A method and system for inspection path decision-making based on quantum implicit map states

By using the quantum implicit map state method, a unified representation of the semantics of inspection tasks and the environmental scene is achieved in a complex open environment. This solves the problems of high computational complexity and insufficient real-time decision-making in existing technologies, and improves the real-time performance and security of path decision-making.

CN122590909BActive Publication Date: 2026-11-10FUJIAN YIXINHAI INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611082674.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-21
Publication Date
2026-11-10
Estimated Expiration
2046-07-21

AI Technical Summary

Technical Problem

Existing technologies struggle to achieve unified representation and collaborative reasoning of inspection task semantics and environmental scenarios in complex and open environments, resulting in high computational complexity, insufficient real-time decision-making, and inadequate security.

Method used

The quantum implicit map state method is adopted to generate quantum states by quantum encoding environmental point clouds and task text data. Quantum fusion and entanglement enhancement techniques are used to generate quantum fusion operators to achieve efficient alignment of task semantics with map position and electrical safety constraints.

Benefits of technology

It improves the real-time performance, globality, and security of path decision-making in complex environments, reduces computational complexity, and ensures absolute avoidance of paths and electrical restricted areas.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122590909B_ABST
    Figure CN122590909B_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for inspection path decision-making based on quantum implicit map states. The method generates environmental quantum states and task quantum states by quantum encoding environmental point cloud data and task text data. It then constructs quantum safety operators based on electrical exclusion zone data, mapping geometric, semantic, and safety information uniformly into the quantum state space. The current environmental quantum state is entangled and fused with historical map states to achieve multi-moment information accumulation, generating a dynamic map state that does not require explicit storage. Entanglement between the map state and the task quantum state is enhanced, establishing a quantum correlation between semantics and location for efficient alignment. Evolution is based on the observed quantum state, utilizing superposition states for parallel path evaluation to improve efficiency, and implementing hard constraints through quantum safety operators to ensure exclusion zone avoidance. This method unifies perception, understanding, planning, and safety within a quantum state evolution framework, improving the real-time performance and safety of path decision-making in complex environments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of quantum information processing technology, and in particular to a method and system for inspection path decision-making based on quantum implicit map states. Background Technology

[0002] With the development of artificial intelligence technology, the demand for autonomous operation of intelligent agents in complex and open environments such as inspection, maintenance, and patrol is constantly increasing. In order to enable intelligent agents to autonomously execute inspection tasks, it is usually necessary to build a scenario model based on environmental perception data and combine it with task objectives to complete path planning and decision control.

[0003] In existing technologies, the autonomous inspection process of intelligent agents usually involves first acquiring environmental information through LiDAR, visual sensors, or multi-source sensors to construct environmental representation models such as grid maps, point cloud maps, or topology maps; then generating inspection paths using path search algorithms, reinforcement learning algorithms, or navigation control algorithms, and driving the intelligent agent to complete the inspection task.

[0004] However, in complex and open environments, environmental scenes typically exhibit characteristics such as complex spatial structures, frequent dynamic changes, incomplete observation information, and diverse task objectives. Existing map construction methods primarily focus on the expression and storage of environmental geometric information, making it difficult to simultaneously describe the semantic information contained in the inspection task. This results in a lack of a unified representation mechanism between the task objective and the environmental scene. When executing inspection tasks, the system usually needs to process task semantic information and environmental scene information separately, and then achieve information association and path generation through multi-stage modules. This not only increases computational complexity but also easily leads to errors in the association between semantic and spatial information.

[0005] Furthermore, most existing inspection path methods rely on explicit map representations and classical reasoning mechanisms, and their decision-making processes are typically based on discrete environmental states and local path evaluations. As the scale of the environment and the complexity of the scene increase, the system needs to process a large amount of state information and path combination relationships, which can easily lead to a rapid increase in the dimensionality of the state space, thus affecting the efficiency of task understanding and the real-time performance of decision-making. Moreover, traditional methods generally adopt an independent processing architecture for task understanding modules, scene mapping modules, and path planning modules, lacking the ability to conduct collaborative reasoning between task semantics and environmental scenes, making it difficult to achieve global decision optimization under complex inspection task conditions. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for inspection path decision-making based on quantum implicit map states, which can realize the unified representation and collaborative reasoning of inspection task semantics and inspection environment scene in complex open environment, thereby improving the path decision-making ability of intelligent agents in autonomous inspection process.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A method for inspection path decision-making based on quantum implicit map states includes:

[0009] Acquire environmental point cloud data, task text data, and electrical exclusion zone data from wind and solar power plants;

[0010] Quantum encoding is performed on the environmental point cloud data and the task text data respectively to generate environmental quantum states and task quantum states, and quantum safety operators are constructed based on the electrical exclusion zone data;

[0011] The current quantum state of the environment is quantumly entangled and fused with the historical quantum implicit map state of the previous moment to generate the current quantum implicit map state.

[0012] The quantum implicit map state and the task quantum state are quantum entangled to enhance quantum entanglement, generating a quantum fusion operator to characterize the matching degree between task semantics and map position.

[0013] An observation quantum state is constructed based on the quantum fusion operator and the quantum implicit map state, and the observation quantum state is subjected to quantum evolution to generate candidate inspection paths;

[0014] Electrical safety constraints are applied to the candidate inspection paths based on the quantum safety operator to generate safe inspection paths.

[0015] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows:

[0016] A patrol path decision system based on quantum implicit map states includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the various steps of the patrol path decision method based on quantum implicit map states described above.

[0017] The beneficial effects of this invention are as follows: By quantum encoding environmental point cloud data and task text data respectively, environmental quantum states and task quantum states are generated. Simultaneously, quantum safety operators are constructed based on electrical exclusion zone data, uniformly mapping environmental geometric information, task semantic information, and safety constraint information to the quantum state space. By fusing the current environmental quantum state with historical quantum implicit map states through quantum entanglement, the non-local correlation characteristics of quantum entanglement are utilized to achieve coherent accumulation of information across multiple time points, generating a dynamic map state that does not require explicit storage of point clouds and is naturally adapted to large-scale sparse scenes. By enhancing the quantum entanglement between the map state and the task quantum state, a quantum correlation is established directly between task semantics and map spatial location, achieving efficient alignment between task and space. Quantum evolution is performed based on the fused observation quantum state, utilizing quantum superposition states to evaluate multiple paths in parallel, improving decision-making efficiency. Finally, physical hard constraints are implemented through quantum safety operators to ensure that the path absolutely avoids electrical exclusion zones. The entire process unifies perception, understanding, planning, and safety within a quantum state evolution framework, improving the real-time performance, globality, and safety of path decision-making in complex wind and solar power environments. Attached Figure Description

[0018] Figure 1 A flowchart of an inspection path decision method based on quantum implicit map states provided for embodiments of the present invention;

[0019] Figure 2 A flowchart illustrating a method for generating inspection interactive text conforming to power dispatch specifications, provided as an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of the architecture of an inspection path decision system based on quantum implicit map states is provided for an embodiment of the present invention;

[0021] Label Explanation:

[0022] 100. A patrol path decision system based on quantum hidden map states; 101. Memory; 102. Processor. Detailed Implementation

[0023] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0024] This invention provides a method for inspection path decision-making based on quantum implicit map states, comprising:

[0025] Acquire environmental point cloud data, task text data, and electrical exclusion zone data from wind and solar power plants;

[0026] Quantum encoding is performed on the environmental point cloud data and the task text data respectively to generate environmental quantum states and task quantum states, and quantum safety operators are constructed based on the electrical exclusion zone data;

[0027] The current quantum state of the environment is quantumly entangled and fused with the historical quantum implicit map state of the previous moment to generate the current quantum implicit map state.

[0028] The quantum implicit map state and the task quantum state are quantum entangled to enhance quantum entanglement, generating a quantum fusion operator to characterize the matching degree between task semantics and map position.

[0029] An observation quantum state is constructed based on the quantum fusion operator and the quantum implicit map state, and the observation quantum state is subjected to quantum evolution to generate candidate inspection paths;

[0030] Electrical safety constraints are applied to the candidate inspection paths based on the quantum safety operator to generate safe inspection paths.

[0031] Among them, the quantum security operator is a linear combination Hermitian operator based on the Pauli-Z operator, and its weight coefficients are determined by mapping the binary indicator function of the electrical exclusion zone data on the spatial voxel to the corresponding qubit through hash index; at the initial moment, the first frame of environmental quantum state or zero amplitude superposition state is used as the initial value of the historical quantum implicit map state; quantum entanglement fusion includes establishing inter-scale correlation through quantum controlled NOT gate, and reversibly mixing the current correlation quantum state with the historical quantum implicit map state through parameterized unitary transformation.

[0032] As described above, the beneficial effects of this invention are as follows: By quantum encoding environmental point cloud data and task text data respectively, environmental quantum states and task quantum states are generated. Simultaneously, a quantum safety operator is constructed based on electrical exclusion zone data, uniformly mapping environmental geometric information, task semantic information, and safety constraint information to the quantum state space. Based on this, the current environmental quantum state is quantum-entangled and fused with the historical quantum implicit map state from the previous moment. Utilizing the non-local correlation characteristics of quantum entanglement, environmental information from different moments is integrated through quantum correlation, generating a current-moment quantum implicit map state with temporal continuity. This allows for coherent cumulative mapping of large-scale scenes without explicit storage of point cloud maps. Furthermore, this quantum implicit map state is quantum-entangled and enhanced with the task quantum state to generate a quantum fusion operator characterizing the matching degree between task semantics and map position, achieving efficient alignment of task targets and spatial positions at the quantum level. Subsequently, an observation quantum state is constructed based on this quantum fusion operator and the quantum implicit map state, and quantum evolution is performed on this observation quantum state to generate candidate inspection paths. Finally, electrical safety constraints are applied to candidate inspection paths using quantum-safe operators to generate safe inspection paths, ensuring absolute avoidance of electrical restricted areas. This invention unifies environmental perception, task understanding, path planning, and safety constraints within the entire process of quantum state encoding, fusion, enhancement, evolution, and constraint. This avoids error accumulation caused by traditional multi-module serial processing and improves the real-time performance, global optimization capability, and safety of inspection path decision-making in complex wind and solar power environments.

[0033] Furthermore, based on the environmental point cloud data and the task text data, quantum encoding is performed to generate environmental quantum states and task quantum states, and a quantum safety operator is constructed based on the electrical exclusion zone data, including:

[0034] The environmental point cloud data is downsampled to obtain a key point dataset;

[0035] The keypoint dataset is divided into voxels at multiple spatial scales to obtain multi-scale voxel features;

[0036] Based on the multi-scale voxel features, target environment features are extracted, and the target environment features are mapped to the first quantum state space through amplitude encoding to generate environment quantum states.

[0037] The task text data is encoded into target text features, and the target text features are mapped to the second quantum state space through angle encoding to generate the task quantum state;

[0038] Based on the electrical exclusion zone data, the preset exclusion zone space identifier is encoded into the weight coefficients of the operators corresponding to the qubits, thus obtaining the quantum safe operator.

[0039] Specifically, the dimensions of the target environmental features are reduced to integer powers of 2 through principal component analysis or zero-filling to meet the normalization and dimension matching requirements of amplitude encoding; each dimension component of the target text features corresponds to the Bloch spherical rotation angle of a single qubit, and the number of dimensions is equal to the number of qubits in the second quantum state space; based on the electrical restricted area data, the preset restricted area space identifier is encoded as the weight coefficient of the corresponding operator of the qubit, and the quantum safety operator is obtained by mapping the space voxel to the qubit index through hash index, and encoding the restricted area space identifier as the binary weight coefficient of the corresponding Pauli-Z operator.

[0040] As described above, the environmental point cloud data is first downsampled, compressing the original massive point cloud into a keypoint dataset, effectively reducing the input size for subsequent quantum encoding. The keypoint dataset is then divided into voxel segments at multiple spatial scales to obtain multi-scale voxel features, allowing objects of different sizes to be represented at their appropriate resolutions. Based on these multi-scale voxel features, target environmental features are extracted and mapped to the first quantum state space using amplitude encoding to generate environmental quantum states. Amplitude encoding enables n qubits to simultaneously encode 2... n This method achieves efficient compressed mapping from classical feature vectors to quantum superposition states across multiple feature dimensions. Simultaneously, task text data is encoded into target text features and mapped to a second quantum state space via angle encoding to generate task quantum states. Angle encoding maps each semantic component to the rotation angle of a single qubit, enabling the semantic information of natural language to be converted into geometric differences in quantum states. Based on electrical exclusion zone data, pre-defined exclusion zone space identifiers are encoded into weight coefficients of operators corresponding to qubits, yielding quantum-safe operators. This expresses exclusion zone rules as computable quantum mechanical observables. This method achieves a standardized transformation from unstructured point clouds, natural language text, and exclusion zone rules to quantum state spaces and quantum operators, providing standardized quantum input for subsequent processing.

[0041] Furthermore, the current quantum state of the environment is quantumly entangled and fused with the historical implicit quantum map state from the previous moment to generate the current quantum implicit map state, including:

[0042] By establishing the correlation between the quantum states of the environment at different spatial scales through quantum entanglement, the correlated quantum states are obtained;

[0043] The current quantum state is quantumly entangled and fused with the historical quantum implicit map state from the previous time step to obtain the current quantum implicit map state.

[0044] As described above, the process begins by establishing a correlation between the environmental quantum states at different spatial scales through quantum entanglement, resulting in correlated quantum states. Quantum entanglement transforms the spatial information at fine, medium, and coarse scales from being independent to forming cross-scale mutual constraints. Changes at one scale affect corresponding components at other scales through entanglement, ensuring the consistency of multi-scale spatial information. Building upon this, the correlated quantum state at the current moment is fused with the historical implicit quantum map state from the previous moment, yielding the current implicit quantum map state. By fusing new observation information with historical accumulated information at the quantum level, the map state can continuously absorb new observations and update its state over time, forming a dynamic map representation with temporal coherence. This entanglement fusion mechanism allows the implicit quantum map state to carry long-term accumulated environmental information with extremely low qubit overhead, providing temporal consistency guarantees for subsequent tasks and map alignment, and preventing map information loss due to temporary sensor obstruction or interference.

[0045] Furthermore, by establishing the correlation between the environmental quantum states at different spatial scales through quantum entanglement, the correlated quantum states are obtained, including:

[0046] The environmental quantum state of a single frame is divided into subsets of different spatial scales, and each subset is encoded as a quantum voxel feature at the corresponding spatial scale.

[0047] Quantum entanglement between quantum voxel features at different spatial scales is established by quantum controlled NOT gates, thereby obtaining the correlated quantum states between quantum voxel features at different spatial scales under the said environmental quantum state.

[0048] As described above, the environmental quantum state of a single frame is divided into subsets of different spatial scales, and each subset is encoded as a quantum voxel feature at the corresponding spatial scale. The fine-scale subset captures local geometric details, the medium-scale subset captures structural information, and the coarse-scale subset captures the global layout, allowing multiple spatial resolutions of geometric information to be encoded simultaneously within the same quantum state. Subsequently, quantum entanglement between quantum voxel features at different spatial scales is established using quantum controlled NOT gates, resulting in correlated quantum states between these features at different spatial scales within the environmental quantum state. The layered application of controlled NOT gates transforms the subsets at each scale from direct product states into an indivisible entangled whole. When the quantum state at one scale changes, the entanglement synchronously affects the corresponding components at other scales. This cross-scale entanglement mechanism enables the spatial information at the fine, medium, and coarse scales to form a mutually verifying and constraining relationship at the quantum state level, avoiding conflicts and ambiguities between different scales and providing a compact and scale-consistent representation of environmental features for subsequent temporal fusion.

[0049] Furthermore, by quantum entanglement and fusion of the current correlated quantum state with the historical implicit quantum map state from the previous moment, the current implicit quantum map state is obtained, including:

[0050] The current correlated quantum state and the previous historical quantum implicit map state are input into the quantum loop circuit, and the correlated quantum state and the historical quantum implicit map state are fused through unitary transformation to obtain the current quantum implicit map state.

[0051] As described above, the current correlated quantum state and the previous historical implicit quantum map state are input into a quantum loop circuit. A unitary transformation is then used to fuse these two states, resulting in the current implicit quantum map state. The unitary transformation, a linear transformation in quantum mechanics that preserves probability, allows for reversible mixing and recombination of two input quantum states in Hilbert space. This quantum loop circuit, through the repeated action of parameterized unitary operators, enables quantum interference between the long-term accumulated information stored in the historical map state and the currently perceived information carried in the new observation state. The interference result depends on the phase relationship between the two states, thus achieving selective retention and updating of information. The current map state output by the unitary transformation not only fully retains the historically accumulated environmental knowledge but also absorbs the new observation information in real time, giving the map state a temporal memory function similar to that of a classical recurrent neural network. Furthermore, due to the continuity and high dimensionality of the quantum state space, it can achieve efficient storage and evolution of long-term temporal information with extremely low qubit overhead.

[0052] Furthermore, the quantum implicit map state and the task quantum state are quantum entangled to enhance quantum entanglement, generating a quantum fusion operator for characterizing the matching degree between the inspection task and the inspection environment, including:

[0053] The task quantum state is input into a quantum self-attention circuit composed of parameterized single-qubit rotating gates and controlled NOT gates stacked alternately, so as to redistribute the entanglement of each semantic component in the task quantum state through unitary transformation to obtain the aggregated task quantum state.

[0054] The aggregated task quantum state is subjected to a time evolution operation with a preset new energy Hamiltonian to map the general semantics of the task text data to new energy-specific semantics, thereby obtaining a special task quantum state.

[0055] Construct a joint quantum system based on the aforementioned quantum implicit map state and the aforementioned dedicated task quantum state;

[0056] The joint quantum system is input into a quantum cross-attention circuit consisting of a parameterized Pauli rotation gate and a cross-register entanglement gate, so as to entangle the quantum implicit map state with the dedicated task quantum state through a unitary transformation;

[0057] Perform quantum state tomography or partial trace operation on the entangled joint quantum system to reconstruct the quantum fusion operator acting on the subspace corresponding to the quantum implicit map state.

[0058] Among them, the new energy Hamiltonian is a parameterized Hermitian operator of the Ising or Heisenberg type preconstructed based on the topological relationship of new energy station equipment, and the time evolution operation is approximated by Trotter decomposition. Trotter decomposition is an approximate decomposition method based on Lie algebra exponential mapping, which is used to realize the quantum state time evolution operation driven by multi-component noncommutative Hamiltonians in the quantum computing system. The expectation value of the quantum fusion operator under the quantum correlation observable operator is used to characterize the matching degree between the task semantics and the map location.

[0059] As described above, the task quantum state is first input into a quantum self-attention circuit composed of parameterized single-qubit rotation gates and controlled NOT gates stacked alternately. This circuit uses unitary transformation to redistribute the entanglement of semantic components within the task quantum state, resulting in an aggregated task quantum state. The quantum self-attention circuit redistributes the entanglement structure between semantic components, giving the qubits corresponding to key equipment names higher entanglement weights, thus achieving the reorganization of semantic components. The aggregated task quantum state is then approximated by a time evolution operation using Trotter decomposition with a pre-constructed Ising-type or Heisenberg-type parameterized new energy Hamiltonian based on the topological relationship of new energy power station equipment. This maps the general semantics of the task text data to new energy-specific semantics, resulting in a dedicated task quantum state. The physical essence of the time evolution operator is to rotate the quantum state in Hilbert space, rotating the general semantic vector to the new energy-specific semantic subspace, thus injecting domain knowledge into the quantum state. A joint quantum system is constructed based on the quantum implicit map state and the dedicated task quantum state. This joint quantum system is then input into a quantum cross-attention circuit composed of a parameterized Pauli rotation gate and a cross-register entanglement gate to entangle the quantum implicit map state and the dedicated task quantum state through unitary transformation. The quantum cross-attention circuit establishes a deep entanglement correlation between the task register and the map register. Finally, quantum state tomography or partial trace operation is performed on the entangled joint quantum system to reconstruct a quantum fusion operator acting on the map state subspace. The expectation value of the quantum fusion operator under the quantum correlation observable operator directly reflects the degree of matching between each semantic component in the task text and each spatial location in the map; regions with higher matching degrees have greater weights in the fusion operator. This method enables the semantic information of the task text to directly establish a quantum-level correlation with the spatial location of the map, achieving efficient alignment of task semantics and spatial information.

[0060] Furthermore, constructing a joint quantum system based on the quantum hidden map state and the dedicated task quantum state includes:

[0061] The quantum implicit map state density matrix is ​​obtained by taking the outer product of the quantum implicit map state and its conjugate transpose.

[0062] The dedicated task quantum state is outer productted with its conjugate transpose to obtain the dedicated task quantum state density matrix.

[0063] The quantum implicit map state density matrix and the dedicated task quantum state density matrix are combined by tensor product to obtain a joint quantum system.

[0064] One approach is to extract the task-map coupled pure state from the joint quantum system as a valid input for the subsequent unitary transformation, or to directly use the quantum implicit map state density matrix and the dedicated task quantum state density matrix as valid inputs for the subsequent unitary transformation.

[0065] As described above, the outer product of a quantum implicit map state and its conjugate transpose yields the quantum implicit map state density matrix. This outer product operation multiplies a pure state vector by its conjugate transpose, resulting in a square matrix density matrix that fully preserves all information of the original quantum state and naturally satisfies the physical legitimacy requirements of Hermitianness, positive definiteness, and trace of 1. Similarly, the outer product of a dedicated task quantum state and its conjugate transpose yields the dedicated task quantum state density matrix. Subsequently, the quantum implicit map state density matrix and the dedicated task quantum state density matrix are combined using a tensor product to obtain a joint quantum system. The essence of the tensor product is to combine two independent subsystems into a composite system, where the dimension of the composite system is the product of the dimensions of the two subsystems. Extracting the task-map coupled pure state from the joint quantum system, or directly using the density matrix as a legitimate input for subsequent unitary transformations, unifies the quantum information originally belonging to the map register and the task register into the same Hilbert space. This eliminates the need for data transfer or format conversion at the classical level, providing a legitimate input state that conforms to the basic postulates of quantum mechanics for subsequent quantum cross-attention computation, thus ensuring the original integrity of the task semantics and map space at the quantum level.

[0066] Furthermore, the observed quantum state is subjected to quantum evolution to generate candidate inspection paths, including:

[0067] The observed quantum state is input into the quantum path strategy circuit so that the observed quantum state is mapped to the path action subspace through unitary evolution to obtain candidate action quantum states;

[0068] The candidate action probability distribution of the quantum state of the candidate action is calculated based on the quantum Born rule, and the candidate inspection path is determined based on the candidate action probability distribution.

[0069] As described above, the observed quantum state is input into the quantum path policy circuit, and unitary evolution maps the observed quantum state to the path action subspace to obtain candidate action quantum states. Unitary evolution involves parameterized rotation and distortion of the quantum state in Hilbert space, causing the probability amplitudes of each ground state in the observed quantum state to be redistributed according to the policy parameters, with the ground state corresponding to the high-reward action receiving a larger probability amplitude. Subsequently, the candidate action probability distribution of the candidate action quantum state is calculated based on the quantum Born rule. The Born rule is a fundamental law in quantum mechanics that connects quantum states and measurement probabilities. It stipulates that the probability of a measurement result occurring is equal to the square of the projection modulus of the quantum state onto the corresponding ground state. This rule naturally outputs a normalized probability distribution, eliminating the need for an additional softmax layer (normalization exponential layer). Furthermore, due to quantum interference effects, the probabilities of similar actions are coherently superimposed, resulting in a smoother action distribution. Candidate inspection paths are determined based on this candidate probability distribution, and the action sequence of the path is obtained through measurement sampling. This mechanism utilizes the superposition property of quantum states to achieve parallel exploration of the action space, rather than evaluating each action one by one, significantly improving the efficiency of path generation.

[0070] Furthermore, based on the quantum-safe operator, electrical safety constraints are applied to the candidate inspection paths to generate safe inspection paths, including:

[0071] Based on the quantum safe operator, a forbidden zone projection operator is constructed, and the forbidden zone component corresponding to the forbidden zone space in the observed quantum state is removed by the forbidden zone projection operator to obtain a safe quantum state;

[0072] The secure quantum state is input into the quantum path strategy circuit so that the secure quantum state is mapped to the path action subspace through unitary evolution to obtain the secure action quantum state;

[0073] The probability distribution of the safety action quantum state is calculated based on the quantum Born rule, and the safety inspection path is determined according to the probability distribution of the safety action.

[0074] In quantum circuits, the forbidden zone projection operator is implemented by introducing auxiliary qubits to perform controlled operations and then combining them with post-selection measurement, or in classical simulators by zeroing and renormalizing the state vector.

[0075] As described above, a forbidden zone projection operator is constructed based on quantum safety operators. This operator removes the forbidden zone components corresponding to the forbidden zone space from the observed quantum state, resulting in a safe quantum state. The projection operator is a Hermitian and idempotent linear operator. Its function is to project the components belonging to the forbidden zone subspace of the quantum state to this subspace while preserving the non-forbidden zone components. Mathematically, this is equivalent to forcibly setting the probability amplitude of falling into the forbidden zone to zero, while maintaining the normalized relative proportion of the probability amplitude of the feasible region. The safe quantum state is input into the quantum path strategy circuit, and through unitary evolution, it is mapped to the path action subspace to obtain the safe action quantum state. Since the unitary operator is linear, the forbidden zone components that have been set to zero in the input remain zero in the output and will not reappear due to linear transformation. Subsequently, the safe action probability distribution of the safe action quantum state is calculated based on the quantum Born rule, and a safe inspection path is determined according to this candidate probability distribution. If the observed quantum state is entirely within the forbidden zone subspace, a zero vector is obtained after projection, the strategy circuit outputs all zeros, and an emergency stop action is triggered. This safety constraint mechanism severs the connection between the forbidden zone and the feasible zone from the structure of the quantum state space, making it impossible for the generated path to enter the forbidden zone at the physical level, rather than using a penalty function for soft constraints, thus fundamentally ensuring electrical safety.

[0076] Another embodiment of the present invention provides an inspection path decision system based on quantum implicit map states, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the various steps in the above-described inspection path decision method based on quantum implicit map states.

[0077] As described above, by quantum encoding environmental point cloud data and task text data respectively, environmental quantum states and task quantum states are generated. Simultaneously, a quantum safety operator is constructed based on electrical exclusion zone data, uniformly mapping environmental geometric information, task semantic information, and safety constraint information to the quantum state space. By fusing the current environmental quantum state with historical quantum implicit map states through quantum entanglement, the non-local correlation characteristics of quantum entanglement are utilized to achieve coherent accumulation of information across multiple time points, generating a dynamic map state that does not require explicit storage of point clouds and is naturally adapted to large-scale sparse scenes. Further quantum entanglement enhancement between the map state and the task quantum state allows for a direct quantum correlation between task semantics and map spatial location, achieving efficient alignment between task and space. Quantum evolution is performed based on the fused observation quantum state, utilizing quantum superposition to evaluate multiple paths in parallel, improving decision-making efficiency. Finally, a physical hard constraint is implemented through the quantum safety operator to ensure that the path absolutely avoids electrical exclusion zones. The entire process unifies perception, understanding, planning, and safety within a quantum state evolution framework, improving the real-time performance, globality, and safety of path decision-making in complex wind and solar power environments.

[0078] The inspection path decision-making method and system based on quantum implicit map states described above are applicable to autonomous inspection path planning and decision-making scenarios of embodied intelligent agents in complex outdoor environments in wind and solar power stations (including onshore / offshore wind farms and centralized / distributed photovoltaic power stations). This improves the real-time performance, globality, and security of the embodied intelligent agent in path decision-making. The following detailed implementation methods illustrate this:

[0079] Please refer to Figure 1 Embodiment 1 of the present invention:

[0080] A method for inspection path decision-making based on quantum implicit map states, comprising S1 to S6:

[0081] S1. Acquire environmental point cloud data, task text data, and electrical restricted area data of the wind and solar power station.

[0082] The environmental point cloud data is collected by airborne LiDAR or panoramic cameras at wind and solar power plants. The task text data includes specific instructions for new energy dispatch and inspection tasks; for example, the wind turbine inspection task instruction is: inspect the blade appearance, tower foundation settlement, and transformer oil temperature of wind turbine #03; the photovoltaic inspection task instruction is: check for string hot spots and support deformation of the photovoltaic array in area A; the substation safety supervision task instruction is: conduct a safety supervision inspection of the main transformer area of ​​the 110kV substation; and the collector line inspection task instruction is: inspect the corridor of collector line #05. Electrical restricted area data includes high-voltage electrical equipment areas, energized bays, wind turbine blade swept surfaces, and the area inside the transformer substation enclosure at wind and solar power plants.

[0083] S2. Quantum encoding is performed on environmental point cloud data and task text data respectively to generate environmental quantum states and task quantum states, and quantum safety operators are constructed based on electrical exclusion zone data.

[0084] Specifically, S2 includes S21-S25:

[0085] S21. Downsample the environmental point cloud data to obtain the key point dataset.

[0086] In some embodiments, for T Frame environment point cloud data FarthestPoint Sampling (FPS) is used to downsample the environmental point cloud data of each frame from the original... Downsampling at each point Each point, environmental point cloud data per frame It contains three-dimensional spatial coordinate information; , n Let be the number of qubits, satisfying .

[0087] S22. Perform multi-scale voxel partitioning on the keypoint dataset to obtain multi-scale voxel features.

[0088] In some embodiments, a voxel resolution of 0.05m is used for fine equipment such as transformer substations and combiner boxes; a voxel resolution of 0.5m is used for medium-sized structures such as wind turbine towers and photovoltaic array supports; and a voxel resolution of 2m is used for large facilities such as substation buildings and power line corridors.

[0089] Since the lidar of a wind and solar power station can generate 100,000 to 1 million points per frame, directly quantum encoding the point cloud data would require thousands of qubits, far exceeding the capabilities of current NISQ (noisy mesoscale quantum) devices. By using FPS downsampling in S21 and multi-scale voxelization in S22, the point cloud is compressed to... Several key points (e.g., 1024 points corresponding to 10 qubits) satisfy the quantum amplitude encoding pair The dimensionality requirement is met while preserving multi-scale geometric structures (such as centimeter-level details of transformer substations and hundred-meter-level outlines of wind turbines), providing standardized classical feature inputs for subsequent quantum coding.

[0090] S23. Extract target environment features based on multi-scale voxel features, and map the target environment features to the first quantum state space through amplitude encoding to generate the environment quantum state.

[0091] In some embodiments, S23 specifically refers to:

[0092] S231. Extract the target environment features of each frame of environment point cloud data using the classic PointNet++ (a deep learning network architecture for extracting 3D point cloud features):

[0093] ;

[0094] in, Let t be the target environmental features of the environmental point cloud data in frame t; For the environmental point cloud data of frame t, This represents the dimension of the target environment features. It maps the disordered point cloud into an ordered feature vector. PointNet++, through hierarchical sampling and local feature aggregation, exhibits permutation invariance to the point cloud arrangement, making it suitable for unstructured outdoor terrain. Output dimension Directly determines the number of bits in the subsequent quantum encoding. Therefore, we obtain T Frame target environment features .

[0095] when When the feature dimension is not a power of 2, principal component analysis or zero-padding can be used to normalize the feature dimension. This is to meet the dimension matching requirements of amplitude coding.

[0096] S232. Normalize the target environment features and map them to... Amplitude of each qubit:

[0097] ;

[0098] in, Let t be the environmental quantum state corresponding to the environmental point cloud data of frame t; Let i be the amplitude coefficient corresponding to the i-th calculated ground state; To calculate the ground state, the binary representation of i corresponds to the feature index; Let be the number of qubits. The above formula is the standard form of quantum amplitude encoding. yes The pure state of a quantum bit, its ground state complex amplitude Based on the characteristics of the target environment The decision; the physical meaning of the above formula is as follows: Physical qubits are superimposed in an exponentially large Hilbert space. The code encodes the global geometric information of the point cloud, laying the foundation for subsequent quantum parallel computing.

[0099] The amplitude coefficient is specifically expressed as:

[0100] ;

[0101] in, Target environmental characteristics The i-th component; The target environment features are defined by the L2 norm, making the encoding process reversible (i.e., known). It can be deduced by measurement. (relative distribution); amplitude coefficients ensure that the encoded quantum state satisfies the probability amplitude normalization condition, normalization constraint .

[0102] S24. Encode the task text data into target text features, and map the target text features to the second quantum state space through angle encoding to generate the task quantum state.

[0103] In some embodiments, S24 specifically includes:

[0104] S241. Encode the task text data into target text features using the classic BERT (Bidirectional Encoder-Representation Transformer). , is the dimension of the target text feature, corresponding to the number of qubits.

[0105] when When the number of available qubits is greater than the number of available qubits, principal component analysis is used to reduce the dimensionality of the target text features to the specified number of qubits, so that each dimensional component uniquely corresponds to the rotation angle of a qubit.

[0106] S242. Map the task text features to component-wise angle encoding. One quantum bit:

[0107] ;

[0108] in, For the task quantum state; For tensor product operations; Let be the rotation angle of the j-th qubit; , This represents the computational ground state for a single quantum bit. The components of each target text feature... Bloch spherical rotation angle corresponding to one qubit The physical meaning of the above formula is that it maps the semantic information of the task text to the polarization direction of the qubits on the Bloch sphere, so that the semantic differences between "wind turbine inspection" and "photovoltaic hot spot investigation" are reflected as geometric differences in quantum states, which facilitates the subsequent differentiation and processing by quantum circuits.

[0109] Encoding angle:

[0110] ;

[0111] in, The first feature of the target text j There are several components; arctan is the arctangent function, used to compress unbounded target text features to an effective angular range. The arctangent function compresses unbounded classical features to... Range, then scaled to This ensures that the rotation angle of each qubit is within the effective range of the Bloch sphere, avoiding encoding distortion caused by angle overflow.

[0112] S25. Based on the electrical restricted area data, the preset restricted area space identifier is encoded into the weight coefficients of the corresponding operators of the qubits to obtain the quantum safe operator.

[0113] In some embodiments, S25 specifically refers to:

[0114] S251. Discretize the three-dimensional spatial region corresponding to the electrical restricted area data into voxels, and map the voxel index to the qubit index through a hash function to establish a one-to-one correspondence between spatial location and qubit.

[0115] S252. Encode the electrical exclusion zone data into the weight coefficients of the Pauli-Z operator to obtain the quantum safety operator:

[0116] ;

[0117] in, A quantum-safe operator, acting on One quantum bit; The weight coefficient for the forbidden zone indicator corresponding to the k-th qubit is... This indicates that the spatial region corresponding to the qubit is an electrically hazardous area; Let $\frac{ ...

[0118] As described above, by mapping point cloud features to quantum superposition states through amplitude encoding, the... One qubit can be encoded simultaneously It achieves exponential information compression by using multiple feature dimensions (such as 10 qubits encoding 1024-dimensional features); it maps scheduling task text to quantum states through angle encoding, injecting prior knowledge from the new energy field; and it pre-encodes electrical exclusion zone information to provide an operator basis for subsequent hard constraints.

[0119] S3. Perform quantum entanglement fusion between the current environment quantum state and the previous historical quantum implicit map state to generate the current quantum implicit map state.

[0120] Specifically, S3 includes S31-S32:

[0121] S31. By establishing the correlation between the quantum states of the environment at different spatial scales through quantum entanglement, the correlated quantum states are obtained.

[0122] Specifically, S31 includes S311-S312:

[0123] S311. Divide the environmental quantum state of a single frame into subsets of different spatial scales, and encode each subset as a quantum voxel feature at the corresponding spatial scale.

[0124] In some embodiments, S311 specifically refers to:

[0125] S3111, will The qubits are divided into three subsets (sized as follows): ,satisfy Each subset encodes voxel features at one scale:

[0126] ;

[0127] in, Let be the quantum voxel feature of the point cloud in frame t at the s-th spatial scale; These correspond to fine scale (0.05m), medium scale (0.5m), and coarse scale (2m), respectively. Let be the amplitude coefficient of the i-th ground state at the s-th scale; Let be the number of qubits in the s-th scale subset. The three scales correspond to different spatial resolutions: the fine scale captures the local geometry of transformer substations and combiner boxes; the medium scale captures the tower and support structures; and the coarse scale captures the global layout of the building and corridors. Because... The total dimension of the three-scale tensor product is equal to that of the original tensor product. Strict matching with no information loss.

[0128] S312. By establishing quantum entanglement between quantum voxel features at different spatial scales through quantum controlled NOT gates, the correlated quantum states between quantum voxel features at different spatial scales under the environmental quantum state are obtained.

[0129] In some embodiments, S312 specifically includes:

[0130] Achieving multi-scale correlations through quantum entanglement gates:

[0131] ;

[0132] in, It is a multi-scale correlated quantum state; This represents a controlled NOT gate operation with qubits of a scale 1 subset as control bits and qubits of a scale 2 subset as target bits; This represents a controlled NOT gate operation with qubits from a subset at scale 2 as control bits and qubits from a subset at scale 3 as target bits. CNOT (Controlled NOT Gate) is a fundamental building block of quantum entanglement. The physical meaning of the above formula is that it entangles quantum subsets at three independent scales into a unified state, enabling quantum correlation between spatial information at different scales; when measurements reveal that a certain region is a wind turbine tower at the coarse scale, the entanglement effect automatically constrains that the region should be a surface texture of the tower at the fine scale, achieving consistent reasoning across scales.

[0133] S32. Perform quantum entanglement fusion between the current correlated quantum state and the historical quantum implicit map state of the previous time to obtain the current quantum implicit map state.

[0134] Specifically, S32 includes S321:

[0135] S321. Input the current correlated quantum state and the previous historical quantum implicit map state into the quantum loop circuit, so as to fuse the correlated quantum state and the historical quantum implicit map state through unitary transformation to obtain the current quantum implicit map state.

[0136] In some embodiments, an initial map state is set. This could be the first frame's associated quantum state, serving as the initial value for the historical quantum implicit map state; historical observation information is integrated through a quantum loop circuit to construct the quantum implicit map state of the wind and solar power station:

[0137] ;

[0138] in, This represents the quantum implicit map state at the current time t; The historical quantum implicit map state of the previous moment; For parameterized quantum loop circuits, Its learnable parameters, number of circuit layers It contains entangled layers and rotational layers; This represents the multi-scale correlated quantum state at the current moment. The physical meaning of the above formula is that the map state at each moment contains all historical information, eliminating the need to maintain an explicit point cloud database like in classic SLAM (Simultaneous Localization and Mapping), thus fundamentally solving the memory explosion problem in large-scale outdoor scenes.

[0139] Specifically, in quantum implicit map states, three types of regions are encoded through the phase difference of quantum states:

[0140] ;

[0141] in, Let be the phase angle of the i-th ground state in the quantum hidden map state; For complex probability amplitude, In quantum mechanics, phase typically does not directly affect measurement probability, but it does influence interference effects. This invention leverages this characteristic: it encodes the safety level of an electrical region as phase rather than amplitude, allowing safety region information to play a role in quantum interference (subsequent alignment calculations) while remaining implicit in direct measurement, thus avoiding navigation errors caused by misclassification. Phase information indirectly affects the coherent superposition of action probability amplitudes by participating in the quantum interference of subsequent quantum path strategy circuits; specifically, regions with higher safety levels experience more significant phase cancellation during interference, resulting in an avoidance tendency in Born rule measurements, ensuring that the phase-encoded safety semantics are ultimately translated into an observable action probability distribution.

[0142] As described above, traditional SLAM is prone to failure in featureless desert areas and high-dynamic areas under wind turbine shadows, while quantum implicit maps encode historical observations into single quantum states through temporal quantum circuits. It eliminates the need for explicit point cloud map maintenance, naturally adapting to large-scale sparse scenarios. Multi-scale encoding preserves both centimeter-level details of transformer substations and hundred-meter-level contours of wind turbines through qubit subset partitioning, while phase partitioning directly encodes electrical safety semantics, providing a structured quantum map representation for subsequent alignment.

[0143] S4. Enhance the quantum entanglement between the quantum implicit map state and the task quantum state to generate a quantum fusion operator for characterizing the matching degree between task semantics and map position.

[0144] Specifically, S4 includes S41-S45:

[0145] S41. Input the task quantum state into the quantum self-attention circuit to redistribute the semantic components in the task quantum state through unitary transformation to obtain the aggregated task quantum state.

[0146] In some embodiments, S41 specifically includes:

[0147] S411, using a quantum self-attention circuit composed of parameterized single-qubit rotating gates and controlled NOT gates stacked alternately. of Entangled redistribution of semantic components:

[0148] ;

[0149] in, For the aggregation task quantum state; For parameterized quantum self-attention circuits, Its learnable parameters; This is the task quantum state output in step S242. Quantum self-attention redistributes the entanglement structure of each qubit through unitary transformation, giving higher weights to the qubits corresponding to the key equipment names (such as "#03 wind turbine"). The physical meaning of the above formula is to realize quantum correlation between semantic components, so that the blade appearance and tower base settlement form an entangled pair in the quantum state, indicating that they belong to different sub-targets of the same inspection task.

[0150] S42. Perform time evolution operation on the aggregated task quantum state and the preset new energy Hamiltonian to map the general semantics of the task text data to the new energy-specific semantics, and obtain the special task quantum state.

[0151] In some embodiments, S42 specifically includes:

[0152] S421. By pre-constructing Ising-type or Heisenberg-type parameterized new energy Hamiltonian based on the topological relationship of new energy power station equipment, and approximating time evolution through Trotter decomposition, the prior information of wind turbines, photovoltaics, or booster stations is injected:

[0153] ;

[0154] in, For dedicated task quantum states; The Hamiltonian pre-trained for the new energy field is used to map general semantics to a new energy-specific semantic subspace; These are time evolution parameters; This refers to the time evolution operator in quantum mechanics. The physical significance of the above formula lies in: "rotating" the semantics of a general language to a new energy-specific subspace; for example, enabling hot spot detection to automatically associate with photovoltaic strings, rather than general hot spots; enabling safety inspections to automatically associate with fence integrity, rather than general walls, thus realizing the quantum state injection of domain knowledge.

[0155] S43. Construct a joint quantum system based on quantum hidden map states and dedicated task quantum states.

[0156] Specifically, S43 includes S431-S433:

[0157] S431. The quantum hidden map state density matrix is ​​obtained by performing an outer product of the quantum hidden map state and its conjugate transpose.

[0158] In some embodiments, S431 specifically refers to:

[0159] ;

[0160] in, This is the quantum hidden map state density matrix; It is a quantum hidden map state; for The conjugate transpose of; L represents the number of map locations.

[0161] S432. The dedicated task quantum state is externally productted with the conjugate transpose of the dedicated task quantum state to obtain the dedicated task quantum state density matrix.

[0162] In some embodiments, S432 specifically includes:

[0163] ;

[0164] in, For dedicated mission quantum state density matrix; The dedicated task quantum state output in step S42; of Conjugate transpose; , The number of qubits for the task.

[0165] S433. Combine the quantum hidden map state density matrix with the dedicated task quantum state density matrix by tensor product to obtain a joint quantum system.

[0166] In some embodiments, the joint quantum system specifically refers to: Extract the task-map coupled pure state from the joint quantum system, or directly use the density matrix as a legitimate input for subsequent unitary transformations.

[0167] S44. Input the joint quantum system into a quantum cross-attention circuit consisting of a parameterized Pauli rotation gate and a cross-register entanglement gate, so as to entangle the quantum implicit map state with the dedicated task quantum state through a unitary transformation.

[0168] In some embodiments, a parameterized quantum attention circuit is constructed:

[0169] ;

[0170] in, For quantum cross-attention circuits, Its set of learnable parameters; This refers to the number of circuit layers. The Hamiltonian of the Pauli string at each level; For the first l Layer k Learnable parameters corresponding to each Hamiltonian; Let be the Hamiltonian associated with the k-th task and the map. This circuit is a Hardware-Efficient Ansatz structure, which achieves deep entanglement between the task register and the map register through layer-by-layer unitary transformation. It is the standard evolutionary form in quantum mechanics, guaranteeing that each step is a reversible unitary transformation, ensuring that quantum information is not lost. Among these:

[0171] ;

[0172] Where i is the qubit index of the task register and j is the qubit index of the map register; For the number of qubits in the task, The number of qubits in the map; Let be the learnable coupling strength between task qubit i and map qubit j in the k-th Hamiltonian; The Pauli operator acts on the i-th qubit; The Pauli operator acts on the j-th qubit; Represents the tensor product. This is the standard form of the exchange interaction between two qubits (Heisenberg interaction). The physical meaning of the above formula is that when a task token i (e.g., wind turbine #03) and map location j (e.g., tower coordinates) should be associated, When the result is positive, quantum entanglement strongly correlates two qubits; when they should not be correlated, Approaching zero enables selective alignment.

[0173] Alignment characteristics include: 1. Equipment-level alignment, such as "#03 wind turbine" being associated with the three-dimensional cluster of tower, blade, or transformer through quantum entanglement; 2. Region-level alignment, such as "Area A photovoltaic" being enhanced with the array region through quantum phase interference; 3. Safety constraint alignment, such as "electrical hazard" being associated with the high-voltage equipment region through quantum forbidden transitions; 4. Channel-level alignment, such as "cleaning channel" being associated with the reserved channel through quantum path integrals.

[0174] S45. Perform quantum state tomography or partial trace operation on the entangled joint quantum system to reconstruct the quantum fusion operator acting on the map state subspace; calculate the expectation value of the quantum fusion operator under the quantum correlation observable operator to obtain the quantum fusion feature used to characterize the matching degree between task semantics and map position.

[0175] In some embodiments, S44 and S45 are specifically:

[0176] First, define the quantum correlation observable operator. ,in For learnable weights, The Pauli-Z tensor product of task qubit i and map qubit j is then performed; subsequently, partial trace operations are performed on the entangled joint quantum system to extract the effective quantum fusion operator acting on the map state subspace.

[0177] ;

[0178] in, A quantum fusion operator that operates on the map state subspace; This indicates that the offset is calculated from the task register; For an associative observable operator that operates only on the map register; the expected value of this operator. Represents the degree of matching between task semantics and map location.

[0179] Compared to the classical attention mechanism, this quantum scheme has the potential to achieve exponential parallelism under ideal quantum hardware conditions. In current NISQ simulation experiments, its performance advantage depends on the size of the qubits, the circuit depth, and the error rate. The above complexity comparison is a theoretical upper limit analysis.

[0180] S5. Construct the observation quantum state based on quantum fusion operators and quantum implicit map states, and perform quantum evolution on the observation quantum state to generate candidate inspection paths.

[0181] In some embodiments, a quantum fusion operator and a quantum implicit map state are combined into an observation quantum state:

[0182] ;

[0183] in, To observe quantum states, the dimension is ; The quantum fusion operator output in step S45; For quantum hidden map states; denominator As a normalization factor, ensure This is a legitimate input for subsequent unitary evolution. Thus, the observed quantum state can carry dual information: where it is (map) and where it is going (task alignment).

[0184] Specifically, in S5, the observed quantum state undergoes quantum evolution to generate candidate inspection paths, including S51-S52:

[0185] S51. Input the observed quantum state into the quantum path strategy circuit so that the observed quantum state can be mapped to the path action subspace through unitary evolution to obtain the candidate action quantum state.

[0186] In some embodiments, constructing a parameterized quantum path strategy circuit specifically involves:

[0187] ;

[0188] in, For quantum path strategy circuits, Its set of learnable parameters; q represents the number of strategy circuit layers; q represents the qubit index. To observe the number of qubits; For the axis Rotation Single-qubit gate; This is a controlled Z-gate acting between qubits p and q. This circuit is a standard variational quantum circuit (VQA): a single-qubit rotation gate implements parameterized characteristic transformation, and a CZ entanglement gate implements the correlation between qubits. The physical meaning of the above formula is: to observe the quantum state... Mapped to the path action subspace, rotation angle The quantum amplitude corresponding to high-reward actions is maximized through reinforcement learning optimization.

[0189] S52. Calculate the candidate action probability distribution of the quantum state of the candidate action based on the quantum Born rule, and determine the candidate inspection path according to the candidate action probability distribution.

[0190] In some embodiments, the probability distribution of candidate actions is output based on the quantum Born rule:

[0191] ;

[0192] in, In order to observe quantum states Select action The quantum strategy probability; For the current action, For action space Any candidate action in the list; Candidate action quantum states; For quantum path strategy circuits; To observe quantum states; Candidate action quantum state The conjugate transpose of the denominator is the sum of the squares of the probability magnitudes of all candidate actions, ensuring probability normalization. Indicates from observation After policy evolution U, it collapses into action. The probability of quantum strategies is naturally output. Quantum strategies output a probability distribution (without requiring an additional Softmax layer), and due to quantum interference effects, the probabilities of similar actions (such as turning 5 degrees left and turning 6 degrees left) coherently superimpose, making the strategy smoother and avoiding the jitter problem of classical strategies. For candidate action quantum states, Linear velocity; Angular velocity; The pitch angle.

[0193] S6. Apply electrical safety constraints to candidate inspection paths based on quantum safety operators to generate safe inspection paths.

[0194] Specifically, S6 includes S61-S63:

[0195] S61. Construct a forbidden zone projection operator based on quantum safe operators, and remove the forbidden zone component corresponding to the forbidden zone space in the observed quantum state through the forbidden zone projection operator to obtain a safe quantum state.

[0196] In some embodiments, a forbidden zone projection operator is constructed based on quantum-safe operators:

[0197] ;

[0198] in, is the forbidden zone projection operator, which acts on the observation quantum state space; s is the forbidden zone quantum state index; Forbidden quantum state; for The conjugate transpose of; This is a set of forbidden quantum states. The projection operator is a fundamental tool in quantum mechanics describing the collapse caused by measurement. The projection operator is idempotent. Once projection is complete, subsequent projections yield the same result. Based on this characteristic, the forbidden zone constraint is absolute, unlike classical neural networks which can be bypassed through adversarial examples. The set of forbidden zone quantum states is determined by a security operator threshold.

[0199] ;

[0200] in, The quantum-safe operator output in step S251; For quantum-safe operators in quantum states The expected value reflects the degree of danger in the area; This represents the safety threshold. The above formula divides the continuous quantum state space into discrete safe or dangerous regions, providing a well-defined domain for projection operations. The safety threshold is determined by the site safety regulations, such as the energized compartments of a booster station. Wind turbine blade sweeping surface Inside the transformer substation fence .

[0201] Quantum projection operators possess idempotency, ensuring constraint stability; they also possess Hermitianness, ensuring physical observability; and they exhibit orthogonality, ensuring strict separation between forbidden and feasible regions. By applying quantum projection constraints to candidate actions, the observed quantum state is projected onto the feasible region: ,in This is a unit operator. If the projected state vector is zero, an emergency stop action is triggered.

[0202] S62. Input the secure quantum state into the quantum path strategy circuit so that the secure quantum state can be mapped to the path action subspace through unitary evolution to obtain the secure action quantum state.

[0203] S63. Calculate the probability distribution of safety action quantum states based on the quantum Born rule, and determine the safety inspection path based on the probability distribution of safety actions.

[0204] In some embodiments, S61-S63 specifically refer to:

[0205] ;

[0206] in, The probability of the quantum strategy after security constraints; For the restricted area projection operator; It is a safe quantum state (the observed quantum state after removing the forbidden component). Acting on the observation space (dimension) ), which removes the forbidden zone component from the observed state (mathematically equivalent to setting the forbidden zone probability amplitude to zero). The secure quantum state is then mapped to the action space. The above formula mathematically guarantees the observation space. With action space Dimensional matching.

[0207] like Then, an emergency stop action will be output. .when When completely within the restricted area, the projected value is zero, with both the numerator and denominator being zero, triggering a preset emergency stop; when When a portion is within a restricted area, only feasible differentiation-driven strategies are retained. The expected value of the action after safety constraints is:

[0208] ;

[0209] in, The expected value of the action after safety constraints; For action space The action in; To perform secondary measurement sampling based on the probability distribution of safe actions, the actions... The number of times it was measured; This represents the total number of samples taken for quantum measurements.

[0210] When the agent approaches the forbidden zone, the projection operator forces the forbidden zone probability amplitude in the observed quantum state to zero:

[0211] ;

[0212] in, To determine the expectation value of the observed quantum state under the forbidden zone projection operator, it is shown that the observed quantum state is completely orthogonal to the forbidden zone subspace. Since the policy circuit is a unitary linear operator, a zero input necessarily results in a zero output, meaning that for all... Established (triggers emergency stop). This achieves a physically impossible safety constraint, corresponding to forbidden transitions in quantum mechanics, which are physically insurmountable and meet the absolute safety requirements of the power industry.

[0213] It should be noted that the safety inspection path refers to a continuous executable trajectory in space and time that the intelligent agent ultimately executes, determined by linear velocity. angular velocity and pitch angle It consists of a continuous sequence of actions, including the agent's position, attitude, and motion state information at each moment. The expected value of the action after safety constraints quantitatively describes the motion state of the safety inspection path at each moment in the form of linear velocity, angular velocity, and pitch angle, and is a statistical characteristic expression of the path.

[0214] The dedicated strategies for wind and solar power stations include: 1. Wind turbine inspection: Automatically navigating to the windward side, with quantum strategy automatically enhancing obstacle avoidance probability in the blade gap area; 2. Photovoltaic inspection: Snake-like traversal along the cleaning channel, with quantum path integral optimizing the traversal order; 3. Substation safety monitoring: Patrol along the outside of the fence, with quantum projection operator ensuring that the quantum state forbidden zone component observed inside the fence is always zero; 4. Collector lines: Patrol along the center line of the corridor, with quantum phase guidance maintaining trajectory stability.

[0215] In a specific application scenario, the above method was applied to a 50MW onshore wind farm in a 32-qubit classical simulator environment. The embodied intelligent agent entered the station for the first time and received the dispatch inspection task instruction: to inspect the blade appearance, tower foundation settlement and transformer oil temperature of wind turbine #07, and to prioritize the inspection of the windward blades. The current wind speed is 8m / s.

[0216] Specifically, the LiDAR on the intelligent agent collects environmental point cloud data at a frequency of 10Hz, retaining 30 frames. The following operations are performed on each frame of the point cloud: First, downsampling is performed using FPS, reducing the original point cloud data in each frame... Downsampling at each point The process involves identifying key points; subsequently, the downsampled key points are divided into voxels at multiple spatial scales, with detailed equipment, medium-sized structures, and large facilities encoded hierarchically at voxel resolutions of 0.05m, 0.5m, and 2m, respectively; finally, PointNet++ is used to extract the target environment features for each frame. The extracted target environmental features are amplitude encoded, and the following method is used: One quantum bit:

[0217] ;

[0218] In the simulation experiment, the task text data was encoded using BERT and then reduced to 8 dimensions through principal component analysis, which was then mapped to 8 qubits:

[0219] ;

[0220] The weights of electrical exclusion zone data are set based on the swept surface of the wind turbine blades and the enclosure area of ​​the transformer substation: wind turbine swept surface transformer substation fence Constructing quantum-safe operators .

[0221] The quantum loop circuit integrates 30 frames of historical observations, Each qubit is divided into a three-scale subset. ,satisfy ,and :

[0222] ;

[0223] Let the initial map state be... This could be the first frame's correlated quantum state; construct a 15-qubit map register and output the quantum implicit map state density matrix: .

[0224] In the simulation experiment, the task quantum state after dimensionality reduction is received. (8 qubits), sentence-level aggregation via quantum self-attention circuit: And incorporate prior knowledge from the new energy field: The final output is the dedicated task quantum state density matrix: , .

[0225] Align the task semantics with the map location:

[0226] First, define the associative observable operator that operates only on the map register. Then, partial trace operations are performed on the joint quantum system to extract the quantum fusion operator acting on the map state subspace:

[0227] ;

[0228] in, This operator represents the partial trace of the task register; the expected value of this operator characterizes the degree of matching.

[0229] Alignment results show: Expected quantum measurement value for the #07 wind turbine tower region Tower base area transformer substation area The windward blade region is enhanced due to quantum phase interference.

[0230] Quantum fusion operator Acting on quantum hidden map states The observation quantum state is constructed, and after evolution through the quantum strategy circuit, the original action probability distribution is output based on the quantum Born rule. The final probability amplitude distribution is: 0.45 around the windward side, 0.35 maintaining a distance of 15m, and 0.20 moving in the blade gap.

[0231] Then, quantum projection constraints are applied to construct the projection operator for the swept surface of the wind turbine:

[0232] ;

[0233] The projection operator for the restricted area is implemented in the classic simulator through zeroing and renormalization of the state vector; the projection operator for the swept surface of the wind turbine satisfies... The probability amplitude of any observation pointing towards the swept surface of the wind turbine is forced to zero: This corresponds to the forbidden transition in quantum mechanics, which is physically insurmountable, ensuring that the intelligent agent will not accidentally enter the blade sweeping area.

[0234] Under the same experimental conditions, the inference latency of the classical neural network method is 120ms / frame, the alignment accuracy is 87%, and the security violation rate is 2.1%. In contrast, under the conditions of a 32-qubit classical simulator, the inference latency of this embodiment is 85ms / frame, the alignment accuracy is 93%, and the security violation rate is 0%. Therefore, this method is superior to existing classical solutions in terms of real-time performance, task alignment accuracy, and absolute security.

[0235] Please refer to Figure 2 Embodiment 2 of the present invention:

[0236] A method for inspection path decision based on quantum hidden map states, which differs from the first embodiment in that it includes step S7 after step S6.

[0237] S7. Based on the quantum fusion characteristics and the expected action value after the security constraints, generate inspection interaction text that conforms to the power dispatching specifications.

[0238] Because existing navigation models only output actions and cannot synchronously generate inspection interaction text that conforms to power dispatching specifications, they cannot be connected to the power grid dispatching automation system. Step S7 uses quantum state tomography to classicalize the quantum fusion features and combines them with the action context to generate an inspection report (including text data such as equipment status, defect alarms, safety events, and dispatching reports) that can be directly connected to the dispatching system, thus completing the inspection loop.

[0239] Specifically, S7 includes S71-S72:

[0240] S71. The quantum fusion feature is converted into a classical fusion feature through quantum state tomography.

[0241] In some embodiments, S71 specifically refers to:

[0242] Quantum fusion operator Pauli operator measurements were performed to obtain the classical fusion features:

[0243] ;

[0244] in, It is a classic fusion feature; For quantum fusion operators; The Pauli operator acts on the i-th qubit; For quantum fusion operators The number of effective qubits, ; For trace calculation; X,Y,Z To form a complete basis for single-qubit observations of the Pauli operator; To extract exist X The projection component of the direction; that is , This represents the Pauli operator vector. Quantum state tomography reconstructs quantum states through complete measurements, converting non-classical correlation information in quantum fusion features into real-number vectors that can be processed by classical neural networks, thus achieving a reverse conversion from quantum to classical interfaces.

[0245] S72. Based on the classic fusion features and the expected action value after the security constraints, generate inspection interaction text through the classic Transformer decoder.

[0246] In some embodiments, S72 specifically includes:

[0247] ;

[0248] in, This is the output of the inspection interaction text; It is a classic Transformer text decoder; It is a classic fusion feature that carries scene semantic information; The expected action value after safety constraints is output in step S63, carrying behavioral context information.

[0249] Based on the above method, the generated text not only describes the inspection scene (i.e. what was seen) but also the inspection action (i.e. what is being done). For example, the generated inspection interaction text is: "I am going around the windward side of wind turbine #07 and found a 2cm crack on the leading edge of the blade, which is in line with the power dispatching specifications."

[0250] The categories of inspection interaction text can include: equipment status, defect alarms, safety events, and dispatch reports. Furthermore, the generated inspection interaction text can be transmitted back to the power grid dispatch automation system via the dispatch communication interface to achieve an integrated system of sensing, quantum mapping, quantum alignment, quantum constrained navigation, and inspection reporting.

[0251] Embodiment 3 of the present invention:

[0252] The method in Embodiment 2 described above can be implemented using a quantum-classical hybrid end-to-end multimodal model. The inputs to this model are environmental point cloud data, task text data, and electrical exclusion zone data. After processing by a quantum coding layer, a quantum point cloud spatiotemporal encoder, a quantum semantic encoder, a quantum task-site alignment fusion module, a quantum policy-action decoder, and an inspection text generation decoder, the output is an action sequence for a safety inspection path and inspection interaction text. This model can be deployed on a quantum processor unit (QPU) or a quantum simulator, supporting amplitude encoding, parameterized quantum circuits, and quantum measurement operations. Specifically, the quantum coding layer executes S2, the quantum point cloud spatiotemporal encoder executes S3, the quantum semantic encoder executes S4, the quantum task-site alignment fusion module executes S5, the quantum policy-action decoder executes S6, and the inspection text generation decoder executes S7.

[0253] In one embodiment, a training sample set is first constructed using a digital twin wind and light field station environment. The training samples include point cloud data of the station environment, text data of inspection tasks, electrical exclusion zone data, and action data of the inspection path. Based on the training samples, a quantum point cloud spatiotemporal encoder, a quantum semantic encoder, a quantum task-station alignment and fusion module, and a quantum policy-action decoder are pre-trained to enable the model to acquire environmental awareness, task understanding, and navigation decision-making capabilities. Subsequently, the pre-trained model's parameters are fine-tuned using inspection data collected from a real wind and light field station to adapt to sensor noise, equipment occlusion, and environmental changes in the actual scenario. A safety operator is constructed according to the station's safety regulations, and a exclusion zone projection operator is generated based on the safety operator. During model training, the exclusion zone projection operator is used to constrain the parameter update process. When the parameter update direction corresponds to the exclusion zone state, the corresponding gradient component is eliminated, and only the gradient corresponding to the feasible region participates in the parameter update, thereby obtaining an inspection decision model that satisfies electrical safety constraints.

[0254] Please refer to Figure 3 Embodiment 4 of the present invention:

[0255] A patrol path decision-making system 100 based on quantum implicit map states includes a memory 101, a processor 102, and a computer program stored in the memory 101 and running on the processor 102. When the processor 102 executes the computer program, it implements the steps of any one of the embodiments in Embodiments 1 to 3. In some embodiments, the processor 102 is a quantum processor or a quantum simulator.

[0256] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for inspection path decision-making based on quantum implicit map states, characterized in that, include: Acquire environmental point cloud data, task text data, and electrical exclusion zone data from wind and solar power plants; Quantum encoding is performed on the environmental point cloud data and the task text data respectively to generate environmental quantum states and task quantum states, and quantum safety operators are constructed based on the electrical exclusion zone data; The current quantum state of the environment is quantumly entangled and fused with the historical quantum implicit map state of the previous moment to generate the current quantum implicit map state. The quantum implicit map state and the task quantum state are quantum entangled to enhance quantum entanglement, generating a quantum fusion operator to characterize the matching degree between task semantics and map position. An observation quantum state is constructed based on the quantum fusion operator and the quantum implicit map state, and the observation quantum state is subjected to quantum evolution to generate candidate inspection paths; Based on the quantum safety operator, electrical safety constraints are applied to the candidate inspection path to generate a safe inspection path; Entangling the quantum implicit map state with the task quantum state to generate a quantum fusion operator for characterizing the matching degree between the inspection task and the inspection environment includes: The task quantum state is input into a quantum self-attention circuit composed of parameterized single-qubit rotating gates and controlled NOT gates stacked alternately, so as to redistribute the entanglement of each semantic component in the task quantum state through unitary transformation to obtain the aggregated task quantum state. The aggregated task quantum state is subjected to a time evolution operation with a preset new energy Hamiltonian to map the general semantics of the task text data to new energy-specific semantics, thereby obtaining a special task quantum state. Construct a joint quantum system based on the aforementioned quantum implicit map state and the aforementioned dedicated task quantum state; The joint quantum system is input into a quantum cross-attention circuit consisting of a parameterized Pauli rotation gate and a cross-register entanglement gate, so as to entangle the quantum implicit map state with the dedicated task quantum state through a unitary transformation; Perform quantum state tomography or partial trace operation on the entangled joint quantum system to reconstruct the quantum fusion operator acting on the subspace corresponding to the quantum implicit map state.

2. The method according to claim 1, characterized in that, Quantum encoding is performed on the environmental point cloud data and the task text data respectively to generate environmental quantum states and task quantum states. A quantum safety operator is constructed based on the electrical exclusion zone data, including: The environmental point cloud data is downsampled to obtain a key point dataset; The keypoint dataset is divided into voxels at multiple spatial scales to obtain multi-scale voxel features; Based on the multi-scale voxel features, target environment features are extracted, and the target environment features are mapped to the first quantum state space through amplitude encoding to generate environment quantum states. The task text data is encoded into target text features, and the target text features are mapped to the second quantum state space through angle encoding to generate the task quantum state; Based on the electrical exclusion zone data, the preset exclusion zone space identifier is encoded into the weight coefficients of the corresponding operators of the qubits to obtain quantum safe operators.

3. The method according to claim 1, characterized in that, The quantum entanglement fusion of the current environment quantum state with the historical implicit quantum map state from the previous moment to generate the current quantum implicit map state includes: By establishing the correlation between the quantum states of the environment at different spatial scales through quantum entanglement, the correlated quantum states are obtained; The current quantum state is quantumly entangled and fused with the historical quantum implicit map state from the previous time step to obtain the current quantum implicit map state.

4. The method according to claim 3, characterized in that, By establishing the correlation between the quantum states of the environment at different spatial scales through quantum entanglement, the correlated quantum states include: The environmental quantum state of a single frame is divided into subsets of different spatial scales, and each subset is encoded as a quantum voxel feature at the corresponding spatial scale. Quantum entanglement between quantum voxel features at different spatial scales is established by quantum controlled NOT gates, thereby obtaining the correlated quantum states between quantum voxel features at different spatial scales under the said environmental quantum state.

5. The method according to claim 3, characterized in that, By fusing the current correlated quantum state with the historical implicit quantum map state from the previous moment through quantum entanglement, the current implicit quantum map state is obtained, including: The current correlated quantum state and the previous historical quantum implicit map state are input into the quantum loop circuit, and the correlated quantum state and the historical quantum implicit map state are fused through unitary transformation to obtain the current quantum implicit map state.

6. The method according to claim 1, characterized in that, Constructing a joint quantum system based on the aforementioned quantum implicit map state and the aforementioned dedicated task quantum state includes: The quantum implicit map state density matrix is ​​obtained by taking the outer product of the quantum implicit map state and its conjugate transpose. The dedicated task quantum state is outer productted with its conjugate transpose to obtain the dedicated task quantum state density matrix. The quantum implicit map state density matrix and the dedicated task quantum state density matrix are combined by tensor product to obtain a joint quantum system.

7. The method according to claim 1, characterized in that, The quantum evolution of the observed quantum state to generate candidate inspection paths includes: The observed quantum state is input into the quantum path strategy circuit so that the observed quantum state is mapped to the path action subspace through unitary evolution to obtain candidate action quantum states; The candidate action probability distribution of the quantum state of the candidate action is calculated based on the quantum Born rule, and the candidate inspection path is determined based on the candidate action probability distribution.

8. The method according to claim 7, characterized in that, Based on the quantum safety operator, electrical safety constraints are applied to the candidate inspection paths to generate safe inspection paths, including: Based on the quantum safe operator, a forbidden zone projection operator is constructed, and the forbidden zone component corresponding to the forbidden zone space in the observed quantum state is removed by the forbidden zone projection operator to obtain a safe quantum state; The secure quantum state is input into the quantum path strategy circuit so that the secure quantum state is mapped to the path action subspace through unitary evolution to obtain the secure action quantum state; The probability distribution of the safety action quantum state is calculated based on the quantum Born rule, and the safety inspection path is determined according to the probability distribution of the safety action.

9. A patrol path decision-making system based on quantum implicit map states, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements each step of the inspection path decision method based on quantum implicit map states as described in any one of claims 1 to 8.

Citation Information

Patent Citations

  • Bridge and tunnel inspection maintenance management system and method based on three-dimensional panorama and GIS map

    CN119991098A

  • Reservoir safety intelligent inspection method and system based on unmanned aerial vehicle

    CN121856984A