Resource scheduling method and system based on multi-dimensional load prediction

Abstract

The invention discloses a resource scheduling method and system based on multi-dimensional load prediction, relates to the technical field of cloud computing and intelligent scheduling, and is used for solving the problems of efficient prediction and adaptive scheduling under multi-dimensional resource load fluctuation. The method comprises the following steps: collecting and preprocessing historical resource use data to construct a multi-dimensional time sequence sample; secondly, predicting a future load, and inhibiting an abnormal fluctuation influence in combination with a sudden change detection mechanism; thirdly, according to task request features and a prediction load result, task priority is evaluated, a preliminary scheduling scheme is generated, and reasonable resource allocation is achieved; finally, key feedback indexes in the scheduling execution process are collected, a scheduling quality scoring system is constructed, and a scheduling strategy is dynamically adjusted and optimized through a multi-objective optimization algorithm. The method has the characteristics of high prediction precision, fast scheduling response, strong adaptability and the like, can effectively improve the resource utilization rate and the task completion efficiency, and reduces the scheduling delay and default risk.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud computing and intelligent scheduling. More specifically, the present invention relates to a resource scheduling method and system based on multi-dimensional load prediction. Background Art

[0002] In modern cloud computing platforms, resource scheduling is a key technology to ensure the efficient operation of the system and service quality. With the diversification of business types and the dynamic changes in user requirements, the number of scheduling tasks has increased sharply, and resource requests have shown characteristics such as multi-dimensionality, high frequency, and strong suddenness, resulting in severe fluctuations in system load and a decline in resource utilization efficiency. Traditional scheduling methods mostly rely on static rules, fixed priorities, or simple resource utilization predictions, and are difficult to adapt to complex dynamic scheduling scenarios, easily causing problems such as uneven resource allocation, task execution delays, and even service defaults.

[0003] At the same time, resource usage behavior has obvious temporal and periodic characteristics, and may exhibit non-stationary jumps during sudden loads or task type switches, which poses higher requirements for load prediction and scheduling strategies. Existing scheduling systems generally lack the ability to deeply model multi-dimensional historical data, cannot accurately capture future load change trends, and also lack an effective feedback optimization mechanism, making it difficult to continuously improve the scheduling effect.

[0004] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a resource scheduling method and system based on multi-dimensional load prediction to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions: In a preferred embodiment, it includes: Step 1: Collect and preprocess historical resource usage data, and construct a normalized time series input sample; Step 2: Predict future resource loads, identify task switches, request fluctuations, and sudden jumps, and optimize model inputs; Step 3: Construct feature vectors, combine the predicted load trends, calculate task priorities, and generate a preliminary scheduling plan; Step 4: Monitor the actual scheduling effect, adjust the scheduling strategy through feedback analysis and multi-objective optimization algorithms, and perform closed-loop iterative optimization.

[0007] In a preferred embodiment, in Step 1, historical resource data is extracted, normalized, and then converted into a set of input sequence samples of a fixed length to generate a historical resource sequence data set.

[0008] In a preferred embodiment, in step 2, through a gated recurrent unit and a convolutional neural network, long-term dependencies and local patterns in the historical resource sequence dataset are extracted, and a predicted value of the resource load at the next moment or future time point is output.

[0009] In a preferred embodiment, in step 2, the task request type switching frequency, resource request pattern irregularity, and historical resource sequence dataset mutation jump recognition index in the input historical resource sequence dataset are monitored in real time, and the mutation degree of the historical resource sequence dataset at this time is determined by weighted summation. A mutation degree threshold of the historical resource sequence dataset is set. When the mutation degree of the historical resource sequence dataset at this time is greater than or equal to the mutation degree threshold of the historical resource sequence dataset, the current time step is marked as a mutation point. When the mutation degree of the historical resource sequence dataset at this time is less than the mutation degree threshold of the historical resource sequence dataset, no mutation processing operation is required.

[0010] In a preferred embodiment, in step 2, the increment between the current input and the previous moment is calculated, and the hyperbolic tangent function is used to compress the increment of the input change at the mutation point and replace the original mutation input into the GRU; Calculate the state difference of the historical resource sequence dataset at this time, set a state difference threshold ϵ. If the state difference of the historical resource sequence dataset at this time is greater than or equal to the state difference threshold ϵ, it is considered that the state of the historical resource sequence dataset jumps at this time. The output of the gated recurrent unit at this time is clipped by an amplitude threshold, and a resource load prediction value adjustment model is established by combining the state difference and the input increment.

[0011] In a preferred embodiment, in step 3, task request data from the upstream scheduling interface is received and converted into a feature vector available for scheduling optimization.

[0012] In a preferred embodiment, in step 3, a weighted scoring function is used to define the task priority score, form an initial scheduling sort list, input the predicted load trend data obtained in step 2, and set a scheduling window according to the load threshold to form a first resource scheduling plan.

[0013] In a preferred embodiment, in step 4, the scheduling plan is deployed and feedback monitoring is performed, and scheduling optimization is carried out based on the feedback results.

[0014] In a preferred embodiment, it includes: a data processing module, a load prediction module, a scheduling decision module, and a feedback optimization module, and the modules are signal-connected to each other; The data processing module is mainly used to collect and preprocess historical load resource data and construct a normalized time series input sample; The load prediction module is mainly used to predict future resource loads, identify task switches, request fluctuations, and sudden jumps, and optimize the model input; The scheduling decision module constructs feature vectors, combines the predicted load trend, calculates task priorities, and generates a preliminary scheduling plan; The feedback optimization module is mainly used to monitor the actual scheduling effect, and through feedback analysis and multi-objective optimization algorithms, adjust the scheduling strategy and perform closed-loop iterative optimization.

[0015] The technical effects and advantages of an energy-saving and environment-friendly rail transit turnout snow melting system of the present invention: By integrating multi-dimensional load prediction and intelligent scheduling optimization, the present invention realizes the coordination of the forward-looking and dynamic nature of resource management. Compared with traditional methods, the present solution has significant technical advantages in the following aspects: First, it can comprehensively utilize multi-source monitoring data such as historical CPU, memory, and bandwidth to achieve fine-grained time series modeling; Second, the introduction of a GRU-CNN hybrid structure combined with a mutation detection mechanism effectively improves the prediction accuracy and stability in sudden load scenarios; Third, the scheduling strategy is not only based on the static attributes of tasks, but also fully considers the predicted load trend and system resource status to achieve dynamic priority evaluation and reasonable arrangement of tasks; Fourth, a closed-loop feedback optimization mechanism is designed to reverse-optimize the strategy parameters through the scheduling execution results, continuously improving the system scheduling efficiency and resource utilization rate. In summary, the present invention significantly enhances the adaptive ability and service guarantee ability of the resource scheduling system in a complex cloud environment. Brief Description of the Drawings

[0016] Figure 1 It is an optimization flowchart of a resource scheduling method and system based on multi-dimensional load prediction of the present invention.

[0017] Figure 2 It is an implementation flowchart of a resource scheduling method and system based on multi-dimensional load prediction of the present invention.

[0018] Figure 3 It is a timing diagram of a resource scheduling method and system based on multi-dimensional load prediction of the present invention. Detailed Embodiment

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Embodiment: The present invention discloses a resource scheduling method based on multi-dimensional load prediction, as Figure 3 shown, including: Step 1: Preprocess the historical resource usage data and construct the time series input required for prediction.

[0021] First, extract the historical resource data from the time series logs generated by resource management platforms such as Prometheus and Kubernetes Metrics Server. Among them, the extracted fields include: CPU utilization (unit: %); Memory usage (unit: MB); Bandwidth usage (unit: Mbps); Task request frequency (unit: times / minute); Among them, all fields need to contain a unified timestamp and set a fixed sampling interval. If the original records are discontinuous and there are time gaps, the forward fill method (forwardfill) or linear interpolation method is used for time alignment and missing value processing to restore data continuity.

[0022] After that, generate a multi-dimensional time series data table based on the extracted historical resource data, denoted as Rt; Furthermore, traverse each column of historical resource data to independently calculate its mean μ and standard deviation σ, and use the Z-score normalization method to normalize the corresponding value at each time point. Specifically, according to the formula: zi = (xi - μ) / σ; Among them, xi represents the original value, and zi represents the normalized value; After that, for features with too small fluctuations or constant values, set anomaly detection conditions to avoid division by zero, and finally generate the normalized data table Rt'.

[0023] Construct a sliding time window sequence to convert the normalized continuous historical data into a set of input sequence samples of a fixed length. Specifically, first set the time window length T representing that each input sample contains data of the past T time points, and the prediction target offset step tf representing that each input sample corresponds to predicting the data at the T + tf-th time point, and traverse and slice the normalized historical resource data Rt'. Each time, a pair of samples is generated; Among them, the sliding logic is as follows: The first group of samples: Input: {R1', R2',..., RT'}; Output target: RT + tf'; The second group of samples: Input: {R2', R3',..., RT + 1'}; Output target: RT + 1 + tf - 1'; And so on, the window slides forward one step each time until the end, and finally generates a historical resource sequence data set.

[0024] Step 2: Use a deep learning model to train the processed historical resource sequence data and predict future load demands.

[0025] First, based on the GRU-CNN hybrid deep neural network model, the gated recurrent unit GRU is used to recursively model the front and back states in the training dataset finally generated in Step 1, dynamically memorizing the key load trends; Specifically, the control unit includes: Update gate: Decide whether to retain past information at the current moment; Reset gate: Decide the degree of fusion between the current input and the past state; Among them, the calculation process is as follows: First, calculate the output vector of the update gate , specifically according to the formula: , where \(x_t\) represents the input vector at the current time step, \(h_{t - 1}\) represents the hidden state at the previous time step, \(I_z\) represents the weight matrix input to the update gate, and \(U_z\) represents the weight matrix from the hidden state to the update gate; Calculate the output vector of the reset gate , specifically according to the formula: , where \(I_r\) represents the weight matrix input to the reset gate, and \(U_r\) represents the weight matrix from the hidden state to the reset gate; After that, calculate the candidate hidden state , specifically according to the formula: , where \(I_h\) represents the weight matrix input to the candidate hidden state, \(U_h\) represents the weight matrix from the hidden state to the candidate hidden state, and \(\odot\) represents element-wise multiplication; Finally, calculate the final hidden state at the current time step , specifically according to the formula: , where represents the weight of the old information, represents the weight of the new information.

[0026] After that, map each input sequence to a set of hidden state sequences \(H = [h_1, h_2,..., h_T]\), \(H\in\mathbb{R}^{T\times D}\), where the output \(H\) contains the overall trend and periodic characteristics in the historical load data; Furthermore, take the \(H\) output by GRU as a one-dimensional "feature map" and apply multiple 1D convolutional kernels of different sizes along the time dimension: \(C_i = ReLU(W_i * H + b_i)\), \(i = 1, 2,..., K\); where \(W_i\) represents the \(i\)-th convolutional kernel, \(*\) represents the one-dimensional convolutional operation; \(K\) represents the number of groups of convolutional kernels for multi-scale feature extraction.

[0027] After that, the convolution result is sent to the max-pooling layer for dimensionality reduction, retaining the most significant responses, specifically according to the formula: Pi = max(Ci); All Pi are concatenated into a one-dimensional feature vector P = [P1, P2,..., PK] for final prediction, further improving the accuracy and generalization performance.

[0028] After training is completed, assuming the output vector is P, it is input into one or more fully connected neurons for linear or non-linear mapping: Ŷ = f(WpP + bp); Among them, WpP represents the weight of the fully connected layer, and f represents the ReLU or linear activation function; After that, single-step prediction of the CPU load at the next moment or multi-step prediction of the complete load sequence at future time points is performed to form the final resource load prediction value.

[0029] It should be noted that considering that in the heterogeneous cloud platform environment facing multiple users with high-frequency elastic scaling, there are problems such as sudden changes in task types and a sharp increase in the randomness of resource request patterns. The load status will show high instability, and there will be non-stationary mutation characteristics in the historical resource sequence dataset. The GRU model is difficult to capture the mutation patterns, and at the same time, it will cause the convolution weights of the model to deviate from the actual load response. Therefore, in this embodiment, during the process of learning the long-term dependence relationship of the historical resource sequence dataset by the gated recurrent unit GRU and extracting features by CNN, the following three aspects of feature monitoring are performed on the input historical resource sequence dataset: Monitoring the switching frequency of task request types: Specifically, a task type classification index set is established based on the input historical resource sequence dataset, and the switching times ϕt of task types within each time window [t−Δt, t] are counted; Detecting the irregularity of resource request patterns: The coefficient of variation CV is used to calculate the volatility of the resource request distribution, specifically according to the formula: CVt = σ(Xt) / μ(Xt); Identifying sudden jumps in historical load data: The input historical resource sequence dataset is analyzed to calculate the first-order difference sequence, specifically according to the formula: ΔXt = Xt - X(t - 1) Furthermore, according to the task request type switching frequency, resource request pattern irregularity, and historical load data sudden jump identification indicators calculated above, the mutation degree Ψt of the historical resource sequence dataset at this time is determined through a weighted summation evaluation function, specifically according to the formula: Ψt = α1×ϕt + α2×CVt + α3×|||ΔXt||; Among them, α1 represents the weight of the task request type switching frequency, α2 represents the weight of the resource request mode irregularity, α3 represents the weight of the historical load data mutation jump recognition. Set the historical load data mutation degree threshold YΨ, and compare the mutation degree Ψt of the historical resource sequence data set at this time with the historical load data mutation degree threshold YΨ. When the mutation degree Ψt of the historical resource sequence data set at this time is greater than or equal to the historical load data mutation degree threshold YΨ, mark the current time step as a mutation point. When the mutation degree Ψt of the historical resource sequence data set at this time is less than the historical load data mutation degree threshold YΨ, no mutation processing operation is required; When a mutation situation is detected, immediately interrupt the existing operation and calculate the increment between the current input and the previous moment, specifically according to the formula: Δin = Xt - X(t - 1); And use the hyperbolic tangent function to compress the increment of the input change at the mutation point and control it within the interval (−1, 1), specifically according to the formula: Xt- = X(t - 1) + tanh(Δin); The compressed input Xt- replaces the original mutation input and enters the GRU, converting the mutation point into a gradual change point, reducing the fluctuation amplitude of the gating unit, avoiding over-strong stimulation of the gating unit by the input mutation, and weakening the prediction deviation caused by the outlier; Furthermore, set the current GRU hidden state as ht, and introduce the exponential weighted moving average (EWMA) for smoothing processing: ht- = β × ht + (1 - β) × h(t - 1); Then, calculate the state difference of the historical load data at this time after the output of the smoothed gated recurrent unit, specifically according to the formula: Δht = ||ht - h(t - 1)||; And set the state difference threshold ϵ. If the state difference Δht of the historical load data at this time is greater than or equal to the state difference threshold ϵ, it is considered that the state of the historical load data jumps at this time, and the output of the gated recurrent unit at this time needs to be clipped by the amplitude threshold; Furthermore, establish a resource load prediction value adjustment model by combining the state difference and the input increment: Y^t = f(ht′) + λ1 × tanh(Δht) + λ2 × tanh(Δin); Among them, Yt^ represents the finally optimized resource load prediction value, f(ht′) represents the initial prediction value output by the CNN - GRU, and λ1, λ2 represent the strengths of controlling the optimization terms.

[0030] Step 3: Describe the scheduling request in multiple dimensions to determine its importance and resource requirements, and generate a preliminary scheduling plan based on the predicted value and feature scores.

[0031] First, automatically receive task request data from the upstream scheduling interface through the task scheduling interface RESTAPI or message queue, including: request ID, request timestamp, requested resource volume, task type, request priority, and estimated task execution time, etc., and form a structured data table; Furthermore, use the following technical means for multi-dimensional feature engineering processing to convert the above structured task information into feature vectors that can be used for scheduling optimization: Type vectorization: Use One-Hot encoding to map the task type to a discrete vector; Time normalization: Convert the submission timestamp to a working period factor and normalize it to [0,1]; Resource request volume normalization: Use the Min-Max normalization method to process memory and bandwidth in the same way; Represent each task as a normalized feature vector that can be used for scoring and sorting, forming the first scheduling feature set.

[0032] After that, use a weighted scoring function to define the task priority score: Si = α1×Ti + α2×Ri + α3×Pi; Where Ti represents the time feature score, Ri represents the resource request feature score, and Pi represents the task priority; And sort all tasks in descending order according to the score Si to form an initial scheduling list; Furthermore, input the predicted load trend data obtained in step two and set a scheduling window according to the load threshold; After that, select tasks one by one from the task list, and sequentially find the earliest time period that can meet its resource request and estimated duration. If the current load + the resource consumption of this task < the set upper limit, insert the task. If the task delay time exceeds its maximum allowable scheduling delay, mark it as unschedulable; and record each scheduled task, including: task ID, start time, allocated resource volume, and execution duration; Form the first resource scheduling plan, which includes all task information that can be reasonably scheduled under the current predicted load, and clarifies the scheduling order, execution time period, and resource configuration quantity of each task.

[0033] Step 4: Deploy the scheduling plan and perform feedback monitoring, and optimize the scheduling based on the feedback results; First, translate the scheduling instructions into resource control instructions, such as container deployment scripts or task orchestration instructions; After that, use the container Kubernetes or the virtual machine scheduler OpenStack to deploy tasks according to the scheduling plan, deploy the first resource scheduling scheme to the scheduler. At the same time, deploy resource monitoring agents on each node to collect the following key performance indicators during the task scheduling process: Resource utilization rate; Task completion rate and SLA violation rate; Energy consumption data; Scheduling latency and task response time, etc.; Structurize the feedback data to form the first feedback result; And use a time sliding window to statistically analyze the feedback data to form the first feedback result data set; Furthermore, construct a comprehensive scheduling quality scoring function to comprehensively evaluate the first feedback result. Among them, the comprehensive scheduling quality scoring function is as follows: Q = λ1×U + λ2×(1 - D) + λ3×(1 - T); Where, U represents the average resource utilization rate, D represents the task SLA violation rate, T represents the average scheduling latency, and λi represents the weight coefficient; Set a preset target Qth, compare the score Q of the first feedback result with the preset target Qth. When the score Q of the first feedback result is greater than or equal to the preset target Qth, it indicates that the preset target has been achieved; when the score Q of the first feedback result is less than the preset target Qth, it indicates that the scheduling result has not reached the preset target, and enter the optimization process.

[0034] The optimization process is as follows, as Figure 1 shown: Step Y1: Convert the feedback information into a structured metric matrix, where each row represents the feedback record of a task scheduling cycle, and each column represents a key metric; after that, use correlation analysis such as the Pearson correlation coefficient ρ to identify the influence degree between each feedback metric and the total task completion rate; After that, use threshold detection and Z-score anomaly detection methods to identify the performance imbalance points, judge whether the corresponding scheduling policy configuration causes system bottlenecks, and finally output the bottleneck constraint set as the optimization input.

[0035] Step Y2: Define a set of objective functions, including the following typical objectives: f1: Maximize the task completion rate; f2: Minimize the total energy consumption; f3: Minimize the scheduling latency; f4: Minimize the SLA violation rate; Furthermore, an optimization model is constructed, and the multi-objective optimization algorithm NSGA-II is applied to generate candidate scheduling schemes that meet the balance of different scheduling requirements.

[0036] Step Y3: Based on the candidate scheduling schemes generated in Step Y2, each objective value is processed using a normalization method to obtain a standardized evaluation matrix, and the direction of the objective function is adjusted to convert all objectives into a minimized form. Furthermore, weights wi are assigned to each objective, satisfying ∑wi = 1, and the comprehensive score of each candidate scheme is calculated; then, the optimal scheme is selected based on the comprehensive score of each candidate scheme to replace the original scheduling scheme, forming the first scheduling optimization scheme. The optimized scheme replaces the original scheduling scheme and is redeployed to enter the next scheduling cycle, forming a closed-loop scheduling system of prediction - execution - feedback - optimization.

[0037] The present invention also proposes a resource scheduling system based on multi-dimensional load prediction, as Figure 2 shown, including: a data processing module, a load prediction module, a scheduling decision module, and a feedback optimization module, with signal connections between the modules; The data processing module is mainly used to collect and preprocess historical resource usage data and construct a normalized time series input sample; The load prediction module is mainly used to predict future resource loads, identify task switching, request fluctuations, and sudden jumps, and optimize the model input; The scheduling decision module constructs a feature vector, combines the predicted load trend, calculates the task priority, and generates a preliminary scheduling scheme; The feedback optimization module monitors the actual scheduling effect, and adjusts the scheduling strategy and performs closed-loop iterative optimization through feedback analysis and multi-objective optimization algorithms.

[0038] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0039] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0040] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application of the technical solution and the inventive constraints. Those skilled in the art can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0041] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0042] As described above, this is only a specific implementation of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0043] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A resource scheduling method based on multi-dimensional load prediction, characterized in that Including: Step 1: Collect and preprocess historical resource usage data, and construct a normalized time series input sample; Step 2: Predict future resource loads, identify task switches, request fluctuations, and sudden jumps, and optimize model inputs; Step 3: Construct feature vectors, combine predicted load trends, calculate task priorities, and generate a preliminary scheduling plan; Step 4: Monitor the actual scheduling effect, adjust the scheduling strategy through feedback analysis and multi-objective optimization algorithms, and perform closed-loop iterative optimization.

2. The resource scheduling method based on multi-dimensional load prediction according to claim 1, wherein: In Step 1, extract historical load resource data, perform normalization processing, and then convert it into a set of input sequence samples with a fixed length to generate a historical resource sequence data set.

3. The resource scheduling method based on multi-dimensional load prediction according to claim 2, wherein: In Step 2, through a gated recurrent unit and a convolutional neural network, extract long-term dependencies and local patterns in the historical resource sequence data set, and output the predicted value of the resource load at the next moment or future time points.

4. The resource scheduling method based on multi-dimensional load prediction according to claim 3, characterized in that; In Step 2, real-time monitor the task request type switching frequency, resource request pattern irregularity, and historical resource sequence data set mutation jump recognition index in the input historical resource sequence data set, and determine the mutation degree of the historical resource sequence data set at this time by weighted summation. Set the mutation degree threshold of the historical resource sequence data set. When the mutation degree of the historical resource sequence data set at this time is greater than or equal to the mutation degree threshold of the historical resource sequence data set, mark the current time step as a mutation point. When the mutation degree of the historical resource sequence data set at this time is less than the mutation degree threshold of the historical resource sequence data set, no mutation processing operation is required.

5. The resource scheduling method based on multi-dimensional load prediction according to claim 4, wherein: In Step 2, calculate the increment of the current input historical resource sequence data set and the historical resource sequence data set at the previous moment, and use the hyperbolic tangent function to compress the increment of the historical resource sequence data set for the input change at the mutation point, and replace the original mutation input into the GRU; Calculate the state difference of the historical resource sequence data set at this time, set the state difference threshold ϵ. If the state difference of the historical resource sequence data set at this time is greater than or equal to the state difference threshold ϵ, it is considered that the state of the historical resource sequence data set jumps at this time. Clip the output of the gated recurrent unit at this time by the amplitude threshold, and establish a resource load prediction value adjustment model in combination with the state difference and the input increment.

6. The resource scheduling method based on multi-dimensional load prediction according to claim 5, characterized in that: In Step 3, receive task request data from the upstream scheduling interface and convert it into a feature vector available for scheduling optimization.

7. The resource scheduling method based on multi-dimensional load prediction according to claim 6, characterized in that: In Step 3, use a weighted scoring function to define task priority scores, form an initial scheduling sort list, input the predicted load trend data obtained in Step 2, and set a scheduling window according to the load threshold to form the first resource scheduling plan.

8. The resource scheduling method based on multi-dimensional load prediction according to claim 7, wherein; In Step 4, deploy the scheduling plan and perform feedback monitoring, and perform scheduling optimization based on the feedback results.

9. A resource scheduling system based on multi-dimensional load prediction, characterized in that Including: A data processing module, a load prediction module, a scheduling decision module, and a feedback optimization module, with signal connections between the modules; The data processing module is mainly used to collect and preprocess historical load resource data and construct a normalized time series input sample; The load prediction module is mainly used to predict future resource loads, identify task switches, request fluctuations, and sudden jumps, and optimize model inputs; The scheduling decision-making module constructs feature vectors, combines the predicted load trend, calculates the task priorities, and generates a preliminary scheduling plan; The feedback optimization module is mainly used to monitor the actual scheduling effect, and through feedback analysis and multi-objective optimization algorithms, adjusts the scheduling strategy and performs closed-loop iterative optimization.

Images