A hardware-aware intelligent scheduling system for quantum tasks

Abstract

This invention discloses a hardware-aware intelligent scheduling system for quantum tasks, relating to the field of quantum technology. The system includes a hardware monitoring module, a feature extraction module, a matching prediction module, and a scheduling optimization module. The hardware monitoring module collects physical parameters and constructs an instantaneous state matrix. The feature extraction module extracts resource and noise features. The matching optimization module maximizes the global expected success rate of the target quantum system. The scheduling optimization module updates the global expected success rate and the optimal task queue based on the changed instantaneous state matrix. This invention periodically collects and records key physical parameters of physical qubits and dynamically adjusts the task queue and optimizes the expected success rate in real time, enabling the quantum system to adjust according to the real-time state of the hardware. This ensures improved task success rate and overall system stability even with increased task load and resource requirements.

Description

Technical Field

[0001] This invention relates to the field of quantum technology, and more specifically to a hardware-aware intelligent scheduling system for quantum tasks. Background Technology

[0002] In quantum computing systems, quantum task scheduling, a key component of the system software stack, is responsible for managing user-submitted quantum computing tasks. This includes mapping quantum circuits to physical qubit resources and determining the execution order of tasks, thereby achieving the rational utilization of quantum computing resources. However, quantum computing environments differ significantly from traditional classical computing environments in terms of hardware characteristics and task execution mechanisms, making it difficult to directly apply existing classical computing task scheduling strategies to quantum computing systems.

[0003] Existing quantum task scheduling methods mostly employ static scheduling strategies or make decisions based on simple heuristics, such as sorting and allocating tasks solely based on indicators like task length, task priority, or quantum circuit size, without fully considering the dynamic changes in the quantum hardware's operating state and the interactions between tasks. Due to the lack of real-time awareness and modeling capabilities of the hardware state, these scheduling methods typically cannot dynamically adjust the task queue according to hardware state changes, and are also ill-equipped to effectively avoid noise interference and resource contention issues in multi-tasking scenarios.

[0004] As the scale of quantum computing systems continues to expand, the number of physical qubits and concurrent tasks continues to increase. Existing scheduling methods are gradually revealing their shortcomings in terms of scheduling efficiency, task execution success rate, and overall system stability, making it difficult to meet the high reliability and high resource utilization requirements of large-scale quantum computing systems. Therefore, it is necessary to propose a quantum task scheduling technique that can adapt to the dynamic state changes of quantum hardware and take into account multi-task interference and resource competition, in order to overcome the aforementioned shortcomings of existing quantum technologies. Summary of the Invention

[0005] This invention provides a hardware-aware intelligent scheduling system for quantum tasks, which solves the problem that the scheduling efficiency and success rate of quantum tasks in existing quantum computing systems decrease as the system scale increases.

[0006] This invention is achieved through the following technical solution: A hardware-aware intelligent scheduling system for quantum tasks, comprising a hardware monitoring module, a feature extraction module, a matching prediction module, and a scheduling optimization module; The hardware monitoring module is configured to periodically collect physical parameters recorded along the time axis for each physical qubit in the target quantum system. The physical parameters include at least relaxation time, decoherence time, and single quantum gate fidelity. Based on the relaxation time, decoherence time, and single quantum gate fidelity, an instantaneous state matrix is ​​constructed to represent the hardware state environment. The feature extraction module is configured to extract resource features and noise features for each scheduled task in the target quantum system, and construct a task feature vector for each scheduled task using the resource features and noise features. The resource features include the number of qubits required, the depth of the quantum circuit, and the number of dual quantum gates. The noise features include the noise sensitivity value. The matching optimization module is configured to: use a machine learning model to generate the individual expected success rate of each task to be scheduled and the current hardware state environment based on the task feature vector and the instantaneous state matrix; calculate the global expected success rate of the target quantum system based on the maximum of all individual expected success rates; and mark the current global expected success rate as the global critical value. The scheduling optimization module is configured as follows: when the instantaneous state matrix changes in subsequent time steps and the change reaches a preset state threshold, and the global expected success rate at the current time step is lower than the global critical value, the global expected success rate is recalculated based on the new instantaneous state matrix after the change, and the expected success rate is generated and maximized to update the optimal task queue.

[0007] Furthermore, let the relaxation time be represented as T1. i (t), the decoherence time is denoted as T2. i (t), the single-quantum gate fidelity is expressed as F. g i (t), The instantaneous state matrix is ​​then expressed in the following form: .

[0008] Furthermore, a hardware health score is set based on the instantaneous state matrix, and the state threshold is set for the hardware health score; the hardware health score is represented as H. system (t), then the hardware health score is expressed as: , Where w1, w2, and w3 are weighting coefficients, and satisfy w1 + w2 + w3 = 1; T1 in the formula max T2 is the preset maximum reference relaxation time. max This is the preset reference maximum value for the decoherence time.

[0009] Furthermore, let the master order number of the task to be scheduled be k; let the task feature vector be denoted as F, and the task feature vector also include the estimated execution time of the task to be scheduled, denoted as T. est Let Q represent the number of qubits required, D represent the depth of the quantum circuit, C represent the number of two quantum gates, and S represent the noise sensitivity value. The task feature vector is then represented as: , Among them, F k T est k Q k D k C k S k , respectively representing the task feature vector, estimated execution time, number of qubits required, quantum circuit depth, number of two quantum gates, and noise sensitivity value of the k-th task to be scheduled.

[0010] Furthermore, let the individual's expected success rate be represented by P. succ Let the machine learning model be denoted as ε, then the individual expected success rate of the k-th task to be scheduled is denoted as P. succ k , The relationship between individual expected success rate and other factors can be expressed as: P succ k =ε{F k ,H(t)}.

[0011] Furthermore, let M represent the matching score between the task to be scheduled and the hardware state environment. Then, the matching score of the k-th task to be scheduled is represented as M. k And let M k =P succ k Let there be L tasks to be scheduled. Let Z be the objective function for calculating the global expected success rate, and let λ be the weight coefficient of the interference penalty term. The formula for maximizing the objective function is then expressed as: , Where x represents a binary decision variable, and x∈{0,1}; i and j represent the interference analysis ordinal numbers of the task to be scheduled, and I ij This represents the interference coefficient between the i-th and j-th tasks to be scheduled; The x i and x j This indicates whether the i-th and j-th tasks to be scheduled are in the current execution queue. When x is 0, x... i and x jThis indicates that the corresponding task to be scheduled is not in the current execution queue. When x is 1, x... i and x j This indicates that the corresponding task to be scheduled is in the current execution queue.

[0012] Furthermore, the physical parameters also include the crosstalk error T between physical qubits. pq , where p and q represent the ordinal numbers of the physical qubits.

[0013] Furthermore, constraints are set for the target quantum system: the total number of physical bits occupied by all scheduled tasks cannot exceed the total number of sub-bits of the system.

[0014] Furthermore, let the total number of quantum bits in the target quantum system be expressed as N. total The specific constraint conditions are then set as follows: .

[0015] Furthermore, the state threshold is set as H. min Let Δt be the time difference between the moment when the hardware health score changes and the moment when the physical parameters were last collected in the previous cycle, and let H be the hardware health score of the k-th task to be scheduled at time t. k system (t); when H k system (t) <H min When the current queue is paused, the latest H is used as the buffer. k system With (t+Δt) as input, rerun the matching degree prediction and scheduling optimization process to generate a new optimal task queue.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. By periodically collecting and recording the key physical parameters of physical qubits, the scheduling system can adjust according to the real-time status of the hardware, making more efficient use of qubit resources and dynamically responding to hardware instability and changes, reducing resource conflicts and unnecessary resource idleness, thereby improving scheduling efficiency. 2. Extract the resource and noise features of each task to be scheduled, and match the task features with the instantaneous state matrix of the hardware state through a machine learning model. By calculating the global expected success rate and comparing it with the global critical value, the system can continuously optimize the overall task execution success rate in multi-task scheduling. 3. By rearranging the task execution order according to the real-time hardware status and dynamically adjusting the task queue and optimizing the expected success rate in real time, the system can allocate and utilize quantum computing resources more efficiently, minimize resource conflicts and noise interference, thereby ensuring that the success rate of tasks and the overall stability of the system are improved when the workload and resource requirements increase. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural block diagram of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention. Example

[0019] like Figure 1 As shown, this embodiment is a hardware-aware quantum task intelligent scheduling system, which includes a hardware monitoring module, a feature extraction module, a matching prediction module, and a scheduling optimization module. The hardware monitoring module is configured to periodically collect physical parameters recorded along the time axis for each physical qubit in the target quantum system. The physical parameters include at least relaxation time, decoherence time, and single quantum gate fidelity. Based on the relaxation time, decoherence time, and single quantum gate fidelity, an instantaneous state matrix is ​​constructed to represent the hardware state environment. The feature extraction module is configured to extract resource features and noise features for each scheduled task in the target quantum system, and construct a task feature vector for each scheduled task using the resource features and noise features. The resource features include the number of qubits required, the depth of the quantum circuit, and the number of dual quantum gates. The noise features include the noise sensitivity value. The matching optimization module is configured to: use a machine learning model to generate the individual expected success rate of each task to be scheduled and the current hardware state environment based on the task feature vector and the instantaneous state matrix; calculate the global expected success rate of the target quantum system based on the maximum of all individual expected success rates; and mark the current global expected success rate as the global critical value. The scheduling optimization module is configured as follows: when the instantaneous state matrix changes in subsequent time steps and the change reaches a preset state threshold, and the global expected success rate at the current time step is lower than the global critical value, the global expected success rate is recalculated based on the new instantaneous state matrix after the change, and the expected success rate is generated and maximized to update the optimal task queue.

[0020] The relaxation time refers to the time it takes for a qubit to return from an excited state (high-energy state) to its ground state (low-energy state). This is the process by which a qubit loses its quantum information after interacting with its environment. Specifically, relaxation is the process by which a qubit gradually relinquishes energy and returns to a lower-energy state when it is in a higher-energy state without external interference. A shorter relaxation time means that the qubit's information will be lost quickly, which limits the operation time window in quantum computing and reduces the stability and reliability of quantum computing. The longer the relaxation time, the longer the qubit can maintain its quantum state, thus allowing for more computational steps to be performed. The decoherence time is the time it takes for a qubit to maintain its coherence (i.e., quantum superposition state) without being disturbed by the environment. In quantum computing, the state of a qubit is usually a superposition state, in which multiple states exist simultaneously. However, this state is very fragile and easily affected by environmental noise, leading to decoherence. The decoherence time measures the time it takes for a qubit to maintain its quantum properties before its state becomes decoherent (loses its quantum superposition state). A shorter decorrelation time means that the quantum properties of a qubit are lost earlier, and the superposition state in quantum computing disappears rapidly, thus affecting the accuracy and reliability of the calculation results. A longer decorrelation time helps the quantum algorithm run more accurately. In this embodiment, the single quantum gate fidelity is an indicator used to measure circuit complexity. The higher the fidelity, the more accurate the execution of the quantum gate and the lower the error rate of quantum computing. A lower fidelity means that there are large errors in the quantum gate operation, which may lead to incorrect calculation results and affect the accuracy of the entire quantum computing. The instantaneous state matrix is ​​a two-dimensional or multi-dimensional data structure composed of the key physical parameters of each physical qubit, recording the hardware state of the quantum computer at a certain moment. As time changes, the state parameters of each qubit may change, and the values ​​in the matrix will be adjusted accordingly. This dynamic adjustment enables the quantum task scheduling system to optimize according to the real-time state of the current hardware environment, ensuring the success rate of task execution and the maximum utilization of system resources.

[0021] The qubit requirement refers to the number of physical qubits required to perform a specific quantum task. A qubit is the basic unit of quantum computing, similar to a bit in classical computing, but unlike a classical bit, a qubit can simultaneously represent multiple states (i.e., a superposition state). The qubit requirement reflects the resource requirements of a quantum computing task, i.e., how many qubits must participate in the computation to execute the task. The quantum circuit depth refers to the number of layers of quantum gates in a quantum circuit, usually expressed as the time-series depth of the quantum gate operations. A larger quantum circuit depth means that the quantum computing task needs to perform more quantum operation steps, usually corresponding to higher computational complexity. The number of two quantum gates refers to the number of operation gates involving two qubits in a quantum circuit. Two quantum gates are common quantum operations in quantum computing, often used to create entanglement between qubits. Two quantum gate operations are among the most common and complex quantum gates in quantum computing, involving interactions between qubits. The noise sensitivity value measures the responsiveness of a specific task or quantum circuit to changes in quantum hardware noise, reflecting the sensitivity of the task to a decrease in the probability of successful execution due to hardware noise interference during quantum computing. A high noise sensitivity value for a task means that it is highly sensitive to hardware noise during quantum computing. For example, some quantum circuits may involve complex quantum gate operations and long-duration qubit operations, making them susceptible to quantum decoherence and noise, leading to a high error rate. These tasks require hardware environments with low noise and high stability to ensure a high success rate. Conversely, a low noise sensitivity means that the task is less affected by hardware noise. Even with strong hardware noise, the task can still maintain a high success rate. For example, some tasks may only involve simple quantum gate operations or short-duration qubit operations, making them more tolerant of noise.

[0022] The machine learning model is trained based on task feature vectors and instantaneous state matrices. Its purpose is to learn the complex relationship between task features and hardware states through historical data, thereby predicting the success rate of task execution under the current hardware conditions. In this embodiment, the machine learning model is a direct application of conventional techniques; the model can be any available machine learning model, such as Bayesian optimization, gradient descent, least squares, etc. For each task to be scheduled, the model outputs an individual expected success rate, i.e., the predicted execution success rate of the task under the current hardware environment. The machine learning model predicts the individual expected success rate of each task based on its feature vector and the current hardware state. Through this prediction, the system can schedule tasks more intelligently, improving the success rate of task execution, especially when facing large-scale quantum computing tasks, enabling more efficient and reasonable resource allocation and task scheduling decisions. The global expected success rate refers to the overall system success rate calculated based on the individual expected success rates of all tasks. To ensure the efficient operation of the entire quantum computing system, tasks need to be rationally sorted and resources allocated according to the expected success rate of each task, thereby optimizing the overall system's task execution success rate. The global expected success rate is a comprehensive expectation of the successful execution of all tasks under the current hardware environment. The global threshold is a benchmark for the current system, representing the lowest acceptable success rate for all tasks under the current hardware conditions. During scheduling, the system attempts to maximize the overall success rate of all tasks by comprehensively considering the characteristics of all tasks and the hardware state. In this embodiment, the global threshold is the initially calculated global expected success rate, ensuring that subsequent global expected success rates do not fall below this value, keeping the target quantum system in a state where efficiency does not decline. Since the calculated global expected success rate has already been maximized, the initial calculated value is not too low, effectively serving as the global threshold to support the efficiency state screening of the quantum system.

[0023] A change in the instantaneous state matrix indicates that the hardware state differs from its previous state at a given moment, potentially due to decoherence of the qubits, noise variations, hardware failures, or other external factors. The magnitude of the change refers to the amplitude of the hardware state change, such as variations in relaxation or decoherence time. When the hardware state changes, if this change is large enough to exceed a preset state threshold, the hardware's operating environment is considered to have changed significantly, potentially requiring a reassessment of the current task scheduling. This state threshold can be set based on empirical rules and historical data. Based on the optimized task scheduling strategy, the system updates the optimal task queue. This means that the current task queue will be reordered or adjusted to adapt to the new hardware state and optimized scheduling. If both a change in the instantaneous state matrix and a global expected success rate below the global critical value are simultaneously satisfied, it means the task execution success rate under the current hardware state is too low, and the system needs to optimize the scheduling to ensure an improved task success rate. If the instantaneous state matrix remains unchanged, but the global expected success rate falls below the global critical value, it is determined that the current hardware state environment of the target quantum system does not affect task processing efficiency; the efficiency reduction may be due to other external factors.

[0024] Furthermore, as a feasible implementation method, let the relaxation time be represented as T1. i (t), the decoherence time is denoted as T2. i (t), the single-quantum gate fidelity is expressed as F. g i (t), The instantaneous state matrix is ​​then expressed in the following form: .

[0025] By introducing an instantaneous state matrix, quantum task scheduling systems can comprehensively and dynamically capture changes in hardware state. This matrix not only provides detailed information for hardware state modeling but also offers strong support for task scheduling, helping to optimize task success rate and resource utilization. By integrating the state information of different qubits, the instantaneous state matrix can reflect the real-time performance of the entire quantum computing system. Based on this matrix, dynamic adjustments to task scheduling can be made, such as selecting appropriate qubits to execute specific tasks and avoiding executing sensitive tasks on unstable qubits, thereby optimizing task execution success rate and efficiency.

[0026] As a feasible implementation, a hardware health score is set based on the instantaneous state matrix, and the state threshold is set according to the hardware health score; the hardware health score is represented as H. system (t), then the hardware health score is expressed as: , Where w1, w2, and w3 are weighting coefficients, and satisfy w1 + w2 + w3 = 1; T1 in the formula max T2 is the preset maximum reference relaxation time. max This is the preset reference maximum value for the decoherence time.

[0027] In the formula The term represents the ratio of the relaxation time of the i-th qubit to the reference maximum relaxation time. A longer relaxation time results in better state retention of the qubit and higher hardware stability; therefore, this ratio should be as close to 1 as possible. In the formula... This represents the ratio of the decoherence time of the i-th qubit to the reference maximum decoherence time. A longer decoherence time means the qubit maintains coherence for a longer period, resulting in better system stability; therefore, a ratio closer to 1 is preferable. The weighting coefficients represent the importance of different hardware state parameters, and their sum must be 1. The single-quantum gate fidelity F... g i The higher the value of (t), the smaller the error in quantum gate operations, and the higher the reliability and accuracy of task execution. Since fidelity is a value that varies within the range [0,1], the closer the fidelity is to 1, the better the quality of the qubit. The weighting coefficients can be adjusted according to the sensitivity of the task to various hardware characteristics. If the task is particularly sensitive to the relaxation time of the qubit, w1 can be given a larger weight; if the task is more sensitive to the decoherence time, more weight can be allocated to w2; if the quantum gate fidelity is more important to the success rate of task execution, then the weight of w3 should be higher.

[0028] Furthermore, as a feasible implementation, let the master order number of the task to be scheduled be k; let the task feature vector be represented by F, and the task feature vector also include the estimated execution time of the task to be scheduled, the estimated execution time being represented by T. est Let Q represent the number of qubits required, D represent the depth of the quantum circuit, C represent the number of two quantum gates, and S represent the noise sensitivity value. The task feature vector is then represented as: , Among them, F k T est k Q k D k C k S k , respectively representing the task feature vector, estimated execution time, number of qubits required, quantum circuit depth, number of two quantum gates, and noise sensitivity value of the k-th task to be scheduled.

[0029] The successful execution of quantum computing tasks depends on hardware resources such as qubits, quantum circuits, and quantum gates. The number of qubits required and the depth of quantum circuits in the task's feature vector help the scheduling system accurately understand the hardware resources needed for the task, thus avoiding excessive competition or waste during resource allocation. The system can determine the computational complexity of the task based on the quantum circuit depth and the number of dual quantum gates. The noise sensitivity value indicates the task's sensitivity to noise in quantum computing.

[0030] Furthermore, as a feasible implementation method, the individual's expected success rate is denoted as P. succ Let the machine learning model be denoted as ε, then the individual expected success rate of the k-th task to be scheduled is denoted as P. succ k , The relationship between individual expected success rate and other factors can be expressed as: P succ k =ε{F k ,H(t)}.

[0031] Furthermore, if we set a matching score M representing the degree of matching between the task to be scheduled and the hardware state environment, then the matching score of the k-th task to be scheduled is represented as M. k And let M k =P succ k Let there be L tasks to be scheduled. Let Z be the objective function for calculating the global expected success rate, and let λ be the weight coefficient of the interference penalty term. The formula for maximizing the objective function is then expressed as: , Where x represents a binary decision variable, and x∈{0,1}; i and j represent the interference analysis ordinal numbers of the task to be scheduled, and I ij This represents the interference coefficient between the i-th and j-th tasks to be scheduled; The x i and x j This indicates whether the i-th and j-th tasks to be scheduled are in the current execution queue. When x is 0, x... i and x j This indicates that the corresponding task to be scheduled is not in the current execution queue. When x is 1, x... i and x j This indicates that the corresponding task to be scheduled is in the current execution queue.

[0032] The formula for individual expected success rates represents the prediction of each task's execution success rate using a machine learning model, based on a combination of task feature vectors and real-time hardware status. This provides crucial input data for subsequent scheduling optimization. The matching score reflects the suitability of the k-th task under the current hardware status, i.e., the matching degree between the task and hardware resources. This matching score will be used for subsequent global expected success rate calculation, forming the basis for scheduling optimization. In the formula... This part represents the interference situation when task i and task j are executed at the same time. If x i =x j =1 indicates that task i and task j are selected to be executed in the same queue, which may cause interference. In this case, the interference needs to be penalized. The interference coefficient I ij This is used to quantify the degree of conflict between two tasks; the greater the interference, the stronger the penalty. The objective function achieves global optimization of task scheduling by maximizing the task matching score and minimizing the interference penalty term. The binary decision variable x represents whether to arrange the k-th task to be scheduled into the current execution queue. The value of this decision variable is 0 or 1, where 0 indicates that the task is not scheduled and 1 indicates that the task is scheduled. By adjusting these decision variables, the system can make optimal scheduling decisions based on factors such as hardware resource status, task characteristics, and interference.

[0033] Furthermore, as a feasible implementation, the physical parameters also include the crosstalk error T between physical qubits. pq Where p and q represent the ordinal numbers of physical qubits; constraints are set for the target quantum system: the total number of physical bits occupied by all scheduled tasks cannot exceed the total number of qubits in the system.

[0034] In quantum computing, crosstalk error T pq This represents the interference or coupling error between two physical qubits p and q. Due to the interaction between qubits, especially in multi-qubit operations, this can lead to computational errors or task instability. Quantum computing tasks often require multiple physical qubits to work together, and crosstalk errors can affect the accuracy of information transfer and quantum gate operations between these qubits. If the crosstalk error between two physical qubits is large, it may cause task failure or errors. Therefore, when scheduling tasks, it is necessary to consider the crosstalk error between physical qubits and avoid scheduling tasks with significant mutual influence on these qubits simultaneously.

[0035] Specifically, let the total number of quantum bits in the target quantum system be expressed as N. total The specific constraint conditions are then set as follows: .

[0036] This constraint ensures that the system's resource utilization does not exceed the upper limit of the number of physical qubits, avoiding excessive resource consumption or competition. It guarantees that the total number of physical qubits occupied by scheduled tasks will not exceed the total number of physical qubits in the quantum computing system, thus effectively managing and allocating quantum computing resources. By considering this constraint during scheduling, the scheduling algorithm ensures that each physical qubit resource in the system is allocated to task execution rationally and efficiently. If the number of physical qubits is over-occupied, the system may experience task conflicts, performance degradation, or execution failures. This constraint effectively avoids this situation, ensuring that task scheduling is performed within the system's resource capacity, thereby improving system reliability and task execution success rate. If the current total number of physical qubits exceeds N... total The system will then dynamically adjust the task queue to avoid over-allocating physical bits and ensure that the constraints are met.

[0037] Furthermore, as a feasible implementation method, the state threshold is set as H. min Let Δt be the time difference between the moment when the hardware health score changes and the moment when the physical parameters were last collected in the previous cycle, and let H be the hardware health score of the k-th task to be scheduled at time t. k system (t); when H k system (t) <H min When the current queue is paused, the latest H is used as the buffer. k system With (t+Δt) as input, rerun the matching degree prediction and scheduling optimization process to generate a new optimal task queue.

[0038] The time difference Δt is defined as the time interval between the change in the hardware health score and the last time physical parameters were collected periodically. By periodically collecting the hardware's physical parameters and calculating the health score, the system can monitor changes in the hardware status in real time. If the health score changes significantly, the system will determine whether the task scheduling strategy needs to be readjusted. The time difference Δt represents the time interval for updating the hardware health score. Within this time difference Δt, the hardware health may change, affecting the task's execution environment, thus requiring dynamic adjustment of the scheduling strategy. When the hardware health score H of the k-th task to be scheduled... k system (t) decreases to H min At this point, the current task queue will be paused, and the updated hardware health score H corresponding to the k-th task to be scheduled will be used. k system (t+Δt) Re-optimize task scheduling.

[0039] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A hardware-aware intelligent scheduling system for quantum tasks, comprising a hardware monitoring module, a feature extraction module, a matching prediction module, and a scheduling optimization module, characterized in that: The hardware monitoring module is configured to periodically collect physical parameters recorded along the time axis for each physical qubit in the target quantum system. The physical parameters include at least relaxation time, decoherence time, and single quantum gate fidelity. Based on the relaxation time, decoherence time, and single quantum gate fidelity, an instantaneous state matrix is ​​constructed to represent the hardware state environment. The feature extraction module is configured to extract resource features and noise features for each scheduled task in the target quantum system, and construct a task feature vector for each scheduled task using the resource features and noise features. The resource features include the number of qubits required, the depth of the quantum circuit, and the number of dual quantum gates. The noise features include the noise sensitivity value. The matching optimization module is configured to: use a machine learning model to generate the individual expected success rate of each task to be scheduled and the current hardware state environment based on the task feature vector and the instantaneous state matrix; calculate the global expected success rate of the target quantum system based on the maximum of all individual expected success rates; and mark the current global expected success rate as the global critical value. The scheduling optimization module is configured as follows: when the instantaneous state matrix changes in subsequent time steps and the change reaches a preset state threshold, and the global expected success rate at the current time step is lower than the global critical value, the global expected success rate is recalculated based on the new instantaneous state matrix after the change, and the expected success rate is generated and maximized to update the optimal task queue.

2. The hardware-aware quantum task intelligent scheduling system according to claim 1, characterized in that, ... The relaxation time is represented as T1. i (t), the decoherence time is denoted as T2. i (t), the single-quantum gate fidelity is expressed as F(t), g i (t), The instantaneous state matrix is ​​then expressed in the following form: .

3. The hardware-aware quantum task intelligent scheduling system according to claim 2, characterized in that, A hardware health score is set based on the instantaneous state matrix, and the state threshold is set based on the hardware health score; the hardware health score is represented as H. system (t), then the hardware health score is expressed as: ， Where w1, w2, and w3 are weighting coefficients, and satisfy w1 + w2 + w3 = 1; T1 in the formula max T2 is the preset maximum reference relaxation time. max This is the preset reference maximum value for the decoherence time.

4. A hardware-aware quantum task intelligent scheduling system according to claim 2, characterized in that, ... The master order number of the task to be scheduled is k; let the task feature vector be represented by F, and the task feature vector also includes the estimated execution time of the task to be scheduled, which is represented by T. est Let Q represent the number of qubits required, D represent the depth of the quantum circuit, C represent the number of two quantum gates, and S represent the noise sensitivity value. The task feature vector is then represented as: , Among them, F k T est k Q k D k C k S k , respectively representing the task feature vector, estimated execution time, number of qubits required, quantum circuit depth, number of two quantum gates, and noise sensitivity value of the k-th task to be scheduled.

5. A hardware-aware quantum task intelligent scheduling system according to claim 4, characterized in that, Let P represent the individual's expected success rate. succ Let the machine learning model be denoted as ε, then the individual expected success rate of the k-th task to be scheduled is denoted as P. succ k , The relationship between individual expected success rate and other factors can be expressed as: P succ k =ε{F k ,H(t)}.

6. A hardware-aware quantum task intelligent scheduling system according to claim 5, characterized in that, Let M represent the matching score between the task to be scheduled and the hardware state environment. Then, the matching score of the k-th task to be scheduled is represented as M. k And let M k =P succ k Let there be L tasks to be scheduled. Let Z be the objective function for calculating the global expected success rate, and let λ be the weight coefficient of the interference penalty term. The formula for maximizing the objective function is then expressed as: ， Where x represents a binary decision variable, and x∈{0,1}; i and j represent the interference analysis ordinal numbers of the task to be scheduled, and I ij This represents the interference coefficient between the i-th and j-th tasks to be scheduled; The x i and x j This indicates whether the i-th and j-th tasks to be scheduled are in the current execution queue. When x is 0, x... i and x j This indicates that the corresponding task to be scheduled is not in the current execution queue. When x is 1, x... i and x j This indicates that the corresponding task to be scheduled is in the current execution queue.

7. A hardware-aware quantum task intelligent scheduling system according to claim 5, characterized in that, The physical parameters also include the crosstalk error T between physical qubits. pq , where p and q represent the ordinal numbers of the physical qubits.

8. A hardware-aware quantum task intelligent scheduling system according to claim 6, characterized in that, Constraints are set for the target quantum system: the total number of physical bits occupied by all scheduled tasks cannot exceed the total number of sub-bits of the system.

9. A hardware-aware quantum task intelligent scheduling system according to claim 8, characterized in that, ... The total number of quantum bits in the target quantum system is expressed as N total The specific constraint conditions are then set as follows: .

10. A hardware-aware quantum task intelligent scheduling system according to claim 3, characterized in that, The state threshold is set as H. min Let Δt be the time difference between the moment when the hardware health score changes and the moment when the physical parameters were last collected in the previous cycle, and let H be the hardware health score of the k-th task to be scheduled at time t. k system (t); when H k system (t)< H min When the current queue is paused, the latest H is used as the reference. k system With (t+Δt) as input, rerun the matching degree prediction and scheduling optimization process to generate a new optimal task queue.

Images