierarchical graph conversion algorithm, determining the initial hierarchical graph, and constructing an optimized global emergency planning tree; the initial hierarchical graph currently only contains effect actions; If the next node traversed is a leaf node, return to the previous level graph; If the next node traversed is an observation action node, the observation action is added to the original hierarchical graph as a new level in the hierarchical graph, and the two observation action branch structures corresponding to the observation action node are processed separately; The hierarchical graph is traversed from the initial hierarchical layer to the last hierarchical layer, and actions in the nodes in the hierarchical graph are executed level by level.

6. An emergency planning device based on action parallelization and graph transformation, characterized in that: The emergency planning device based on action parallelization and graph conversion adopts the emergency planning method based on action parallelization and graph conversion according to any one of claims 1 to 5, and the emergency planning device based on action parallelization and graph conversion includes: The contingency planning tree construction module is used to select a contingency planner and construct a contingency planning tree based on the task corresponding to the user's needs; the contingency planning tree includes multiple nodes; the nodes include observation action nodes and effect action nodes, and the edge between each two nodes in the contingency planning tree has a belief state; there are multiple effect actions between the observation actions corresponding to two observation action nodes; the observation actions are used to obtain environmental state information; and the effect actions are used to change the environmental state; An observation action branch structure construction module is used to construct an observation action branch structure based on each observation result of the observation action; each observation action branch structure includes a linear effect action sequence, and the linear effect action sequence includes multiple effect actions; An extraction module, configured to extract effect actions and belief states between two observation action nodes in the emergency planning tree; a parallelizable effect-action pair determination module, configured to determine, based on the belief state, parallelizable effect-action pairs in the linear effect-action sequence; A hierarchical graph construction module is used to represent the parallelizable effect-action pairs as a hierarchical graph based on a planning graph model; the same-level effect-actions in the hierarchical graph can be executed in parallel; The optimized global emergency planning tree construction module is used to recursively merge all hierarchical graphs and the observation action branch structure to construct an optimized global emergency planning tree, and execute tasks corresponding to user needs according to the global emergency planning tree.

7. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the emergency planning method based on action parallelization and graph transformation according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the emergency planning method based on action parallelization and graph transformation according to any one of claims 1 to 5 is implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the emergency planning method based on action parallelization and graph transformation according to any one of claims 1 to 5 is implemented.

Images