Intelligent decision-making methods, devices, electronic equipment and storage media for resource scheduling

Abstract

This application provides a resource scheduling intelligent decision-making method, apparatus, electronic device, and storage medium, relating to the field of resource operation and management technology. By inputting historical feature data into a trained target decision-making model, the model predicts the target scheduling state of a target node and its corresponding state prediction probability based on its trained neural network. Furthermore, when the state prediction probability is not less than a preset state probability threshold, the aforementioned target scheduling state is taken as the resource scheduling result of the target node. Based on this method, the management object can quickly and accurately determine the resource scheduling result of the target node through the target scheduling state predicted by the target decision-making model, reducing the steps required for adjusting business weights and improving the decision-making efficiency of resource scheduling.

Description

Technical Field

[0001] This invention relates to the field of resource operation and management technology, and in particular to a resource scheduling intelligent decision-making method, device, electronic device and storage medium. Background Technology

[0002] In large organizations / departments, there are often multiple branch nodes, each capable of handling a certain number of business requests. For example, in practice, organizations can rationally divide their coverage area and deploy a specified number of branch nodes in each area. By relying on these branch nodes to handle the business within their respective areas, the overall business operations within the coverage area can be effectively maintained.

[0003] Furthermore, to ensure the quality of business operations in each segmented area, it is often necessary to conduct reasonable and effective resource scheduling for each branch node within it, based on the total amount of operational resources corresponding to that area. For example, in some practical scenarios, it is usually necessary to coordinate all operational resources within the same segmented area as a whole and decide on the resource scheduling results for each branch node, thereby avoiding business interruptions caused by simultaneous opening or closing of different branch nodes.

[0004] To address the aforementioned issues, relevant technologies typically employ node business evaluation. Based on the historical business data of each branch node within a specified area, the actual business status of each branch node is evaluated. Specifically, taking a branch node as an example, relevant technologies use a pre-defined evaluation model to analyze the corresponding historical business data triggered by the branch node in each specified business dimension to obtain the corresponding evaluation score for the branch node in each business dimension. Based on this, and according to the dimension weights corresponding to each business dimension, the node evaluation value of the branch node is determined. Thus, based on the obtained node evaluation value, the resource scheduling result of the branch node is decided.

[0005] However, based on the above approach, the relevant management entities need to adjust the dimension weights corresponding to each business dimension according to their own experience. When there are many business dimensions, the above approach often requires a lot of waiting time, resulting in low efficiency in resource scheduling decisions in actual situations. Summary of the Invention

[0006] This application provides a resource scheduling intelligent decision-making method, apparatus, electronic device, and storage medium to improve the decision-making efficiency of resource scheduling.

[0007] In a first aspect, embodiments of this application provide a resource scheduling intelligent decision-making method, including:

[0008] Obtain historical feature data of the target node. The historical feature data includes at least the target node's historical business data and historical scheduling status, wherein the target node has at least one specified business function.

[0009] Based on a pre-defined target decision model, features are extracted from historical business data to obtain the business characteristics of the target node. Based on the historical scheduling status, the target scheduling status of the target node is predicted to obtain the corresponding state prediction probability.

[0010] When the state prediction probability is not less than the preset state probability threshold, the target scheduling state represented by the state prediction probability is taken as the resource scheduling result of the target node.

[0011] Secondly, embodiments of this application provide a resource scheduling intelligent decision-making device, comprising:

[0012] The acquisition module is used to acquire historical feature data of the target node. The historical feature data includes at least the target node's historical business data and historical scheduling status, wherein the target node has at least one specified business function.

[0013] The prediction module is used to extract features from historical business data based on a preset target decision model to obtain the business features of the target node, and to predict the target scheduling status of the target node based on the historical scheduling status to obtain the corresponding state prediction probability.

[0014] The determination module is used to take the target scheduling state represented by the state prediction probability as the resource scheduling result of the target node when the state prediction probability is not less than a preset state probability threshold.

[0015] In an optional embodiment, before acquiring the historical feature data of the target node, the acquisition module is further configured to:

[0016] Obtain a training sample set, wherein each training sample includes: input information and entity label corresponding to a sample node; the sample node has at least one specified business function.

[0017] Using training samples from the training sample set, the pre-defined decision model is trained iteratively through multiple rounds. When the pre-defined convergence condition is met, the converged decision model is output. During one round of iterative training, the following operations are performed:

[0018] A decision model is used to obtain prediction results based on input information in training samples, and the parameters of the decision model are adjusted based on the loss value between the prediction results and the corresponding entity labels.

[0019] In one alternative embodiment, the input information of a training sample is obtained in the following manner:

[0020] Obtain sample feature data of a sample node, wherein the sample feature data includes at least: sample business data and sample scheduling status of a sample node.

[0021] Based on preset partitioning rules, the sample business data is partitioned, and based on the partitioning results, corresponding sample observation information is generated.

[0022] The sample observation information and sample scheduling status are combined to generate corresponding sample transfer information, which is then used as input information for a training sample.

[0023] In one optional embodiment, when dividing the sample business data based on a preset quantile rule and generating corresponding sample observation information based on the division results, the acquisition module is specifically used for:

[0024] Based on the preset partitioning rules, the sample business data is partitioned to obtain the corresponding partitioned business data, and based on the preset quantile rules, the quantile interval mapped by each partitioned business data is determined.

[0025] Based on the determined quantile intervals, the quantile characteristic values ​​of each segmented business data are determined respectively.

[0026] The quantile features are combined to generate corresponding sample observation information.

[0027] In an optional embodiment, when adjusting the parameters of the decision model based on the loss value between the prediction result and the corresponding entity label, the acquisition module is specifically used for:

[0028] The maximum likelihood estimation (MLE) method is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value.

[0029] or,

[0030] The expectation-maximum algorithm (EM) is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value.

[0031] In one optional embodiment, after outputting the convergent decision model, the acquisition module is further configured to:

[0032] Based on the training samples in the training sample set, the convergent decision model is trained iteratively for multiple rounds, and the target decision model is output when the preset correlation constraints are met. During one round of iterative training, the following operations are performed:

[0033] A convergent decision model is adopted to obtain test results based on the input information in the training samples, and the parameters of the convergent decision model are adjusted based on the loss value between the test results and the corresponding entity labels.

[0034] Thirdly, embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and when the computer program is executed by the processor, the processor enables the processor to implement any of the resource scheduling intelligent decision-making methods described in the first aspect above.

[0035] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the resource scheduling intelligent decision-making method of the first aspect.

[0036] Fifthly, embodiments of this application also provide a computer program product that, when invoked by a computer, causes the computer to execute the method described in the first aspect.

[0037] This application provides an intelligent resource scheduling decision-making method. By inputting historical feature data into a trained target decision-making model, the target decision-making model predicts the target scheduling state of a target node and its corresponding state prediction probability based on its trained neural network. Furthermore, when the state prediction probability is not less than a preset state probability threshold, the target scheduling state is taken as the resource scheduling result of the target node. Based on the above method, the management object can quickly and accurately determine the resource scheduling result of the target node through the target scheduling state predicted by the target decision-making model, thereby avoiding a large amount of waiting time caused by dimension weight adjustment and further improving the efficiency of resource scheduling decision-making. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating a possible application scenario provided by an embodiment of this application;

[0039] Figure 2 This is a diagram illustrating the architecture of a resource scheduling decision-making system provided in an embodiment of this application.

[0040] Figure 3 A flowchart of a target decision model training method provided in an embodiment of this application;

[0041] Figure 4 A flowchart of an input information acquisition method provided in an embodiment of this application;

[0042] Figure 5 A flowchart of a resource scheduling decision-making method provided in an embodiment of this application;

[0043] Figure 6 A logical schematic diagram of a resource scheduling decision-making method provided in an embodiment of this application;

[0044] Figure 7 This is a schematic diagram of a resource scheduling decision-making scenario provided in an embodiment of this application;

[0045] Figure 8 A schematic diagram of a resource scheduling decision-making device provided in an embodiment of this application;

[0046] Figure 9 This is a schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0047] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this invention.

[0048] The inventive concept of this application is as follows:

[0049] In related technologies, a preset evaluation model is used to analyze the corresponding historical business data triggered by the branch node on each specified business dimension. This method requires the management object to adjust the dimension weights corresponding to each business dimension based on its own experience. When there are many business dimensions, the above method often requires a lot of waiting time, resulting in low decision efficiency for resource scheduling in actual situations.

[0050] To improve the decision-making efficiency of resource scheduling, this application provides an intelligent decision-making method for resource scheduling. By inputting historical feature data into a trained target decision-making model, the model predicts the target scheduling state of a target node and its corresponding state prediction probability based on its trained neural network. Furthermore, when the state prediction probability is not less than a preset state probability threshold, the target scheduling state is taken as the resource scheduling result of the target node. Based on this method, the management object can quickly and accurately determine the resource scheduling result of the target node through the target scheduling state predicted by the target decision-making model, thereby avoiding the significant waiting time caused by dimension weight adjustments and further improving the efficiency of resource scheduling decisions.

[0051] For ease of understanding, the following explanations are provided for some of the nouns or terms that appear in the description of the embodiments of this application:

[0052] Resources refer to all material, financial, and human resources within a region. Specifically, they can be divided into two main categories: natural resources and social resources. Natural resources include material resources such as sunlight, air, water, and forests; social resources include human resources, information resources, and various material wealth created through labor.

[0053] For example, in the embodiments of this application, the operational resources in each divided area can be represented as human resources, information resources or other optional material resources deployed by the relevant organization in that area; in some more specific scenarios, they can also refer to the relevant activities, plans, etc. carried out by the relevant organization in that area. The above scenarios are only examples and will not be elaborated here.

[0054] Branch Node: An independent business unit that constitutes a large organization. Specifically, in the embodiments of this application, a node can be understood as an actual carrier capable of independently executing one or more business functions, such as a department. In some more specific scenarios, it can also refer to the actual business carrier unit corresponding to the relevant organization at the geographical level, such as a business hall.

[0055] Resource scheduling refers to the process of effectively allocating, analyzing, or using schedulable job resources within a certain time frame and in a relevant scenario.

[0056] For example, in this embodiment of the application, the process of coordinating all operational resources within the same partitioned area can be regarded as a resource scheduling for each branch node within that partitioned area. In some more specific scenarios, it can also be regarded as the allocation of operational resources based on the actual operational status of the branch nodes. For example, if the aforementioned branch nodes represent business carrying units, then the resource scheduling in this scenario can be regarded as the coordinated allocation of their associated human resources, information resources, or other material resources, including but not limited to personnel scheduling and permission allocation.

[0057] Based on the above explanations of terms and concepts, the intelligent resource scheduling decision-making method provided in the embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0058] See Figure 1 The diagram illustrates a possible application scenario, which includes multiple branch nodes (branch node A to branch node D) belonging to the same partitioned area. Each branch node has at least one designated business function. In an optional embodiment, the branch node can represent a specific business institution or department, such as a bank branch. In the above application scenario, the one or more business functions of each branch node can be characterized as the various business functions that a bank branch should possess.

[0059] For further details, please refer to [link / reference]. Figure 2 As shown, the management platform can connect to the node devices associated with each branch node based on the created data transmission channel, and collect and organize the historical feature data of each branch node. Based on the collected historical feature data, a corresponding target decision model is constructed, and the resource scheduling decision method proposed in this application embodiment is implemented. The data transmission channel includes, but is not limited to, PSTN, LAN, WAN, etc. Furthermore, the management platform can be represented as a computer terminal or similar computing device, such as a server cluster, for recording, storing and managing the historical feature information of each branch node.

[0060] It is worth noting that the above method is only an example. In an optional embodiment, data interoperability between various branch nodes can also be achieved based on blockchain, cloud technology, etc., so that a corresponding target decision model can be built in each branch node, and the resource scheduling decision method mentioned above can be implemented. Similarly, the above target decision model can also be deployed on other designated related devices, which will not be elaborated here. In this embodiment, for ease of description, the above management platform is used as an example for the main description.

[0061] Based on the above infrastructure, see Figure 3 As shown, the management platform can obtain the aforementioned target decision model based on the acquired historical feature data and using a Hidden Markov Model architecture. Specifically, this includes the following steps:

[0062] S301: Obtain the training sample set.

[0063] Specifically, in the early stages of model training, data should be collected from multiple sample nodes with the same business functions to obtain the sample feature data of each sample node, corresponding to one or more specified business functions that the target node should have. Each sample feature data includes the sample business data and sample scheduling status of the corresponding sample node.

[0064] For example, if branch nodes A through D all represent business halls, then each of these branch nodes, or other "business halls" with the same business functions, can be selected as the aforementioned sample nodes. For ease of description, in this embodiment, each of these branch nodes is used as a sample node during model training. Furthermore, data is collected from each sample node to determine the corresponding sample business data and sample scheduling status of each sample node, which serve as the sample feature data for each sample node. For example, assuming that the sample feature data collected by each of these branch nodes at a certain moment within a specified time window T is shown in Table 1 below:

[0065] Table 1

[0066] Sample Nodes Sample business data Sample scheduling status Branch node A Business Data A Scheduled Branch node B Business Data B Unscheduled Branch node C Business Data C Scheduled Branch node D Business Data D Unscheduled

[0067] Furthermore, data processing is performed on the feature data of each sample to obtain a training sample set determined for the preset decision model. In one optional embodiment, each training sample may correspond to the feature data collected by a sample node. Therefore, in the above data processing process, corresponding data processing can be performed on multiple determined sample nodes separately, thereby improving the sample processing efficiency. For details, please refer to... Figure 4 As shown, the data processing in a sample node may include:

[0068] S3011: Obtain the sample feature data of a sample node.

[0069] Specifically, within a specified time window T, sample service data for a sample node is collected, and the corresponding sample scheduling status is determined. Optionally, based on preset preprocessing rules, the aforementioned sample feature data is preprocessed, specifically including:

[0070] 1. Preprocessing of continuous data.

[0071] In practice, sample business data often includes continuous business data. For example, in the aforementioned business hall scenario, it is often necessary to consider the actual business situation of the sample node based on continuous business data such as total business volume, queuing time, and key business volume. Therefore, preprocessing of this type of business data may include:

[0072] 1) Abnormal data preprocessing.

[0073] Specifically, for the continuous business data in the above sample business data, based on the preset continuous anomaly handling rules, corresponding preprocessing is performed. For example, in an optional embodiment, an anomaly truncation threshold can be set based on the sample values ​​corresponding to the above continuous business data, so as to truncate the abnormal data with large differences in the sample business data, thereby avoiding the adverse effects of the abnormal data in the sample business data on the subsequent model training.

[0074] 2) Preprocessing of missing data.

[0075] Specifically, for the continuous business data in the above sample business data, corresponding preprocessing is performed based on the preset continuous missing data handling rules. For example, in an optional embodiment, the missing data contained in the sample business data can be filled based on the sample mean corresponding to the above continuous business data, thereby avoiding the adverse effects of the missing data in the sample business data on subsequent model training.

[0076] 3) Standardized preprocessing.

[0077] Specifically, since the data volume corresponding to each business dimension is often different in actual situations, in order to ensure the standardization of sample business data, it is mapped to the preset standardized range [0,1] to improve the computational efficiency of the model.

[0078] 2. Preprocessing of discrete data.

[0079] Similarly, in practice, sample business data often includes discrete business data. For example, in the aforementioned business hall scenario, the criteria for evaluating sample nodes often include: whether it is a key node, whether it is an urban node, etc. Therefore, preprocessing this type of business data can include:

[0080] 1) Abnormal data preprocessing.

[0081] Specifically, for the discrete business data present in the above sample business data, based on the preset discrete anomaly handling rules, corresponding preprocessing is performed. For example, in an optional embodiment, there may be a situation where the category of a certain abnormal data cannot be determined. In this case, based on the above discrete anomaly handling rules, the frequency of occurrence of the abnormal data in the sample business data can be counted. If the frequency of occurrence is greater than or equal to the preset abnormal data threshold, the abnormal data is treated as a separate category; otherwise, the category of the abnormal data is determined to be the business category to which the business data with the highest frequency of occurrence in the sample business data belongs.

[0082] 2) Preprocessing of missing data.

[0083] Specifically, for the discrete business data present in the above sample business data, based on the preset discrete missing data handling rules, corresponding preprocessing is performed. For example, in an optional embodiment, for the data missing situation in the sample business data, the frequency of occurrence of the missing data in the sample business data can be counted. If the frequency of the attribute is greater than or equal to the preset missing data threshold, the missing value is treated as a separate category; otherwise, the category to which the abnormal data belongs is determined to be the business category to which the business data with the highest frequency in the sample business data belongs.

[0084] The above preprocessing rules are merely illustrative examples. In this embodiment, taking the business data A corresponding to the branch node A as an example, the business data A is represented as [L A ,R A ], where L A Representing the continuous business data contained in business data A, R AIf discrete business data is contained in business data A, then business data A can be processed accordingly based on the above preprocessing rules to ensure the accuracy of subsequent model training.

[0085] S3012: Based on preset partitioning rules, the sample business data is partitioned, and based on the partitioning results, corresponding sample observation information is generated.

[0086] Furthermore, in a Hidden Markov Model, the actual business status of each sample node is often represented as a corresponding observation sequence and state sequence. Therefore, during the training process of the target decision model, it is also necessary to map the corresponding sample business data to an observation element in the model. In an optional embodiment, binning can be used to achieve the mapping of the above observation element. After determining the sample business data of the above sample node, the sample business data is divided using a preset partitioning rule, thereby generating corresponding sample observation information based on the partitioning result.

[0087] Specifically, in one optional embodiment, the numerical range corresponding to the continuous business data can be further divided, setting multiple different quantile intervals to achieve the mapping of sample business data to corresponding observation elements. Optionally, a fixed quantile method can be used to divide the above-mentioned sample business data and determine the quantile interval mapped to each division of business data. Then, based on the quantile feature values ​​corresponding to each determined quantile interval, the corresponding observation elements are combined and mapped as the corresponding sample observation information.

[0088] For better illustration, let's still use the business data A = [L] corresponding to the above branch node A. A ,R A For example, L A Represented as continuous business data contained in business data A, R A The discrete business data contained in business data A is represented as follows: Assume that the continuous business types considered by branch node A are: {total business volume, queuing time, number of employees, key business volume, and passenger flow}; and the discrete business types considered by A are: {whether it is a key node or a city node}; then the above business data A can be divided based on the above 5 continuous business types and 2 discrete business types.

[0089] Furthermore, for each of the aforementioned business types, fixed quantile binning is performed. For example, for each continuous business type, binning is performed using a fixed quintile method. This allows the quantile interval to be determined based on the actual data volume corresponding to each partition of business data, as shown in Table 2 below:

[0090] Table 2

[0091] Divide business data Business type Actual data volume quantile interval Divide business data 1 Total business volume 0.12 [0,0.2] Divide business data 2 Queueing time 0.27 [0.2,0.4] Divide business data into 3 Number of employees 0.18 [0,0.2] Divide business data into 4 Key business volume 0.47 [0.4,0.6] Divide business data into 5 Passenger flow 0.15 [0,0.2]

[0092] Based on the above method, one-hot encoding can be performed on the continuous business data based on the quantile interval mapped to each partitioned business data and the corresponding quantile feature value. For example, in the binning process with a fixed quantile of 5, each partitioned business data can be mapped to a unique quantile feature value. Then, based on the obtained quantile feature values, the sample business data can be mapped to a corresponding observation element, which can be represented as O A , of which O A This is the feature matrix corresponding to the aforementioned fixed quantiles.

[0093] It is worth noting that the above fixed quantiles are only illustrative examples. In actual situations, the fixed quantiles used can be flexibly adjusted based on factors such as the number of business types and the model training accuracy, which will not be elaborated further here.

[0094] S3013: Combine sample observation information and sample scheduling status to generate corresponding sample transfer information, and use the sample transfer information as input information for a training sample.

[0095] Furthermore, the sample observation information and sample scheduling status are combined as input information for a preset decision model. The sample scheduling status includes two states: scheduled and unscheduled. These two scheduling states can be encoded and represented as a state element in the Hidden Markov Model.

[0096] S302: Using training samples from the training sample set, perform multiple rounds of iterative training on the preset decision model, and output the converged decision model when the preset convergence condition is met.

[0097] Specifically, during one round of iterative training, the following operations are performed: using a decision model, based on the input information in the training samples, to obtain prediction results, and adjusting the parameters of the decision model based on the loss value between the prediction results and the corresponding entity labels.

[0098] Since each input information can be understood as an observation element and a state element in the Hidden Markov Model, the combination of the above input information can be understood as the observation sequence and state sequence determined for each of the above business types in the Hidden Markov Model, respectively expressed as:

[0099] O' = (O1, O2, ..., O T )

[0100] Q'=(Q1,Q2,...,Q T )

[0101] Where O' represents the corresponding observation sequence and Q' represents the corresponding state sequence.

[0102] Based on the above method, the corresponding observation sequence and state sequence are determined. The initial scheduling probability distribution of this decision model can then be expressed as: π = P(Q1); the state transition probability distribution can be expressed as: A = P(Q1). T |Q T-1 The observation probability distribution can be expressed as: B = P(O) T |Q T If the above decision model can be represented as a hidden Markov model λ determined based on the above parameters, then λ = (A, B, π).

[0103] Optionally, the parameters of the above decision model can be learned using any of the following methods:

[0104] Method 1: Use the maximum likelihood estimation method to determine the loss value between the prediction result and the corresponding entity label, and adjust the parameters of the decision model based on the loss value.

[0105] Method 2: Use the expectation-maximum algorithm to determine the loss value between the prediction result and the corresponding entity label, and adjust the parameters of the decision model based on the loss value.

[0106] Based on the above method, a preset decision model can be trained iteratively in multiple rounds, and a converged decision model can be output when the preset convergence conditions are met.

[0107] Furthermore, optionally, to improve the accuracy of the resource scheduling results output by the above decision model, after obtaining the corresponding convergent decision model, the convergent decision model can be further iterated and trained multiple times based on the training samples in the above training sample set. It should be understood that the training samples for the convergent decision model can be either the training samples used in the above training process or the corresponding training samples output after other sample feature data have been processed by the above method.

[0108] For example, in an optional embodiment, the sample feature data with a time window of T can be time-segmented so that after training the decision model based on the training sample data corresponding to the first half of the segmented time, the training sample data corresponding to the second half of the segmented time can be used to test the convergent decision model, thereby further ensuring the accuracy of the model's output resource scheduling results.

[0109] See Figure 5 As shown, based on the above method, after obtaining the target decision model, this application embodiment provides a resource scheduling decision-making method, specifically including:

[0110] S501: Obtain historical feature data of the target node.

[0111] Specifically, assuming that the above-mentioned branch node A is the designated target node, the historical characteristic data of the target node should include: the historical business data and historical scheduling status of the branch node. In actual situations, in order to ensure the accuracy of resource scheduling results, it is also necessary to determine the historical time corresponding to the historical business data and the target time corresponding to the resource scheduling results. The corresponding time span should meet the preset association conditions.

[0112] For example, in a real-world scenario, if it is necessary to obtain the resource scheduling results of the target node at time t, the historical business data and historical scheduling status of the target node at time t-1 can be input into the target decision model to ensure the accuracy of the input resource scheduling results.

[0113] S502: Based on the preset target decision model, feature extraction is performed on historical business data to obtain the business characteristics of the target node, and based on the historical scheduling status, the target scheduling status of the target node is predicted to obtain the corresponding state prediction probability.

[0114] Specifically, the target decision model determines the business characteristics of the target node at a historical moment based on the input historical feature data, and predicts the determined target scheduling state based on the business characteristics and historical scheduling state, thereby obtaining the state prediction probability determined by the target decision model for the target scheduling state.

[0115] For example, based on the historical feature data input above, the target decision model determines that at time t-1, the business feature of the target node is O. t-1 And determine the historical scheduling state as Q. t-1 Then, for time t, the state prediction probability of the target scheduling state can be equivalently expressed as: P(Q t |Q t-1 O t-1 ).

[0116] S503: When the state prediction probability is not less than the preset state probability threshold, the target scheduling state represented by the state prediction probability is used as the resource scheduling result of the target node.

[0117] Furthermore, after the target decision model outputs the predicted state probability for the target scheduling state based on its trained model parameters, it is determined whether this predicted state probability meets a preset state probability threshold. The most likely resource scheduling state at time t is then determined based on the magnitude of the predicted state probability. This approach ensures the fairness and accuracy of resource scheduling decisions while avoiding the significant waiting time caused by dimensional weight adjustments, thus further improving the efficiency of resource scheduling decisions.

[0118] See Figure 6The diagram illustrates the logic of the resource scheduling decision-making method described above. In this diagram, historical feature data obtained by the target node at time t-1 is input into a pre-trained target decision-making model. This model determines the predicted state probability of the target node for the target scheduling state based on the extracted business features and the corresponding historical scheduling states. When the predicted state probability is not less than a preset state probability threshold, the target scheduling state represented by the predicted state probability is taken as the resource scheduling result of the target node. Based on this method, the management object can quickly and accurately obtain the corresponding resource scheduling result without performing too many operations, thereby significantly improving the decision-making efficiency of resource scheduling and effectively enhancing the refined management level of the relevant organization.

[0119] See Figure 7 The diagram illustrates a specific scenario example of the resource scheduling decision-making method described above. In this scenario, assuming the target node is branch node A, on March 1, 2022, the current business data of branch node A, along with its current scheduling status (unscheduled), is input into the preset target decision model. The target decision model then determines the state prediction probability P for March 2, 2022, based on the business feature X extracted from the current business data and its corresponding current scheduling status (unscheduled). Assuming this state prediction probability corresponds to the target scheduling status (scheduled), the above process is equivalent to solving the conditional probability P(scheduled|unscheduled, X). Assuming the state prediction probability P(scheduled|unscheduled, X) output by the target decision model is 0.7, and the preset state probability threshold is 0.5, the managed object can determine, based on the above method, that the resource scheduling result for the branch node on March 2, 2022, is "scheduled," thereby quickly and accurately obtaining the corresponding resource scheduling result, ensuring the efficiency of resource scheduling decisions and the refined management level of the relevant organization.

[0120] See Figure 8 As shown, this application embodiment provides a resource scheduling intelligent decision-making device, including an acquisition module 801, a prediction module 802, and a determination module 803, wherein:

[0121] The acquisition module 801 is used to acquire historical feature data of the target node. The historical feature data includes at least the historical business data and historical scheduling status of the target node, wherein the target node has at least one specified business function.

[0122] The prediction module 802 is used to extract features from historical business data based on a preset target decision model to obtain the business features of the target node, and to predict the target scheduling state of the target node based on the historical scheduling state to obtain the corresponding state prediction probability.

[0123] The determination module 803 is used to take the target scheduling state represented by the state prediction probability as the resource scheduling result of the target node when the state prediction probability is not less than the preset state probability threshold.

[0124] In an optional embodiment, before acquiring the historical feature data of the target node, the acquisition module 801 is further configured to:

[0125] Obtain a training sample set, wherein each training sample includes: input information and entity label corresponding to a sample node; the sample node has at least one specified business function.

[0126] Using training samples from the training sample set, the pre-defined decision model is trained iteratively through multiple rounds. When the pre-defined convergence condition is met, the converged decision model is output. During one round of iterative training, the following operations are performed:

[0127] A decision model is used to obtain prediction results based on input information in training samples, and the parameters of the decision model are adjusted based on the loss value between the prediction results and the corresponding entity labels.

[0128] In one alternative embodiment, the input information of a training sample is obtained in the following manner:

[0129] Obtain sample feature data of a sample node, wherein the sample feature data includes at least: sample business data and sample scheduling status of a sample node.

[0130] Based on preset partitioning rules, the sample business data is partitioned, and based on the partitioning results, corresponding sample observation information is generated.

[0131] The sample observation information and sample scheduling status are combined to generate corresponding sample transfer information, which is then used as input information for a training sample.

[0132] In an optional embodiment, when dividing the sample business data based on a preset quantile rule and generating corresponding sample observation information based on the division results, the acquisition module 801 is specifically used for:

[0133] Based on the preset partitioning rules, the sample business data is partitioned to obtain the corresponding partitioned business data, and based on the preset quantile rules, the quantile interval mapped by each partitioned business data is determined.

[0134] Based on the determined quantile intervals, the quantile characteristic values ​​of each segmented business data are determined respectively.

[0135] The quantile features are combined to generate corresponding sample observation information.

[0136] In an optional embodiment, when adjusting the parameters of the decision model based on the loss value between the prediction result and the corresponding entity label, the acquisition module 801 is specifically used for:

[0137] The maximum likelihood estimation (MLE) method is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value.

[0138] or,

[0139] The expectation-maximum algorithm (EM) is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value.

[0140] In an optional embodiment, after outputting the convergent decision model, the acquisition module 801 is further configured to:

[0141] Based on the training samples in the training sample set, the convergent decision model is trained iteratively for multiple rounds, and the target decision model is output when the preset correlation constraints are met. During one round of iterative training, the following operations are performed:

[0142] A convergent decision model is adopted to obtain test results based on the input information in the training samples, and the parameters of the convergent decision model are adjusted based on the loss value between the test results and the corresponding entity labels.

[0143] Based on the same inventive concept as the embodiments described above, this application also provides an electronic device that can be used for intelligent resource scheduling decisions. In one embodiment, the electronic device can be a server, a terminal device, or other electronic devices. In this embodiment, the structure of the electronic device can be as follows: Figure 9 As shown, it includes a memory 901, a communication interface 903, and one or more processors 902.

[0144] The memory 901 is used to store computer programs executed by the processor 902. The memory 901 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0145] Memory 901 may be volatile memory, such as random-access memory (RAM); memory 901 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 901 may be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 901 may be a combination of the above-mentioned memories.

[0146] The processor 902 may include one or more central processing units (CPUs) or digital processing units, etc. The processor 902 is used to implement the aforementioned intelligent resource scheduling decision-making method when calling computer programs stored in the memory 901.

[0147] Communication interface 903 is used to communicate with terminal devices and other servers.

[0148] This application embodiment does not limit the specific connection medium between the memory 901, the communication interface 903, and the processor 902. This application embodiment... Figure 9 The memory 901 and the processor 902 are connected via a bus 904, which is in... Figure 9 The connections between other components are shown in thick lines and are for illustrative purposes only, not as limiting information. The 904 bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0149] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform any of the resource scheduling intelligent decision-making methods described in the above embodiments. The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0150] According to one aspect of this application, this application also provides a computer program product that, when invoked by a computer, causes the computer to perform the method as described in the first aspect.

[0151] This application provides an intelligent resource scheduling decision-making method. By inputting historical feature data into a trained target decision-making model, the target decision-making model predicts the target scheduling state of a target node and its corresponding state prediction probability based on its trained neural network. Furthermore, when the state prediction probability is not less than a preset state probability threshold, the target scheduling state is taken as the resource scheduling result of the target node. Based on the above method, the management object can quickly and accurately determine the resource scheduling result of the target node through the target scheduling state predicted by the target decision-making model, thereby avoiding a large amount of waiting time caused by dimension weight adjustment and further improving the efficiency of resource scheduling decision-making.

[0152] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0153] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0155] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A resource scheduling intelligent decision-making method, characterized in that, include: Obtain a training sample set, wherein each training sample includes: input information and entity label corresponding to a sample node; the sample node has at least one specified business function; the input information of a training sample is obtained in the following way: Obtain sample feature data of a sample node, wherein the sample feature data includes at least: sample service data and sample scheduling status of the sample node; Based on preset partitioning rules, the sample business data is partitioned, and based on the partitioning results, corresponding sample observation information is generated. The sample observation information and the sample scheduling status are combined to generate corresponding sample transfer information, and the sample transfer information is used as the input information for a training sample. Using training samples from the aforementioned training sample set, a preset decision model is trained iteratively through multiple rounds. When a preset convergence condition is met, a converged decision model is output. During one round of iterative training, the following operations are performed: Using the aforementioned decision model, prediction results are obtained based on input information from training samples, and the parameters of the decision model are adjusted based on the loss value between the prediction results and the corresponding entity labels. Obtain historical feature data of the target node, wherein the historical feature data includes at least: historical business data and historical scheduling status of the target node, and the target node has at least one specified business function; Based on a preset target decision model, feature extraction is performed on the historical business data to obtain the business characteristics of the target node, and based on the historical scheduling status, the target scheduling status of the target node is predicted to obtain the corresponding status prediction probability. When the state prediction probability is not less than a preset state probability threshold, the target scheduling state represented by the state prediction probability is used as the resource scheduling result of the target node; the target scheduling state is used to indicate whether the target node has been scheduled or the target node has not been scheduled.

2. The method as described in claim 1, characterized in that, The sample business data is divided based on a preset quantile rule, and corresponding sample observation information is generated based on the division results, including: Based on a preset partitioning rule, the sample business data is partitioned to obtain corresponding partitioned business data, and based on a preset quantile rule, the quantile interval mapped by each partitioned business data is determined. Based on the determined quantile intervals, the quantile feature values ​​of each segmented business data are determined respectively; The quantile feature values ​​are combined to generate corresponding sample observation information.

3. The method according to any one of claims 1-2, characterized in that, The step of adjusting the parameters of the decision model based on the loss value between the prediction result and the corresponding entity label includes: The maximum likelihood estimation (MLE) method is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value. or, The expectation-maximum algorithm (EM) is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value.

4. The method according to any one of claims 1-2, characterized in that, Following the output convergence decision model, it also includes: Based on the training samples in the training sample set, the convergent decision model is further trained through multiple rounds of iterative training, and the target decision model is output when the preset correlation constraints are met; wherein, during one round of iterative training, the following operations are performed: Using the convergent decision model, test results are obtained based on the input information in the training samples, and the parameters of the convergent decision model are adjusted based on the loss value between the test results and the corresponding entity labels.

5. A resource scheduling intelligent decision-making device, characterized in that, include: The acquisition module is used to acquire a training sample set, wherein each training sample includes: input information and entity labels corresponding to a sample node; the sample node has at least one specified business function; the input information of a training sample is obtained in the following way: Obtain sample feature data of a sample node, wherein the sample feature data includes at least: sample service data and sample scheduling status of the sample node; Based on preset partitioning rules, the sample business data is partitioned, and based on the partitioning results, corresponding sample observation information is generated. The sample observation information and the sample scheduling status are combined to generate corresponding sample transfer information, and the sample transfer information is used as the input information for a training sample. Using training samples from the aforementioned training sample set, a preset decision model is trained iteratively through multiple rounds. When a preset convergence condition is met, a converged decision model is output. During one round of iterative training, the following operations are performed: Using the aforementioned decision model, prediction results are obtained based on input information from training samples, and the parameters of the decision model are adjusted based on the loss value between the prediction results and the corresponding entity labels. The acquisition module is used to acquire historical feature data of the target node. The historical feature data includes at least: historical business data and historical scheduling status of the target node, wherein the target node has at least one specified business function. The prediction module is used to extract features from the historical business data based on a preset target decision model to obtain the business features of the target node, and to predict the target scheduling state of the target node based on the historical scheduling state to obtain the corresponding state prediction probability. The determination module is used to take the target scheduling state represented by the state prediction probability as the resource scheduling result of the target node when the state prediction probability is not less than a preset state probability threshold; the target scheduling state is used to indicate whether the target node has been scheduled or the target node has not been scheduled.

6. The apparatus as claimed in claim 5, characterized in that, The sample business data is divided based on a preset quantile rule, and corresponding sample observation information is generated based on the division results, including: Based on a preset partitioning rule, the sample business data is partitioned to obtain corresponding partitioned business data, and based on a preset quantile rule, the quantile interval mapped by each partitioned business data is determined. Based on the determined quantile intervals, the quantile feature values ​​of each segmented business data are determined respectively; The quantile feature values ​​are combined to generate corresponding sample observation information.

7. The apparatus according to any one of claims 5-6, characterized in that, The step of adjusting the parameters of the decision model based on the loss value between the prediction result and the corresponding entity label includes: The maximum likelihood estimation (MLE) method is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value. or, The expectation-maximum algorithm (EM) is used to determine the loss value between the prediction result and the corresponding entity label, and the parameters of the decision model are adjusted based on the loss value.

8. The apparatus according to any one of claims 5-6, characterized in that, Following the output convergence decision model, it also includes: Based on the training samples in the training sample set, the convergent decision model is further trained through multiple rounds of iterative training, and the target decision model is output when the preset correlation constraints are met; wherein, during one round of iterative training, the following operations are performed: Using the convergent decision model, test results are obtained based on the input information in the training samples, and the parameters of the convergent decision model are adjusted based on the loss value between the test results and the corresponding entity labels.

9. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the resource scheduling intelligent decision-making method as described in any one of claims 1-4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-4.

11. A computer program product, characterized in that, When the computer program product is invoked by a computer, it causes the computer to perform the method as described in any one of claims 1-4.

Images