Quantum computing task scheduling method and system, storage medium and electronic equipment

Abstract

The invention discloses a quantum computing task scheduling method and system, a storage medium and electronic equipment. The quantum computing task scheduling method comprises the steps that coding information corresponding to a scheduling task is obtained; based on the coding information, screening out a first computing task meeting a first preset priority, and based on a computing node state of the computing cluster, performing grading processing on the first computing task according to a time sequence; based on the coding information, screening out a second computing task meeting a second preset priority, matching out a corresponding computing node in an idle state from the computing cluster, and processing the second computing task; and matching execution is performed according to the priority and the computing nodes of the quantum computing cluster, and the idle computing nodes are used for matching the second computing task to perform scheduling execution, so that timeout caused by long-term scheduling incapability of some tasks can be prevented, computing power resources of the quantum computing cluster are utilized under a relatively fair condition, and the scheduling efficiency of the quantum computing cluster is improved. The overall utilization rate of computing power resources is improved.

Description

Technical Field

[0001] The present invention relates to the field of quantum computer technology, and in particular to a quantum computing task scheduling method, scheduling system, storage medium and electronic equipment. Background Art

[0002] Quantum computing, an emerging computing paradigm, is gaining global attention due to its advantages in solving complex problems and parallel computing. Compared to traditional electronic computers, quantum computing demonstrates tremendous potential for solving specific complex problems, particularly for tasks requiring high levels of parallelism. Efficiently allocating and utilizing the resources of quantum computing clusters has become a current research hotspot. Quantum computing tasks are highly complex, placing extremely stringent demands on the type of quantum computer equipment.

[0003] Existing quantum computing task scheduling methods are designed for heterogeneous quantum computing tasks. They are based on the earliest idle quantum computing node and whether the node meets the basic requirements of the task. However, they ignore whether the device type of the quantum computing node matches the task requirements. They lack a clear task selection mechanism and implement scheduling solely through trial and error, resulting in poor overall scheduling performance. This limitation is significantly insufficient when addressing the scheduling needs of heterogeneous quantum computing clusters and fails to fully consider the inherent characteristics of quantum computing tasks and the heterogeneity of quantum computing clusters. For example, quantum computing tasks have strict requirements on the quantum computer device type and hardware and software configuration. Tasks may have dependencies (the results of some tasks need to serve as inputs for subsequent tasks), and some tasks may have higher priority and real-time requirements. These key factors are not taken into account during the scheduling process. Existing quantum computing task scheduling methods lack an effective solution for situations where the requirements of the quantum computing task cannot be met during scheduling, further limiting their applicability and efficiency in complex quantum computing environments. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and to propose a quantum computing task scheduling method, scheduling system, storage medium and electronic device.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a quantum computing task scheduling method, comprising:

[0006] Obtaining encoding information corresponding to the scheduling task, wherein the encoding information is obtained by encoding the waiting queue of the quantum computing task and the state information of the quantum cluster;

[0007] Based on the coding information, first computing tasks meeting a first preset priority are screened out, and based on the computing node status of the computing cluster, the first computing tasks are hierarchically processed in a time sequence;

[0008] Based on the coding information, a second computing task that meets a second preset priority is screened out, and a corresponding computing node in an idle state is matched from the computing cluster to process the second computing task.

[0009] As a further description of the above technical solution: the method for screening out the first computing task that meets the first preset priority includes:

[0010] Performing feature extraction on the encoded information to obtain feature parameters of the quantum computing task, wherein the feature parameters include a priority identifier of the quantum computing task;

[0011] The first computing tasks meeting the first preset priority are screened out based on the priority identifier, the first computing tasks including first-category computing tasks and second-category computing tasks, and the first-category computing tasks have a higher priority than the second-category computing tasks.

[0012] As a further description of the above technical solution: the first computing task includes a first category computing task and a second category computing task, the first category computing task has a higher priority than the second category computing task, and the method for hierarchically processing the first computing task according to the time sequence includes:

[0013] According to the idle status of the computing nodes of the computing cluster, computing nodes are matched for the first type of computing tasks first; for the remaining unmatched computing nodes of the first type of computing tasks or the second type of computing tasks, corresponding computing nodes in a non-idle state are matched and reserved according to priority.

[0014] As a further description of the above technical solution: the method for screening out the second computing task that meets the second preset priority includes:

[0015] Acquiring characteristic parameters of the quantum computing task, wherein the characteristic parameters include a priority identifier of the quantum computing task;

[0016] The second computing tasks meeting the second preset priority are screened out based on the priority identifier, where the second preset priority is lower than the first preset priority.

[0017] As a further description of the above technical solution: a method for matching a corresponding computing node in an idle state from the computing cluster and processing the second computing task includes:

[0018] According to the computing power requirement of the second computing task, an idle computing node with corresponding computing power requirement is matched in the computing cluster to perform the calculation.

[0019] As a further description of the above technical solution: the scheduling task is updated, and the updating method includes:

[0020] Establishing a reward function related to the characteristic parameters and adjusting the weights of the characteristic parameters to adjust the priority of the quantum computing task;

[0021] After the quantum computing task is scheduled, the reward brought by scheduling the quantum computing task is calculated through the reward function, and the scheduled task is updated in real time according to the sum of the several rewards obtained.

[0022] Also included is a quantum computing task scheduling system, which is applicable to the scheduling method described in any one of the above technical solutions, including:

[0023] The encoding module is used to encode the waiting queue of the quantum computing task and the state information of the quantum cluster to obtain encoding information and form the feature vector of the corresponding scheduling task;

[0024] A neural network module generates a scheduling strategy based on the feature vector, including screening out a first computing task that meets a first preset priority; and

[0025] Screening out a second computing task that meets a second preset priority;

[0026] a scheduling module, configured to perform task scheduling according to the scheduling policy, including hierarchically processing the first computing task in a time sequence based on the computing nodes of the computing cluster according to the first computing task; and

[0027] According to the second computing task, a corresponding computing node in an idle state is matched from the computing cluster, and the second task is processed.

[0028] As a further description of the above technical solution: it also includes: a dynamic adjustment module, which establishes a reward function for the characteristic parameters of the quantum computing task, adjusts the priority of the quantum computing task, calculates the reward brought by scheduling the quantum computing task through the reward function, and updates the scheduled task in real time according to the sum of several rewards obtained.

[0029] As a further description of the above technical solution: the neural network module includes a first-level neural network and a second-level neural network, the first-level neural network is used to screen out first computing tasks that meet a first preset priority, and the second-level neural network is used to screen out second computing tasks that meet a second preset priority.

[0030] As a further description of the above technical solution: the first computing task is to decide whether the quantum task is to be executed immediately or reserved for execution; the second computing task is responsible for performing the backfill of the quantum computing task.

[0031] As a further description of the above technical solution: the architecture of the first-level neural network and the second-level neural network are the same;

[0032] The first-level neural network or the second-level neural network includes an embedding layer for receiving the feature vector;

[0033] An output end of the embedding layer is connected to at least one or more encoder units to perform feature extraction on the feature vector;

[0034] The output end of the encoder unit is connected to a calculation unit, and the calculation unit performs weight calculation on the extracted characteristic parameters to determine the priority identifier of the quantum computing task.

[0035] It also includes a computer-readable storage medium storing a computer program for running the scheduling method, wherein the computer program enables a computer to execute the scheduling method as described in any one of the above technical solutions.

[0036] Also included is an electronic device comprising:

[0037] one or more processors; memory; and

[0038] One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the scheduling method described in any one of the above technical solutions.

[0039] The above technical solution has the following advantages or beneficial effects:

[0040] 1. Quantum computing tasks are screened based on the encoded information and preset priority, and matched and executed according to the computing nodes of the quantum computing cluster through the first-level neural network and the second-level neural network, so that the quantum computing tasks are more accurately adapted to the quantum computing cluster.

[0041] 2. When scheduling quantum computing tasks, they are matched and executed according to the priority and computing nodes of the quantum computing cluster. Idle computing nodes are used to match the second computing task for scheduling and execution. When there are dependencies between quantum computing tasks, this can prevent certain tasks from timeouts due to long-term inability to schedule, thereby affecting the execution of subsequent quantum computing tasks. It can also utilize the computing power resources of the quantum computing cluster in a relatively fair manner, improve the overall utilization rate of computing power resources, and meet the real-time requirements of quantum computing tasks.

[0042] 3. By obtaining the waiting queue information of quantum computing tasks and the status information of the quantum computing cluster, the quantum computing tasks are processed. By establishing a reward function, the priority of the quantum computing tasks is adjusted, the priority corresponding to each quantum computing task is determined, and then scheduling is performed. The scheduling of quantum computing tasks is dynamically adjusted, and flexible adjustments can be made in complex quantum computing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 Flowchart of the scheduling method proposed by the present invention;

[0045] Figure 2 A flowchart of a method for screening out a first computing task that meets a first preset priority in the present invention;

[0046] Figure 3 A flowchart of a method for screening out a second computing task that meets a second preset priority in the present invention;

[0047] Figure 4 A flowchart of a method for updating a scheduling task in the present invention;

[0048] Figure 5 This is a schematic diagram of the structure of the scheduling system proposed by the present invention;

[0049] Figure 6 This is a structural principle diagram of the neural network module in the present invention.

[0050] Legend:

[0051] 1. Data processing module; 2. Neural network module; 21. Embedding layer; 22. Encoder unit; 23. Computing unit; 3. Scheduling module; 4. Dynamic adjustment module. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] Reference Figure 1, an embodiment of the present invention provides: a quantum computing task scheduling method, comprising:

[0054] S1: Obtaining the encoding information of the corresponding scheduling task, wherein the encoding information is obtained by encoding the waiting queue of the quantum computing task and the state information of the quantum cluster;

[0055] S2: Based on the coding information, select first computing tasks that meet a first preset priority, and hierarchically process the first computing tasks in a time sequence based on the computing node status of the computing cluster;

[0056] S3: Based on the coding information, a second computing task that meets a second preset priority is screened out, and a corresponding computing node in an idle state is matched from the computing cluster to process the second computing task.

[0057] In this embodiment, after receiving a scheduling request, the scheduling system identifies the quantum computing task that currently needs to be scheduled from the waiting queue, and identifies the status information of the quantum computer cluster. It digitizes and serializes the information through encoding and then splices it to obtain encoded information. The encoded information includes a feature vector formed by splicing attributes such as the priority of the quantum computing task, computing power requirements, time limit, and required equipment type.

[0058] The quantum computing tasks to be scheduled are screened based on the encoded information, and a first computing task that meets a first preset priority is selected from the waiting queue. The first computing task is further hierarchically processed in a time sequence based on the status of computing nodes in the computing cluster, where the computing node status includes idle, busy, and remaining computing resources. The first computing task is then scheduled to a matching computing node for execution, ensuring that the first computing task meets the priority requirement for execution.

[0059] After the first computing task is screened, the second computing task is screened out according to the second preset priority. At this time, if there are idle computing nodes in the computing cluster, the second computing task is scheduled.

[0060] Reference Figure 2 , a method for screening out a first computing task that meets a first preset priority includes:

[0061] S21: extracting features from the encoded information to obtain feature parameters of the quantum computing task, where the feature parameters include a priority identifier of the quantum computing task;

[0062] S22: Filter out first computing tasks that meet a first preset priority based on the priority identifier, where the first computing tasks include first-category computing tasks and second-category computing tasks, and the first-category computing tasks have a higher priority than the second-category computing tasks.

[0063] In this embodiment, a first-level neural network is set up to extract features from the encoded information, obtain feature parameters of the quantum computing task, and perform matching and calling according to the priority identifier of the quantum computing task. The first-level neural network adopts an architecture based on the Transformer model. Through the self-attention mechanism of the Transformer model, the importance of different features in the input encoded information is weighted to determine the first preset priority. The calculation formula of the self-attention mechanism is as follows:

[0064]

[0065] Among them, the query vector Q represents the current task scheduling requirements, the key vector matrix K represents the current cluster resource status, the value vector V represents the dynamic coupling relationship between the current tasks, and the key vector dimension is d k , by calculating the normalized dot product similarity between the query vector Q and the key vector matrix K: The interaction intensity between the current quantum computing task and the set of tasks to be scheduled, as well as the nodes in the quantum computing cluster, is quantified. This includes constraints on task device type (e.g., superconducting / ion trap heterogeneity), resource competition (e.g., quantum bit capacity, time window overlap), and priority conflicts. The dot-product similarity is then mapped to probabilistic attention weights using the Softmax function. Finally, the weighted aggregate value vector V (e.g., physical constraints such as the real-time load of the encoding node and the cost of task preemption) is used to screen the quantum computing tasks and obtain the first computing task. The self-attention mechanism dynamically models quantum computing tasks and associates them with computing power resources in multiple dimensions. This allows for adaptive capture of the complex coupling relationships between quantum computing task priorities, device type heterogeneity, and computing power resources in the quantum computing cluster, providing scheduling decision support for high-concurrency, strongly constrained quantum computing environments.

[0066] The first computing tasks screened out include first-category computing tasks and second-category computing tasks. The priority of first-category computing tasks is higher than that of second-category computing tasks. When scheduling, the first-category computing tasks are scheduled first. If the computing power resources of the computing node are sufficient, all the first-category computing tasks can be scheduled and completed, and the second-category computing tasks can be further scheduled. If the computing power resources of the computing node are insufficient, the first-category computing tasks are scheduled first, and the second-category computing tasks wait.

[0067] The first computing tasks include first-category computing tasks and second-category computing tasks. The first-category computing tasks have a higher priority than the second-category computing tasks. The method for hierarchically processing the first computing tasks according to the time sequence includes:

[0068] According to the idle status of the computing nodes in the computing cluster, computing nodes are matched first for the first-category computing tasks; for the remaining first-category computing tasks or second-category computing tasks that have not been matched with computing nodes, the corresponding computing nodes in a non-idle state are matched and reserved according to the priority.

[0069] In this embodiment, the first computing task is hierarchically processed in time sequence according to the idle state of the computing nodes of the computing cluster, and the number of first quantum bits required for the computing power of the first computing task and the number of second quantum bits whose computing nodes in the current quantum computing cluster are idle are obtained. The number of first quantum bits of the first type of computing task is matched with the number of second quantum bits. If the number of first quantum bits is greater than the number of second quantum bits, it means that the computing power resources of the node do not meet the computing power requirements of the quantum computing task, and the next computing node is selected for matching.

[0070] When there are no matching idle nodes for the first type of computing tasks, the second type of computing tasks will be matched again until there are no matching idle nodes for all idle nodes. For the remaining unmatched first type computing tasks or second type computing tasks, the corresponding non-idle computing nodes will be matched and reserved according to their priority.

[0071] Furthermore, the corresponding computing nodes in a non-idle state are matched and reserved according to a preset time resource table. According to the status information of the quantum computer cluster, a time resource table is preset to record the number of quantum bits, operating status, occupied time and other parameters of each computing node in the quantum computer cluster. Although the current quantum computing cluster cannot meet the remaining first-class computing tasks or second-class computing tasks to be scheduled, computing power resources will be released after waiting for a certain period of time. After the release, the first-class computing tasks or second-class computing tasks can be scheduled. According to the time resource table, the computing nodes at the time point of computing power resource release are allocated to the first-class computing tasks or second-class computing tasks to be scheduled for execution.

[0072] Reference Figure 3 , a method for screening out the second computing task that meets the second preset priority includes:

[0073] S31: Acquire characteristic parameters of the quantum computing task, where the characteristic parameters include a priority identifier of the quantum computing task;

[0074] S32: Filter out a second computing task that meets a second preset priority based on the priority identifier, where the second preset priority is lower than the first preset priority.

[0075] In this embodiment, after the first computing task is screened, the second computing task is screened by setting a second-level neural network. The second-level neural network adopts the same Transformer model as the first-level neural network. By adjusting its self-attention mechanism, it is used as the second preset priority to screen out the second computing task, and the second preset priority is lower than the first preset priority. During the scheduling process, the first computing task is scheduled first, and then the second computing task is scheduled.

[0076] The method of matching a corresponding computing node in an idle state from the computing cluster and processing the second computing task includes:

[0077] According to the computing power requirement of the second computing task, an idle computing node with corresponding computing power requirement is matched in the computing cluster to perform the calculation.

[0078] In this embodiment, when the first computing task is being scheduled and the second computing task is waiting, a starvation problem may occur in the quantum computing task. The starvation problem is that the quantum computing task cannot be scheduled for a long time, resulting in timeout. The second computing task is backfilled through the second-level neural network. Since the computing power resources of the current quantum computing cluster cannot meet the needs of the first computing task to be scheduled, but may meet the needs of the waiting second computing task, the executable second computing task is scheduled first for backfilling, thereby reducing the waiting time of the quantum computing task.

[0079] Furthermore, existing backfilling methods include Easy Backfilling and Conservative Backfilling. However, Easy Backfilling only considers the expected scheduling time of the first task in the waiting queue, but may delay tasks waiting later in the queue. Conservative Backfilling, on the other hand, ensures that the tasks selected do not delay the execution of all tasks preceding them in the queue. This embodiment identifies the number of first qubits of the second computing task and the number of second qubits of the idle computing nodes, selects the matching second computing task for scheduling, executes the quantum computing task, and determines whether to backfill based on the number of qubits, making it easier to adapt to the backfilled quantum computing task.

[0080] Reference Figure 4 , update the scheduling task, the update method includes:

[0081] S41: Establish a reward function for the characteristic parameters and adjust the weights of the characteristic parameters to adjust the priority of the quantum computing task;

[0082] S42: After the quantum computing task is scheduled, the reward brought by the scheduled quantum computing task is calculated through the reward function, and the scheduled task is updated in real time based on the sum of several rewards obtained.

[0083] In this embodiment, a reward function is designed to calculate the reward for scheduling quantum computing tasks. The reward reflects the quality of the current quantum computing task scheduling behavior. The scheduled tasks are updated through the reward function to achieve the best scheduling effect. The reward function mainly considers balancing three factors: preventing quantum computing task starvation, task priority, and improving utilization. The calculation model of the reward function is as follows:

[0084]

[0085] Where t represents the waiting time of the selected quantum computing task, represents the average waiting time of quantum computing tasks in the waiting queue, t max Indicates the maximum waiting time of quantum computing tasks in the waiting queue. Similarly, n represents the number of requesting nodes for the selected quantum computing task, N represents the total number of nodes in the quantum computing cluster, and N used Indicates the number of nodes currently occupied by the quantum computing cluster. Used to dynamically adjust the impact of quantum computing task waiting time on rewards: When the maximum waiting time of quantum computing tasks in the waiting queue is small, the contribution of waiting time to rewards will be appropriately reduced, thereby prioritizing utilization and task priority, avoiding the impact of an excessive number of quantum computing tasks with short waiting times on overall scheduling efficiency. By adjusting the weighting coefficient, scheduling tasks can be biased towards a certain factor according to actual needs.

[0086] Furthermore, in this embodiment, the update frequency of the scheduling task is controlled by adjusting the batch parameter (batchsize). The batch parameter is the number of times quantum computing tasks are scheduled. That is, after each batch of quantum computing tasks is scheduled, the model parameters of the scheduling system are updated in real time based on the sum of the rewards fed back after the quantum computing tasks are scheduled. The next batch of quantum computing tasks will be called by the updated scheduling system based on the scheduled tasks. The scheduling tasks can be adaptively adjusted when the encoded information changes, thereby maintaining efficiency and robustness. In the early stages of training, the batch parameter is set to a small value to explore as much solution space as possible and improve the diversity and adaptability of the scheduling system model. As the model gradually stabilizes, the batch parameter can be gradually increased to reduce the frequency of model parameter updates, thereby reducing the impact of small fluctuations on the model and ensuring that the model achieves better results overall. In addition, a switch function for model parameter updates is provided, which allows users to flexibly choose whether to turn parameter updates on or off according to actual needs to adapt to different application scenarios.

[0087] Specifically, in the initial training phase, batchsize = 10 was set to explore a wider range of solution spaces. That is, after every 10 scheduling runs, the scheduling system model parameters were updated based on the sum of the rewards. The updated scheduling system then re-acquired the schedule for subsequent quantum computing tasks. After the model stabilized, batchsize = 50 was set to reduce the frequency of parameter updates.

[0088] The scheduling system's model is trained using the Proximal Policy Optimization (PPO) algorithm. PPO updates scheduled tasks by optimizing a surrogate objective function. This surrogate objective function limits the magnitude of changes in scheduled tasks with each update, ensuring that scheduled tasks do not deviate too far from the previous version. PPO combines a hybrid algorithm based on value-based learning (DQL) and policy gradient (PG). In many applications, PPO often outperforms pure DQL or PG methods, offering greater robustness and higher sample efficiency.

[0089] PPO provides two variants: penalty and clipping. Studies have shown that clipping is more effective and easier to implement than penalty. In this embodiment, clipping is used to implement the scheduling system model as follows:

[0090]

[0091] Among them, r t (θ) represents the scheduling system in the same coded information state S t The probability ratio between the new scheduled task and the old scheduled task in scheduling the same quantum computing task;

[0092]

[0093] Among them, a t Indicates that at time t, the scheduling system is in the coded information state S t According to the probability of taking action (scheduling a quantum computing task) under the scheduling task π, Indicates that in the coded information state S t The degree of advantage of the current quantum computing task scheduled by the scheduling system compared to the average scheduling task, ∈ is a hyperparameter set to 0.2, used to control the clipping range, E t The pruning mechanism ensures the conservatism of scheduled task updates and avoids large deviations that may lead to training instability.

[0094] In a specific embodiment,

[0095] The waiting queue for quantum computing tasks is as follows:

[0096] Task 1: Submission time = 0, Time limit = 3600 seconds, Number of qubits = 20, Device type = A, Priority = High, Number of experiment repetitions = 1000.

[0097] Task 2: Submission time = 60, Time limit = 1800 seconds, Number of qubits = 5, Device type = B, Priority = Medium, Number of experiment repetitions = 2000.

[0098] Task 3: Submission time = 120, Time limit = 7200 seconds, Number of qubits = 15, Device type = A, Priority = Medium, Number of experiment repetitions = 1000.

[0099] Task 4: Submission time = 180, Time limit = 2400 seconds, Number of qubits = 8, Device type = A, Priority = Low, Number of experiment repetitions = 1000.

[0100] Quantum computing cluster status: The state vector of each node includes node ID, number of quantum bits, device type, and task execution status (such as remaining time).

[0101] At time 180, the total number of nodes = 7, and the number of occupied nodes = 4.

[0102] Idle nodes: Node 1 (number of qubits = 20, device type = A), Node 2 (number of qubits = 10, device type = B), Node 3 (number of qubits = 10, device type = A).

[0103] Running nodes: nodes 4-7 (device types are A, B, A, A; running tasks are T1, T2, T3, T4; number of quantum bits are 20, 10, 15, 20; remaining time is 600 seconds, 1200 seconds, 300 seconds, 900 seconds).

[0104] Encode the quantum computing tasks in the waiting queue and the status information of the quantum computing cluster.

[0105] Coding scheme:

[0106] Device type: A=0, B=1.

[0107] Task priority: High = 2, Medium = 1, Low = 0.

[0108] Node status: Idle = 0, Running = Remaining running time.

[0109] The final encoded feature vector results are as follows:

[0110]

[0111] Based on the output of the first-level neural network, the quantum computing task is scheduled for execution or reservation. Task 1 (high priority) is selected. Cluster resources are checked: Node 1 meets Task 1's requirements (device type A, 20 qubits). It is executed immediately. Task 2 (medium priority) is selected. Cluster resources are checked: Node 2 meets Task 2's requirements (device type B, 10 qubits). It is executed immediately. Task 3 (medium priority) is selected. Cluster resources are checked: No idle nodes meet the requirements. It is marked as a reserved task and Node 6 (device type A, 15 qubits) is reserved for availability in 300 seconds.

[0112] The second-level neural network selects quantum computing tasks or reserved tasks that can be backfilled. Task 4 meets the backfill criteria and is selected. Node 3 (number of qubits = 10, device type = A) meets the requirements of Task 4 and is immediately scheduled for execution.

[0113] Reference Figure 5 The present invention also provides an embodiment of a quantum computing task scheduling system, which is applicable to the scheduling method of any one of the above technical solutions, including:

[0114] Encoding module 1 is used to encode the waiting queue of the quantum computing task and the state information of the quantum cluster to obtain encoding information and form a feature vector corresponding to the scheduling task;

[0115] A neural network module 2 generates a scheduling strategy based on the feature vector, including screening out a first computing task that meets a first preset priority; and

[0116] Screening out a second computing task that meets a second preset priority;

[0117] Scheduling module 3, executing task scheduling according to the scheduling policy, including hierarchical processing of the first computing task according to a time sequence based on the computing nodes of the computing cluster according to the first computing task; and

[0118] According to the second computing task, a corresponding computing node in an idle state is matched from the computing cluster, and the second task is processed.

[0119] In this embodiment, the encoding module 1 encodes the waiting queue of the quantum computing task and the state information of the quantum cluster to obtain encoding information, forms a feature vector S corresponding to the scheduling task, and sends the feature vector to the neural network module 2.

[0120] The neural network module 2 performs feature extraction and calculation based on the feature vector, and selects the first calculation task and the second calculation task by setting the first level neural network and the second level neural network.

[0121] Scheduling Module 3 sets a scheduling model, schedules the first computing task based on the status of the computing nodes in the computing cluster, and reserves corresponding non-idle computing nodes based on the remaining unmatched first computing nodes, waiting for subsequent computing power resources to be released before calling for execution. For idle computing nodes that do not match the first computing task, the second computing task is called for execution, scheduling the quantum computing task for execution.

[0122] It also includes a dynamic adjustment module 4, which updates the scheduling model of the scheduling module 3 by setting a reward function.

[0123] The feature vector includes: the characteristics of each quantum computing task in the waiting queue of quantum computing tasks; the number of idle nodes in the quantum computing cluster, the number of quantum bits of the node, the device type corresponding to the node, and the remaining time of the running node.

[0124] In this embodiment, for quantum computing tasks, it is necessary to obtain the characteristics of each task, including: task submission time, task time limit, number of qubits required for the task, type of quantum device required for the task, task priority, and number of experimental repetitions. For quantum computing clusters, it is necessary to obtain the number of idle nodes in the cluster, the number of qubits per node, and the device type corresponding to each node. For running computing nodes, it is necessary to obtain the specific execution status of the task on each computing node. This information is encoded and concatenated into a feature vector S, which serves as the input to the first-level neural network.

[0125] Reference Figure 6 The neural network module 2 includes a first-level neural network and a second-level neural network. The first-level neural network is used to screen out first computing tasks that meet a first preset priority, and the second-level neural network is used to screen out second computing tasks that meet a second preset priority. The first-level neural network and the second-level neural network have the same architecture.

[0126] The first-level neural network or the second-level neural network includes an embedding layer 21 for receiving a feature vector;

[0127] The output end of the embedding layer 21 is connected to at least one or more encoder units 22 to perform feature extraction on the feature vector;

[0128] The output end of the encoder unit 22 is connected to the calculation unit 23, which performs weight calculation on the extracted characteristic parameters to determine the priority identifier of the quantum computing task.

[0129] The existing Transformer model mainly consists of an encoder module and a decoder module, which contains multiple encoders / decoders with the same architecture. Each encoder and decoder consists of a self-attention layer (Self-Attention), a position-wise feed forward network (Position-wise Feed Forward Network) and a residual connection normalization layer (Add&Norm), and each decoder also contains a masked self-attention layer. Parallel computing is achieved through the Transformer model.

[0130] In this embodiment, the first-level neural network and the second-level neural network adopt the architecture of the Transformer model, and the architecture of the Transformer model is improved. Only a number of encoder units 22 are used to extract features from feature vectors. After obtaining feature parameters, the extracted feature parameters are calculated by the calculation unit 23 to determine the scheduling task corresponding to the quantum computing task.

[0131] Specifically, the feature vector is mapped and input to the encoder unit through an embedding layer (Input Embedding), where the embedding dimension is set to 128. Too high a dimension will increase the computational complexity. The encoder unit 22 uses a 3-layer encoder and a multi-head attention layer (Multi-Head Attention), using 4 heads. The multi-head attention layer is a mechanism that can improve the performance of the ordinary self-attention layer, and different feature parameters are extracted through a 3-layer encoder. The computing unit 23 is connected to the back of the encoder unit 22. The computing unit includes: a convolution layer (cov), which performs convolution calculation on the feature parameters, linearly corrects them through the rectified linear unit (ReLU) and the linear layer (Linear), and linearly outputs them to the transfer function (Softmax). The linear layer output is converted into a probability distribution to obtain several quantum computing tasks for determining the scheduling task.

[0132] Also included is a computer-readable storage medium storing a computer program for executing the scheduling method, wherein the computer program causes a computer to execute the following steps:

[0133] S1: Obtaining the encoding information of the corresponding scheduling task, wherein the encoding information is obtained by encoding the waiting queue of the quantum computing task and the state information of the quantum cluster;

[0134] S2: Based on the coding information, select first computing tasks that meet a first preset priority, and hierarchically process the first computing tasks in a time sequence based on the computing node status of the computing cluster;

[0135] S3: Based on the coding information, a second computing task that meets a second preset priority is screened out, and a corresponding computing node in an idle state is matched from the computing cluster to process the second computing task.

[0136] Among them, the computer-readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, a computer-readable storage medium is coupled to a processor so that the processor can read information from the computer-readable storage medium and write information to the computer-readable storage medium. Of course, the computer-readable storage medium can also be an integral part of the processor. The processor and the computer-readable storage medium can be located in an application-specific integrated circuit (ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the computer-readable storage medium can also exist in a communication device as discrete components.

[0137] Specifically, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0138] Also included is an electronic device comprising:

[0139] one or more processors; memory; and

[0140] One or more programs, wherein the one or more programs are stored in a memory and configured to be executed by one or more processors, the programs comprising steps for performing the following steps:

[0141] S1: Obtaining the encoding information of the corresponding scheduling task, wherein the encoding information is obtained by encoding the waiting queue of the quantum computing task and the state information of the quantum cluster;

[0142] S2: Based on the coding information, select first computing tasks that meet a first preset priority, and hierarchically process the first computing tasks in a time sequence based on the computing node status of the computing cluster;

[0143] S3: Based on the coding information, a second computing task that meets a second preset priority is screened out, and a corresponding computing node in an idle state is matched from the computing cluster to process the second computing task.

[0144] Memory is used to store computer programs. This memory may include high-speed random access memory (RAM) and non-volatile memory (NVM), such as at least one disk storage device. It can also be a USB flash drive, a mobile hard drive, a read-only memory, a magnetic disk, or an optical disk.

[0145] A processor is used to execute a computer program stored in a memory. The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), or application-specific integrated circuits (ASICs). A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0146] Optionally, the memory can be independent or integrated with the processor.

[0147] When the memory is a device independent of the processor, the electronic device may further include a bus. The bus is used to connect the memory and the processor. The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be divided into address buses, data buses, control buses, etc.

[0148] It should be noted that, through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiment. In this article, relational terms such as first and second are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. Without further constraints, an element defined by the phrase "comprises a..." does not preclude the existence of additional identical elements in the process, method, article or apparatus that includes the element.

[0149] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A quantum computing task scheduling method, characterized in that: include: Obtaining encoding information corresponding to the scheduling task, wherein the encoding information is obtained by encoding the waiting queue of the quantum computing task and the state information of the quantum cluster; Based on the coding information, first computing tasks meeting a first preset priority are screened out, and based on the computing node status of the computing cluster, the first computing tasks are hierarchically processed in a time sequence; Based on the coding information, a second computing task that meets a second preset priority is screened out, and a corresponding computing node in an idle state is matched from the computing cluster to process the second computing task.

2. The scheduling method according to claim 1, wherein: The method for selecting the first computing task that meets the first preset priority includes: Performing feature extraction on the encoded information to obtain feature parameters of the quantum computing task, wherein the feature parameters include a priority identifier of the quantum computing task; The first computing tasks meeting the first preset priority are screened out based on the priority identifier, the first computing tasks including first-category computing tasks and second-category computing tasks, and the first-category computing tasks have a higher priority than the second-category computing tasks.

3. The scheduling method according to claim 1, wherein: The first computing tasks include first-category computing tasks and second-category computing tasks, the first-category computing tasks have a higher priority than the second-category computing tasks, and the method for hierarchically processing the first computing tasks according to a time sequence includes: According to the idle status of the computing nodes of the computing cluster, computing nodes are matched for the first type of computing tasks first; for the remaining unmatched computing nodes of the first type of computing tasks or the second type of computing tasks, corresponding computing nodes in a non-idle state are matched and reserved according to priority.

4. The scheduling method according to claim 1, wherein: The method for selecting the second computing task that meets the second preset priority includes: Acquiring characteristic parameters of the quantum computing task, wherein the characteristic parameters include a priority identifier of the quantum computing task; The second computing tasks meeting the second preset priority are screened out based on the priority identifier, where the second preset priority is lower than the first preset priority.

5. The scheduling method according to claim 4, characterized in that: The method of matching a corresponding computing node in an idle state from the computing cluster and processing the second computing task includes: According to the computing power requirement of the second computing task, an idle computing node with corresponding computing power requirement is matched in the computing cluster to perform the calculation.

6. The scheduling method according to claim 2, characterized in that: The scheduling task is updated, and the updating method includes: Establishing a reward function related to the characteristic parameters and adjusting the weights of the characteristic parameters to adjust the priority of the quantum computing task; After the quantum computing task is scheduled, the reward brought by scheduling the quantum computing task is calculated through the reward function, and the scheduled task is updated in real time according to the sum of the several rewards obtained.

7. A quantum computing task scheduling system, characterized in that: include: The encoding module is used to encode the waiting queue of the quantum computing task and the state information of the quantum cluster to obtain encoding information and form the feature vector of the corresponding scheduling task; A neural network module generates a scheduling strategy based on the feature vector, including screening out a first computing task that meets a first preset priority; as well as Screening out a second computing task that meets a second preset priority; a scheduling module, configured to execute task scheduling according to the scheduling policy, including hierarchically processing the first computing task in a time sequence based on the computing nodes of the computing cluster according to the first computing task; as well as According to the second computing task, a corresponding computing node in an idle state is matched from the computing cluster, and the second task is processed.

8. The dispatching system according to claim 7, characterized in that: It also includes a dynamic adjustment module that establishes a reward function for the characteristic parameters of the quantum computing task, adjusts the priority of the quantum computing task, calculates the reward brought by scheduling the quantum computing task through the reward function, and updates the scheduled task in real time based on the sum of several rewards obtained.

9. The dispatching system according to claim 7, characterized in that: The neural network module includes a first-level neural network and a second-level neural network. The first-level neural network is used to screen out first computing tasks that meet a first preset priority, and the second-level neural network is used to screen out second computing tasks that meet a second preset priority.

10. The dispatching system according to claim 7, characterized in that: The first computing task is to determine whether the quantum task is to be executed immediately or reserved for execution; the second computing task is responsible for performing quantum computing task backfill.

11. The dispatching system according to claim 7, characterized in that: The architecture of the first-level neural network and the second-level neural network are the same; The first-level neural network or the second-level neural network includes an embedding layer for receiving the feature vector; An output end of the embedding layer is connected to at least one or more encoder units to perform feature extraction on the feature vector; The output end of the encoder unit is connected to a calculation unit, and the calculation unit performs weight calculation on the extracted characteristic parameters to determine the priority identifier of the quantum computing task.

12. A computer-readable storage medium, characterized in that The computer stores a computer program for running the scheduling method, wherein the computer program enables a computer to execute the scheduling method according to any one of claims 1 to 6.

13. An electronic device, characterized in that: include: one or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for executing the scheduling method according to any one of claims 1 to 6.

Images