Resource scheduling methods, equipment, media and program products

Abstract

This disclosure relates to the field of wireless communication technology, and in particular provides a resource scheduling method, device, medium, and program product. The resource scheduling method includes: acquiring multiple historical load index values ​​associated with multiple load indices for multiple cells within a historical time period; inputting the multiple historical load index values ​​associated with each load index for each cell within a historical time period into a load index prediction model to obtain a target load index value associated with each load index for each cell within a target time period; determining the matching score value between two cells in each cell group within the multiple cell groups during the target time period based on the target load index value and a load index threshold; determining the cell pairing result based on the matching score value set and the cell pairing strategy prediction model, and performing resource scheduling during the target time period based on the cell pairing result. This disclosure can improve resource utilization and the rationality of resource allocation.

Description

Technical Field

[0001] This disclosure relates to the field of wireless communication technology, and in particular to a resource scheduling method, apparatus, medium, and program product. Background Technology

[0002] With the rapid development of 5G commercial networks and the rapid increase in the penetration rate of 5G terminals, 5G is gradually becoming the main service carrier layer. Due to the differences in cell coverage scenarios, some cells may experience high network load at one time and low network load at another time. In order to effectively reuse network physical resources, cells with fluctuating high and low loads can be identified and scaled up or down.

[0003] In related technologies, it is common practice to obtain network load indicators and / or other information that can reflect the load status of each cell, and to expand the cell if the cell load is too high, or to shrink the cell if the cell load is too low, in order to perform resource scheduling.

[0004] However, the resource scheduling schemes provided in related technologies mainly schedule resources from the load dimension of a single cell. This often results in situations where high-load cells have no resources available and low-load cells have idle resources, leading to unreasonable resource allocation and even affecting normal business services. Summary of the Invention

[0005] This disclosure is made in view of the above-mentioned problems. This disclosure provides a resource scheduling method, apparatus, medium, and program product that can improve resource utilization and the rationality of resource allocation.

[0006] According to one aspect of this disclosure, a resource scheduling method is provided, comprising:

[0007] Obtain the historical load index values ​​of multiple cells within a historical time period, which are associated with multiple load indices respectively;

[0008] The historical load index values ​​associated with each load index in each cell during the historical period are input into the load index prediction model to obtain the target load index value associated with each load index in each cell during the target period.

[0009] Based on the target load index value and load index threshold associated with each load index in each cell during the target time period, the matching score value of two cells in each cell group in the multiple cell groups is determined during the target time period. The multiple cell groups are obtained by matching two cells in each of the multiple cells. The matching score value is used to characterize the degree of load difference between the two cells.

[0010] Based on the matching score set composed of the matching score values ​​of two cells in each cell group during the target time period, and the cell pairing strategy prediction model, the cell pairing result is determined, and resource scheduling is performed during the target time period based on the cell pairing result. The cell pairing result includes the paired cells determined for resource scheduling for each cell.

[0011] According to another aspect of this disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the resource scheduling method described above.

[0012] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the resource scheduling method described above.

[0013] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the resource scheduling method described above.

[0014] The resource scheduling method, equipment, medium, and program products provided in this disclosure obtain the target load index value associated with each load index for each cell in a target time period by using multiple historical load index values ​​associated with multiple cells in a historical time period, and a load index prediction model. Then, based on each target load index value associated with each load index and a load index threshold for each cell in the target time period, the load difference between two cells in a cell group determined by pairwise matching of multiple cells is determined. Based on the load difference between the two cells in the cell group and a cell pairing strategy prediction model, a paired cell for resource scheduling is matched for each cell. Resource scheduling is then performed based on the paired cells of each cell at the target time. By using the load difference values ​​of each pair of cells in the target time period, a paired cell for resource scheduling is determined for each cell in the target time period, thereby achieving reasonable allocation and scheduling of resources in multiple cells and improving the resource utilization rate of multiple cells.

[0015] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description

[0016] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 This is a schematic diagram illustrating an implementation scenario of a resource scheduling scheme according to an embodiment of this disclosure.

[0018] Figure 2 This is a flowchart illustrating resource scheduling according to an embodiment of the present disclosure.

[0019] Figure 3 This is a flowchart further illustrating the training of a cell pairing strategy prediction model in resource scheduling according to an embodiment of the present disclosure.

[0020] Figure 4 This is a block diagram illustrating a resource scheduling apparatus according to an embodiment of the present disclosure.

[0021] Figure 5 This is a schematic diagram illustrating a computer program product according to an embodiment of the present disclosure.

[0022] Figure 6 This is a hardware block diagram illustrating an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0024] In related technologies, during resource scheduling, after obtaining network load indicators and / or other information that can reflect the load status of each cell, if it is determined that the cell load is too high, the cell is expanded; or if it is determined that the cell load is too low, the cell is reduced in size, in order to perform resource scheduling.

[0025] However, this resource scheduling scheme generally schedules resources for each cell sequentially from the load dimension of a single cell. For high-load cells, resources usually need to be expanded. However, the premise of resource expansion is that resources are available. When there are no available resources in a high-load cell, resource scheduling identification will occur. For low-load cells, after reducing the resource occupancy of the cell, the excess resources are idle, resulting in resource idleness and unreasonable resource scheduling.

[0026] To address the aforementioned problems, embodiments of this disclosure provide a resource scheduling scheme, such as... Figure 1 As shown, Figure 1 This illustration shows a schematic diagram of an implementation scenario for a resource scheduling scheme provided by an exemplary embodiment of this disclosure. For example... Figure 1 As shown, the implementation scenario 100 includes a network device 101 and a resource scheduling device 102. The network device 101 is a device that can monitor the indicator data of the cell, and the resource scheduling device 102 can be a computer, laptop, tablet or server, etc.

[0027] The network device 101 and the resource scheduling device 102 can establish a communication link to implement the resource scheduling scheme provided in this embodiment.

[0028] Figure 2 A flowchart illustrating an exemplary embodiment of the present disclosure of a resource scheduling method is shown. This method can be applied to a resource scheduling device, such as... Figure 2 As shown, the method in this embodiment of the disclosure may include:

[0029] Step S201: Obtain multiple historical load index values ​​of multiple cells that are associated with multiple load indexes within a historical time period;

[0030] Step S202: Input the multiple historical load index values ​​associated with each load index in the historical time period of each cell into the load index prediction model to obtain the target load index value associated with each load index in the target time period of each cell.

[0031] Step S203: Based on the target load index value and load index threshold associated with each load index for each cell in the target time period, determine the matching score value of two cells in each cell group in the target time period.

[0032] Among them, multiple cell groups are obtained by matching two cells from multiple cells, and the matching score is used to characterize the degree of load difference between the two cells;

[0033] Step S204: Based on the matching score set composed of the matching score values ​​of the two cells in each cell group during the target time period, and the cell pairing strategy prediction model, determine the cell pairing result, and perform resource scheduling during the target time period based on the cell pairing result.

[0034] The cell pairing results include the paired cells determined for resource scheduling for each cell.

[0035] In summary, the resource scheduling method provided in this disclosure obtains the target load index value associated with each load index for each cell in the target time period by using multiple historical load index values ​​associated with multiple cells in historical time periods and a load index prediction model. Then, based on each target load index value associated with each load index for each cell in the target time period and the load index threshold, the load difference between two cells in a cell group determined by pairwise matching of multiple cells is determined. Based on the load difference between the two cells in the cell group and the cell pairing strategy prediction model, a paired cell for resource scheduling is matched for each cell. Resource scheduling is then performed based on the paired cells of each cell at the target time. By using the load difference values ​​of each pair of cells in the target time period, a paired cell for resource scheduling is determined for each cell in the target time period, thereby achieving reasonable allocation and scheduling of resources across multiple cells and improving resource utilization of multiple cells.

[0036] The following are Figure 2 The specific implementation methods of each step in the illustrated embodiment are described in detail below:

[0037] In step S201, the resource scheduling device can obtain multiple historical load index values ​​of multiple cells that are associated with multiple load indexes within a historical period.

[0038] In this embodiment of the disclosure, the resource scheduling device can periodically schedule resources of multiple cells. Specifically, it can be determined based on actual needs, and this embodiment of the disclosure does not limit this. For example, the resource scheduling device can perform resource scheduling once every one or two days.

[0039] It should also be noted that the historical time period is a preset duration prior to the current resource scheduling time. Specifically, it can be determined based on actual needs, and this embodiment does not limit it. The historical time period may include multiple unit durations, each of which is determined based on actual needs, and this embodiment does not limit it. For example, the historical time period can be 7 days, and the unit duration can be 1 day; or the historical time period can be 3 days, and the unit duration can be 12 hours. The duration of the resource scheduling cycle is greater than or equal to the unit duration within the historical time period, which can prevent the process of repeatedly determining cell pairing results and prevent waste of resources.

[0040] Load metrics are parameters used to characterize the cell load status. Specifically, they can be determined based on actual needs, and this disclosure does not limit this. For example, load metrics may include: average evolved radio access bearer (E-RAB) traffic, number of radio resource control (RRC) channels with data transmission, physical uplink shared channel (PUSCH) utilization, physical downlink shared channel (PDSCH) utilization, physical downlink control channel (PDCCH) utilization, uplink traffic and / or downlink traffic, etc. Optionally, load metrics can be different for different services. In this disclosure, resource scheduling can be performed for different services, and multiple load metrics are load metrics associated with the target services that need to be resource-scheduled.

[0041] In one optional implementation, the process by which the resource scheduling device obtains multiple historical load indicator values ​​associated with multiple load indicators for multiple cells within a historical time period may include: sending a load data acquisition request to a network device, and obtaining the multiple historical load indicator values ​​associated with multiple load indicators for the cells within a historical time period returned by the network device. This allows for real-time acquisition of historical load indicator data through the network device, improving the efficiency of historical load indicator value acquisition.

[0042] In one optional implementation, the process by which the resource scheduling device obtains multiple historical load index values ​​of multiple cells that are associated with multiple load indicators in a historical period may include: reading multiple historical load index values ​​of multiple cells that are associated with multiple load indicators in a historical period from the resource scheduling device.

[0043] It is understood that, in this embodiment of the disclosure, the multiple historical load index values ​​associated with multiple load indicators respectively include: the historical load index value of each load indicator associated with each unit duration of a historical time period.

[0044] For example, if the historical period is 7 days and the unit duration can be 1 day, and multiple load metrics include: E-RAB traffic, PUSCH utilization, and PDCCH utilization, then the multiple historical load metric values ​​associated with each of the multiple load metrics include: historical E-RAB traffic, PUSCH utilization, and PDCCH utilization associated with each day within the 7 days.

[0045] In step S202, the resource scheduling device inputs multiple historical load index values ​​associated with each load index in the historical time period of each cell into the load index prediction model to obtain the target load index value associated with each load index in the target time period of each cell.

[0046] In this embodiment of the disclosure, the load index prediction model is a time series model. Specifically, it can be determined based on actual needs, and this embodiment of the disclosure does not limit it. For example, the load index prediction model can be a Long Short Term Memory Network (LSTM) model, or an Auto-Regression and Moving Average (ARMA) model, etc.; the target time period is later than the historical time period, and the target time period can include at least one unit of time, wherein the time length of each unit of time included in the target time period is the same as the time length of each unit of time included in the historical time period.

[0047] In one optional implementation, the process of training a load index prediction model for a resource scheduling device may include: acquiring a sample dataset; then, dividing the sample dataset into a training set and a validation set; training a time series network using the training set; validating the time series network using the validation set after the time series network converges; and obtaining the load index prediction model if the validation passes. The sample dataset includes: multiple historical load index values ​​associated with multiple load indices for each cell within a sample historical time period; and load index label values ​​associated with multiple load indices for each cell within a sample target time period, wherein the sample target time period is later than the sample historical time period.

[0048] For example, taking the LSTM model as the load index prediction model, the process of training the load index prediction model for resource scheduling equipment may include: collecting sample historical load index values ​​of n types of load indices associated with M units of time within the target duration; shifting the sample historical load index values ​​of each load index Fi associated with M units of time within the target duration using a time window of length T+S to obtain MT-S+1 samples associated with each load index Fi; and combining the MT-S+1 samples associated with each load index Fi to obtain a sample dataset, where the sample dataset is (MT-S+1, T+S, n), the number of samples associated with each load index in the sample dataset is MT-S+1, and the time step of each sample is T+S; where the time step T is the duration of the sample historical period, indicating that the sample historical period contains T units of time, the time step S is the duration of the sample target period, indicating that the sample target period contains S units of time, and the feature dimension of each sample is n, indicating that it contains n types of load indices, with the value of i ranging from 1 to n.

[0049] Furthermore, the sample dataset is split into a training set tensor_x and a validation set tensor_y. The mean squared error is used as the loss function, and the LSTM model is trained using the backpropagation algorithm in the training set tensor_x. After the model converges, the prediction performance of the trained LSTM model is verified using the validation set tensor_y. The validation metric can be the mean absolute percentage error.

[0050] The process of training an LSTM model using the mean squared error as the loss function and employing the backpropagation algorithm in the training set tensor_x may include: inputting multiple historical load index values ​​for each load index within the historical sample time period from the training set into the LSTM model to obtain the predicted load index value for the target sample time period; then, inputting the predicted load index value and the load index label value for the target sample time period into the loss function to obtain the loss function value for the current iteration; repeating the above iterative process if the loss function value is greater than or equal to a loss function value threshold to obtain the LSTM model; the loss function value threshold can be determined based on actual needs, and this embodiment does not limit it.

[0051] In an optional implementation, when the target time period includes a unit duration, the process of the resource scheduling device inputting multiple historical load index values ​​associated with each load index for each cell in the historical time period into the load index prediction model to obtain the target load index value associated with each load index for each cell in the target time period may include: inputting multiple historical load index values ​​associated with each load index for each cell in the historical time period into the load index prediction model to obtain the target load index value associated with each load index for each cell in the target time period.

[0052] In an optional implementation, when the target time period includes multiple unit durations, the resource scheduling device inputs multiple historical load index values ​​associated with each load index for each cell within a historical time period into the load index prediction model to obtain the target load index value associated with each load index for each cell within the target time period. This process may include: inputting multiple historical load index values ​​associated with each load index for each cell within a historical time period into the load index prediction model to obtain the target load index value associated with each load index for each cell within the earliest unit duration of the target time period; then, determining the earliest unit duration and the target number of unit durations preceding the earliest unit duration as the updated historical time period, and repeating the above process of obtaining the target load index value associated with each load index for each cell within the earliest unit duration of the target time period until the target load index value associated with each load index for each cell within the last unit duration of the target time period is obtained; wherein, the target number is the difference between the number of unit durations within the historical time period and 1.

[0053] For example, if the historical time period contains T units of duration, the target time period contains S units of duration, and the number of load indicators is n, then in the process of the resource scheduling equipment determining the target load indicator value associated with each load indicator for each cell in the target time period, for each cell, for the i-th load indicator among the n load indicators, the input data to the LSTM model is a tensor of (1, T, ni), and the output is a tensor of the form (1, S, ni), representing the target load indicator value of the i-th load indicator among the n load indicators in the target time period S. The above process is repeated to obtain the target load indicator value associated with each load indicator for each cell in the target time period S.

[0054] In step S203, the resource scheduling device determines the matching score of two cells in each cell group within the target time period based on the target load index value and load index threshold associated with each load index for each cell in the target time period.

[0055] In this embodiment of the disclosure, multiple cell groups are obtained by matching two cells from multiple cells, and the matching score is used to characterize the degree of load difference between the two cells. The load index threshold associated with each load index can be determined based on actual needs, and this embodiment of the disclosure does not limit this. The load index threshold associated with each load index may include: the minimum load index threshold and the maximum load index threshold of the load index threshold range associated with each load index; or, the load index threshold associated with each load index may include: a reference value for the load index threshold associated with each load index.

[0056] In one optional implementation, the process by which the resource scheduling device determines the matching score of two cells within each cell group in a target time period based on the target load index value and load index threshold associated with each load index for each cell in a target time period may include: for each cell group, obtaining a first modified target load index value associated with each load index for the first cell in the cell group in the target time period based on the load index threshold associated with each load index and the comparison result of the target load index value associated with each load index for the first cell in the cell group in the target time period; and obtaining a first modified target load index value associated with each load index for the first cell in the cell group in the target time period based on the load index threshold associated with each load index and the comparison result of the target load index value associated with each load index for the second cell in the cell group in the target time period; and obtaining a first modified target load index value associated with each load index for the first cell in the cell group in the target time period based on the comparison result of the target load index value associated with each load index for the second cell in the cell group in the target time period. The comparison results of the target load index values ​​associated with each load index are used to obtain the second modified target load index value associated with each load index in the second cell during the target time period. Next, the first modified target load index value and the second modified target load index value associated with each load index in the first cell and the second cell in each cell group are determined during the target time period, so as to obtain the load difference amount associated with each load index in the first cell and the second cell in each cell group during the target time period. Further, the load difference amounts associated with multiple load indices in the first cell and the second cell in each cell group during the target time period are weighted and summed to obtain the matching score value of the two cells in each cell group during the target time period. Based on the load index threshold associated with each load index, the target load index values ​​associated with each load index for the two cells in each cell group can be corrected within the target time period. The corrected values ​​determine the load difference between the two cells in each cell group and each load index within the target time period. The actual load status of each cell in the cell group can be measured using the load index threshold, and the target load index values ​​associated with each cell and each load index can be corrected to ensure a match between the determined load difference and the actual situation. Furthermore, the load difference between the two cells in each cell group and each load index within the target time period is weighted and summed to obtain the matching score value for the two cells in each cell group within the target time period. This allows for consideration of the impact of different load indices on the cell load status, improving the reliability of the determined matching score value for the two cells in each cell group within the target time period.

[0057] In an optional implementation, when the load indicator threshold associated with each load indicator includes: a minimum load indicator threshold and a maximum load indicator threshold within the load indicator threshold range associated with each load indicator, the process by which the resource scheduling device obtains a first corrected target load indicator value associated with each load indicator for the first cell in the cell group during the target time period, based on the load indicator threshold associated with each load indicator and a comparison result of the target load indicator values ​​associated with each load indicator for the first cell in the cell group during the target time period, may include:

[0058] For each load indicator, if the target load indicator value associated with the load indicator in the first cell of the cell group is greater than or equal to the minimum load indicator threshold associated with the load indicator during the target time period, and the target load indicator value associated with the load indicator in the first cell of the cell group is less than or equal to the maximum load indicator threshold associated with the load indicator during the target time period, then the first modified target load indicator value associated with the load indicator in the first cell is determined to be 0 during the target time period.

[0059] If the target load index value associated with the load index of the first cell in the cell group is less than the minimum load index threshold associated with the load index during the target time period, then the difference between the target load index value associated with the load index and the minimum load index threshold is determined as the first modified target load index associated with the load index of the first cell during the target time period.

[0060] If the target load index value associated with the load index of the first cell in the cell group is greater than the maximum load index threshold associated with the load index during the target time period, then the difference between the target load index value associated with the load index and the maximum load index threshold is determined as the first modified target load index associated with the load index of the first cell during the target time period.

[0061] In an optional implementation, where the load indicator threshold associated with each load indicator includes a reference value for the load indicator threshold associated with each load indicator, the process by which the resource scheduling device obtains a first corrected target load indicator value associated with each load indicator for the first cell in the cell group during the target time period, based on the load indicator threshold associated with each load indicator and a comparison result of the target load indicator values ​​associated with each load indicator for the first cell in the cell group during the target time period, may include:

[0062] For each load indicator, if the target load indicator value associated with the load indicator in the first cell of the cell group is equal to the reference value of the load indicator threshold associated with the load indicator during the target time period, then the first modified target load indicator value associated with the load indicator in the first cell is determined to be 0 during the target time period.

[0063] If the target load index value associated with the load index of the first cell in the cell group is less than the reference value of the load index threshold associated with the load index during the target time period, then the difference between the target load index value associated with the load index and the reference value of the load index threshold is determined as the first modified target load index associated with the load index of the first cell during the target time period.

[0064] If the target load index value associated with the load index of the first cell in the cell group is greater than the reference value of the load index threshold associated with the load index during the target time period, then the difference between the target load index value associated with the load index and the reference value of the load index threshold is determined as the first modified target load index associated with the load index of the first cell during the target time period.

[0065] It should be noted that, in the embodiments of this disclosure, the process by which the resource scheduling device determines the first modified target load index value associated with each load index of the second cell within the target time period is similar to the process by which the resource scheduling device determines the first modified target load index value associated with each load index of the first cell within the target time period, and will not be described in detail in the embodiments of this disclosure.

[0066] For example, for load index Fi in n types of load indices, the maximum load index threshold associated with load index Fi is Fi_high, and the minimum load index threshold associated with load index Fi is Fi_low. In a cell group, the j-th cell is denoted as cellj, the k-th cell is denoted as cellk, j is any one of multiple cells, and k is also any one of multiple cells, but j is different from k. The target load index value associated with load index Fi for the j-th cell cellj during the target time period is Fi_cellj. The process by which the resource scheduling device obtains the first corrected target load index value associated with each load index during the target time period for the first cell in the cell group, based on the load index threshold associated with each load index and the comparison result of the target load index values ​​associated with each load index for the first cell in the cell group during the target time period, can include:

[0067] If the target load index value Fi_cellj is greater than or equal to the minimum load index threshold Fi_low, and the target load index value Fi_cellj is less than or equal to the maximum load index threshold Fi_high, then the first modified target load index value Feature_Tensor_j associated with the load index Fi in cell j is determined to be 0 within the target time period.

[0068] If the target load index value Fi_cellj is less than the minimum load index threshold Fi_low, then the first modified target load index value Feature_Tensor_j associated with the load index Fi in cell j within the target time period is determined to be Fi_cellj-Fi_low; or, if the target load index value Fi_cellj is greater than the maximum load index threshold Fi_high, then the first modified target load index value Feature_Tensor_j associated with the load index Fi in cell j within the target time period is determined to be Fi_cellj-Fi_high.

[0069] It is understood that, in this embodiment of the disclosure, the load difference between cell j and cell k in the target time period associated with the load index Fi, represented by Feature_Tensor_substract_jk, is:

[0070] Feature_Tensor_substract_jk=Feature_Tensor_j-Feature_Tensor_k; (Formula 1)

[0071] In Formula 1, Feature_Tensor_k is the first modified target load index value associated with the load index Fi for the k-th cell k within the target time period.

[0072] The process by which the resource scheduling equipment weights and sums the load differences associated with multiple load indicators for the first and second cells in each cell group during the target time period to obtain the matching score of the two cells in each cell group during the target time period may include: for each cell group, constructing a target matrix based on the load differences associated with each load indicator for the first and second cells in the cell group during the target time period, wherein the columns of the target matrix are n types of load indicator dimensions, and the rows of the target matrix are S units of duration within the target time period; then, based on a preset weight configuration, multiplying the load difference of each column in the target matrix by the weight value associated with each column of load indicator to obtain an updated target matrix; further, solving the sum of the updated target matrices for each cell group to obtain the matching score of the two cells in each cell group during the target time period.

[0073] It is understood that, in the embodiments of this disclosure, the weight value configured for each load indicator can be determined based on actual needs, and the embodiments of this disclosure do not limit this. The sum of the weight values ​​configured for multiple load indicators is 1.

[0074] In step S204, the resource scheduling device can determine the cell pairing result based on the matching score set composed of the matching score values ​​of the two cells in each cell group during the target time period, and the cell pairing strategy prediction model, and perform resource scheduling during the target time period based on the cell pairing result.

[0075] In this embodiment of the disclosure, the cell pairing result includes a paired cell determined for resource scheduling for each cell.

[0076] It should be noted that, in this embodiment of the disclosure, the greater the load difference between cells, the higher the expansion resource demand of the high-load cell and the higher the remaining resources of the low-load cell. Providing the resources of the low-load cell to the high-load cell can make more efficient use of resources. Therefore, in this embodiment of the disclosure, the goal of cell pairing is to maximize the utilization of multiple cells. The strategy prediction model determines the paired cell for resource scheduling for each cell based on the matching score set between each pair of cells in the multiple cells, with the goal of maximizing resource utilization.

[0077] In one optional implementation, the process of training a strategy prediction model for resource scheduling equipment may include: acquiring a sample dataset and training a time series network using the sample dataset; obtaining a load index prediction model after the time series network converges; then, inputting multiple sample historical load index values ​​associated with each load index for each cell in the sample historical time period into the load index prediction model to obtain the sample load index prediction value associated with each load index for each cell in the sample target time period; based on the sample load index prediction value associated with each load index for each cell in the sample target time period and the load index threshold, determining the sample matching score value of two cells in each cell group in the sample target time period, and training a cell pairing strategy prediction model based on the sample matching score value of two cells in each cell group in the sample target time period. In the process of training the cell pairing strategy prediction model, the sample matching score value used is obtained by processing multiple historical load index values ​​associated with each load index for each cell in the historical sample period through the load index prediction model. This results in the predicted sample load index value associated with each load index for each cell in the target sample period, as well as the sample matching score value determined by the load index threshold. This is more in line with the strategy for determining the matching score value in actual applications, thereby improving the reliability of the trained cell pairing strategy prediction model in practical applications.

[0078] It should be noted that, in this embodiment, the process of the resource scheduling equipment training the load index prediction model can be the same as the process described above, and will not be elaborated upon here. The process by which the resource scheduling equipment determines the sample matching score of two cells within each cell group during the target time period, based on the predicted value of the sample load index associated with each load index and the load index threshold for each cell during the target time period, can refer to the process described above, which determines the matching score of two cells within each cell group during the target time period, based on the target load index value associated with each load index and the load index threshold for each cell during the target time period. This will not be elaborated upon here.

[0079] In one optional implementation, the process of training the strategy prediction model for resource scheduling equipment may include: determining the sample matching score of two cells within each cell group during the sample target time period based on the sample target load index value (i.e., load index label value) associated with each load index and the load index threshold for each cell during the sample target time period; further, training the cell pairing strategy prediction model based on the sample matching score of two cells within each cell group during the sample target time period. During the training of the cell pairing strategy prediction model, the sample matching score can be directly determined based on the sample target load index value associated with each load index for each cell during the sample target time period, which is then used to train the cell pairing strategy prediction model. This reduces the amount of data processing and improves the training efficiency of the cell pairing strategy prediction model.

[0080] In one alternative implementation, such as Figure 3 As shown, the process by which the resource scheduling device trains a cell pairing strategy prediction model based on the sample matching scores of two cells within each cell group during the target time period may include:

[0081] Step S301: Construct a sample matching score matrix using the sample matching scores of two cells within each cell group during the target time period.

[0082] In this embodiment of the disclosure, the sample matching score matrix is ​​a NumCell*NumCell dimensional matrix, where NumCell represents the number of cells. For example, when the number of cells is 100, the sample matching score matrix is ​​a 100*100 dimensional matrix.

[0083] Understandably, since the cell cannot match itself, the value of the element on the diagonal of the sample matching score matrix is ​​0.

[0084] Step S302: Multiple cell groups are identified as the action space, and a grid environment is constructed based on the sample matching score matrix;

[0085] In this embodiment of the disclosure, the grid environment includes multiple cell groups and a target matching score value corresponding to each cell group, the target matching score value being determined based on the sample matching score value.

[0086] In one optional implementation, the process of constructing a grid environment based on a sample matching score matrix by the resource scheduling device may include: constructing a target grid with a side length equal to the number of cells, wherein a sub-grid in the target grid represents a cell group; then, setting the target matching score values ​​of multiple sub-grids on the diagonal of the target grid to 0; further, obtaining the mean and variance of multiple sample matching score values ​​in the sample matching score matrix, and constructing a normal distribution of multiple sample matching score values ​​in the sample matching score matrix based on the mean and variance of multiple sample matching score values; and, for each off-diagonal target sub-grid in the target grid, extracting a sample matching score value to be processed from the normal distribution for the target sub-grid, and normalizing the sample matching score value to obtain the target matching score value; finally, filling the target sub-grid with the target matching score value of each target sub-grid to obtain the grid environment. This method can construct a new grid environment based on the sample matching score matrix during the training of a cell pairing strategy prediction model based on a reinforcement learning algorithm, preventing the use of a single grid environment, increasing the diversity of grid environments, and improving the generalization ability of the trained sample matching score matrix.

[0087] Step S303: Repeat the process of determining sample paired cells for each cell in the action space from the current state of the grid environment, obtaining the sample cell pairing result of the current iteration process, the updated state of the grid environment, and the reward value corresponding to the updated state of the grid environment, until the iteration stopping condition is met, and obtain the first target sample dataset and the second target sample dataset.

[0088] In this embodiment of the disclosure, the first target sample dataset includes the current state of the grid environment in each iteration, the sample cell pairing result, the reward value obtained after determining the sample pairing cell for each cell, the reward value being the sample matching score value of the cell group formed by the cell and the sample pairing cell, and the updated state of the grid environment. The second target sample dataset includes the updated state of the grid environment in each iteration, the updated state of the grid environment being the state of the grid environment after determining the sample pairing cell for each cell in each iteration, and the reward value corresponding to the updated state of the grid environment, the reward value being the sum of the reward values ​​obtained after determining the sample pairing cell for each cell.

[0089] The iteration stopping condition can be determined based on actual needs, and this embodiment of the disclosure does not limit it. For example, the iteration stopping condition can be that the reward value obtained in multiple consecutive iterations is greater than the reward value threshold. The reward value threshold can be determined based on actual needs, and this embodiment of the disclosure does not limit it.

[0090] In one optional implementation, the process by which the resource scheduling device determines sample paired cells for each cell in the action space from the current state of the grid environment, obtains the sample cell pairing results for the current iteration, the updated state of the grid environment, and the reward value corresponding to the updated state of the grid environment, may include: initializing the root node based on the current state of the grid environment; then, repeatedly searching for child nodes starting from the root node, and expanding all child nodes of the current child node if it is determined that the current child node has not been expanded; and selecting the next child node to search among all child nodes of the current child node based on the upper bound confidence interval algorithm, until the condition is met. The process of searching for termination conditions yields multiple optional cell pairing results. Further, the optional cell pairing result associated with the most visited search path among the search paths corresponding to each of the multiple optional cell pairing results is determined as the sample cell pairing result for the current iteration. Additionally, the updated state of the grid environment associated with the matching results of the sample optional cells is determined as the updated state of the grid environment for the current iteration. The total value of the sample matching scores for each cell and the sample paired cell in the sample cell pairing results is determined based on the sample matching scores corresponding to multiple cell groups, thus obtaining the reward value corresponding to the updated state of the grid environment. The Monte Carlo tree search algorithm can be used to train the cell pairing strategy prediction model. Collecting a target sample dataset can improve the randomness of the sample data in the collected target sample dataset, further enhancing the generalization ability of the trained cell pairing strategy prediction model.

[0091] It is understood that, in the embodiments of this disclosure, the process of the resource scheduling device searching for child nodes is the process of determining sample paired cells for each cell in the action space. The search termination condition can be that the action space is empty. Determining sample paired cells for a cell can be understood as performing an action. The reward for performing an action is the sample matching score of the cell group consisting of the cell and the paired cell. The reward value of the current iteration is the total sample matching score of the cell group consisting of the sample paired cells determined for each cell after the current iteration is completed.

[0092] In the action space, each cell can form a cell group with multiple other cells. During the process of determining sample paired cells for each cell in the action space, the resource scheduling device searches for one of the candidate cell groups containing the target cell in the action space as the sample target cell group for each target cell without a determined paired cell. Then, it determines another cell in the sample target cell group other than the target cell as the sample paired cell of the target cell. The other candidate cell groups containing the target cell are deleted from the action space to prevent multiple sample paired cells from being determined for the same cell.

[0093] The algorithm for the upper confidence bound (UCB) is as follows:

[0094]

[0095] In Formula 2, ni is the number of times node Si is explored, Vi is the total reward obtained by exploring node Si, N is the total number of explorations, and c is a constant, which can generally be taken as 2.

[0096] Optionally, after repeatedly acquiring the target sample dataset (including the first target sample dataset and the second target sample dataset) a specific number of iterations, the resource scheduling device can update the grid environment based on the grid environment construction method provided in the above embodiments, and continue to acquire the target sample dataset using the updated grid environment. This can further improve the generalization ability of the trained cell pairing strategy prediction model. The specific number of iterations can be set based on actual needs; for example, the specific number of iterations can be 25 or 30.

[0097] Step S304: The cell pairing strategy prediction model is jointly iteratively trained based on the first target sample dataset and the value network based on the second target sample dataset. After the cell pairing strategy prediction model and the value network converge, the cell pairing strategy prediction model is obtained.

[0098] In this embodiment of the disclosure, a cell pairing strategy prediction model can be trained using a reinforcement learning algorithm. During the training process, the reward value is used as the model optimization objective. Since the reward value is the sum of the reward values ​​obtained after determining the sample paired cells for each cell, and the reward value is the sample matching score value of the cell group formed by the cell and the sample paired cells, the matching score value can characterize the load difference between cells. Thus, the optimization objective of maximizing resource utilization is achieved, ensuring that the trained cell pairing strategy can match each cell in multiple cells with a cell pairing result that maximizes the resource utilization of multiple cells.

[0099] It should be noted that, in the embodiments of this disclosure, the cell pairing strategy prediction model and the value network model can be neural network models, and the specific model structure can be set according to actual needs. This disclosure does not limit this.

[0100] In one optional implementation, the process of the resource scheduling device jointly iteratively training a cell pairing policy prediction model based on a first target sample dataset and a value network based on a second target sample dataset may include: determining a first sample from the first target sample dataset, which includes the current state of the grid environment, sample cell pairing results, reward values ​​obtained after determining sample paired cells for each cell, and the updated state of the grid environment in each iteration; and determining a second sample from the second target sample dataset, which includes the updated state of the grid environment and the reward value corresponding to the updated state of the grid environment in each iteration; then, repeatedly inputting the current state of the grid environment from the first sample into the cell pairing policy prediction model. The model is tested to obtain the predicted cell matching result. The updated state of the grid environment in the second sample is input into the value model to obtain the predicted reward value corresponding to the updated state of the grid environment. Then, based on the predicted cell matching result, the predicted reward value corresponding to the updated state of the grid environment, the sample cell pairing result in the first sample, the reward value obtained after determining the sample paired cell for each cell, the updated state of the grid environment, the reward value corresponding to the updated state of the grid environment in the second sample, and the process of determining the loss function value, if the loss function value is less than the loss function threshold, the cell pairing strategy prediction model and the value network are determined to converge. The loss function threshold can be determined based on actual needs, and this embodiment does not limit it.

[0101] The loss function is:

[0102]

[0103] In Equation 3, k is the number of times the cell pairing strategy prediction model and value network are repeatedly trained, P_mcts_i is the probability distribution of the sample cell pairing results in the first sample, P_nn_i is the probability distribution of the predicted cell matching results, value_nn_i is the predicted reward value corresponding to the updated state of the grid environment, and value_mcts_i is the reward value corresponding to the updated state of the grid environment. The probability distribution is determined based on the reward value obtained after determining the sample pairing cell for each cell in the first sample, and the updated state of the grid environment.

[0104] It is understandable that the first and second samples used in each repeated training of the cell pairing strategy prediction model and value network are sample data obtained in the same iteration of step S303 above.

[0105] In one optional implementation, the resource sculpting device can further determine the sample reference cell pairing result for the current iteration in a replica grid of the grid environment based on a greedy strategy. Then, based on the sample matching score values ​​corresponding to multiple cell groups, it determines the total sample reference matching score corresponding to each cell and the sample reference paired cell in the sample reference cell pairing result, obtaining the reference reward value for the current iteration. Further, it determines the difference between the reward value and the reference reward value, obtaining the reward value difference, and uses the reward value difference and the reward value of the current iteration as the updated reward value for the current iteration. Based on the sample reference cell pairing result determined by the greedy strategy, the cell pairing strategy prediction model can be optimized and trained so that, in practical applications, the trained cell pairing strategy prediction model can determine cell pairing results that optimize resource utilization.

[0106] In one optional implementation, the process by which the resource scheduling device determines the cell pairing result based on a set of matching scores of two cells within each cell group during a target time period, and a cell pairing strategy prediction model, may include: constructing a matching score matrix using the set of matching scores; then constructing the current grid environment based on the matching score matrix; and further, inputting the current state of the current grid environment into the cell pairing strategy prediction model to obtain the cell pairing result.

[0107] It should be noted that the process of constructing the matching score matrix by the resource scheduling device is similar to the process of constructing the sample matching score matrix, and the process of constructing the current grid environment based on the matching score matrix is ​​similar to the process of constructing the current grid environment based on the matching score matrix. This disclosure will not elaborate on these aspects.

[0108] In an optional implementation, before performing resource scheduling based on cell pairing results during a target time period, the resource scheduling device may further: find the target matching score value associated with each cell and paired cell in the cell pairing results from the matching score value set, and determine the sum of multiple target matching score values ​​to obtain a cell pairing result score value associated with the cell pairing results; then, based on a greedy strategy and the matching score value set, determine a reference cell pairing result, wherein the reference cell pairing result includes a reference paired cell determined for each cell; and, in the matching score value set, find the reference matching score value associated with each cell and reference paired cell in the reference cell pairing result, and determine the sum of multiple reference matching score values ​​to obtain a reference cell pairing result score value associated with the reference cell pairing result; furthermore, if the reference cell pairing result score value is greater than the cell pairing result score value, then the reference cell pairing result is determined as the updated cell pairing result. A greedy strategy can be used to determine the reference cell pairing result. After determining the reference cell pairing result, which is better than the cell pairing result determined by the cell pairing strategy prediction model, the reference cell pairing result is determined as the final cell pairing result, so as to facilitate the cell matching result that maximizes resource utilization.

[0109] In one optional implementation, the process by which the resource scheduling device performs resource scheduling based on cell pairing results during a target time period may include: during the target time period, for each cell, according to the paired cell determined for the cell in the cell pairing results, if the cell load is greater than that of the paired cell, providing the resources of the paired cell to the cell.

[0110] An exemplary embodiment of this disclosure provides a resource scheduling apparatus, which may be a server or a chip applied to a server. Figure 4 A schematic block diagram of the functional modules of a resource scheduling apparatus according to an exemplary embodiment of the present disclosure is shown. Figure 4 As shown, the resource scheduling device 400 includes:

[0111] The acquisition module 401 is configured to acquire multiple historical load index values ​​of multiple cells within a historical time period, which are respectively associated with multiple load indexes.

[0112] The first determining module 402 is configured to input multiple historical load index values ​​associated with each load index in a historical period into the load index prediction model to obtain the target load index value associated with each load index in a target period for each cell.

[0113] The second determining module 403 is configured to determine the matching score value of two cells in each of the multiple cell groups during the target time period based on the target load index value and load index threshold associated with each load index of each cell during the target time period. The multiple cell groups are obtained by matching two cells in each of the multiple cells, and the matching score value is used to characterize the degree of load difference between the two cells.

[0114] The resource scheduling module 404 is configured to determine the cell pairing result based on the matching score set composed of the matching score values ​​of two cells in each cell group during the target time period, and the cell pairing strategy prediction model, and to perform resource scheduling during the target time period based on the cell pairing result, wherein the cell pairing result includes the paired cells determined for resource scheduling for each cell.

[0115] Optionally, the second determining module 403 is configured to:

[0116] For each cell group, based on the load index threshold associated with each load index and the comparison result of the target load index value associated with each load index in the first cell of the cell group during the target time period, a first modified target load index value associated with each load index in the first cell of the cell group during the target time period is obtained.

[0117] Based on the load index threshold associated with each load index, and the comparison result of the target load index value associated with each load index in the second cell of the cell group during the target time period, a second modified target load index value associated with each load index in the second cell of the cell group is obtained.

[0118] Determine the first and second cells in each cell group, and the first and second modified target load index values ​​associated with each load index during the target time period, to obtain the load difference amount associated with each load index for the first and second cells in each cell group during the target time period;

[0119] The load difference between the first cell and the second cell in each cell group and multiple load indicators during the target time period is weighted and summed to obtain the matching score value of the two cells in each cell group during the target time period.

[0120] Optionally, the device further includes an update module 405, configured to:

[0121] In the set of matching score values, find the target matching score value associated with each cell and the paired cell in the cell matching result, and determine the sum of multiple target matching score values ​​to obtain the cell matching result score value associated with the cell matching result;

[0122] Based on the greedy strategy and the matching score set, a reference cell pairing result is determined, wherein the reference cell pairing result includes a reference paired cell determined for each cell;

[0123] In the set of matching score values, find the reference matching score value associated with each cell in the reference cell pairing result and the reference paired cell, and determine the sum of multiple reference matching score values ​​to obtain the reference cell pairing result score value associated with the reference cell pairing result;

[0124] If the reference cell pairing result score is greater than the cell pairing result score, then the reference cell pairing result is determined as the updated cell pairing result.

[0125] Optionally, the apparatus further includes a model training module 406, configured to:

[0126] Obtain a sample dataset, wherein the sample dataset includes multiple historical load index values ​​of each cell associated with multiple load indices during the sample historical period, and load index label values ​​of each cell associated with multiple load indices during the sample target period.

[0127] The time series network is trained using the sample dataset, and the load index prediction model is obtained after the time series network converges.

[0128] The multiple historical load index values ​​associated with each load index for each cell during the sample historical period are input into the load index prediction model to obtain the predicted value of the sample load index associated with each load index for each cell during the sample target period.

[0129] Based on the predicted value of the sample load index and the load index threshold associated with each load index for each cell in the target sample period, the sample matching score of the two cells in each cell group in the target sample period is determined, and a cell pairing strategy prediction model is trained based on the sample matching score of the two cells in each cell group in the target sample period.

[0130] Optionally, the model training module 406 is configured as follows:

[0131] A sample matching score matrix is ​​constructed using the sample matching score values ​​of two cells within each cell group during the target time period.

[0132] Multiple cell groups are identified as the action space, and a grid environment is constructed based on the sample matching score matrix. The grid environment includes the multiple cell groups and a target matching score corresponding to each cell group. The target matching score is determined based on the sample matching score.

[0133] The process of repeatedly determining sample-paired cells for each cell in the action space from the current state of the grid environment, obtaining the sample cell pairing result of the current iteration, the updated state of the grid environment, and the reward value corresponding to the updated state of the grid environment, continues until the iteration stopping condition is met, resulting in a first target sample dataset and a second target sample dataset. The first target sample dataset includes the current state of the grid environment, the sample cell pairing result, the reward value obtained after determining the sample-paired cells for each cell, and the updated state of the grid environment in each iteration. The second target sample dataset includes the updated state of the grid environment in each iteration and the reward value corresponding to the updated state of the grid environment.

[0134] The cell pairing strategy prediction model is jointly trained based on the first target sample dataset and the value network based on the second target sample dataset. After the cell pairing strategy prediction model and the value network converge, the cell pairing strategy prediction model is obtained.

[0135] Optionally, the model training module 406 is configured as follows:

[0136] Root node initialization is performed based on the current state of the mesh environment;

[0137] The process of repeatedly searching for child nodes starting from the root node, expanding all child nodes of the current child node if it is determined that the current child node has not been expanded, and selecting the next child node to search among all child nodes of the current child node based on the confidence interval upper bound algorithm, until the search termination condition is met, yields multiple optional cell pairing results.

[0138] The optional cell pairing result associated with the search path with the most visits among the search paths corresponding to the multiple optional cell pairing results is determined as the sample cell pairing result of the current iteration process;

[0139] The updated state of the grid environment associated with the matching results of the sample selectable cells is determined as the updated state of the grid environment in the current iteration. The total value of the sample matching score values ​​corresponding to each cell and the sample paired cell in the sample cell pairing results is determined based on the sample matching score values ​​corresponding to multiple cell groups. The reward value corresponding to the updated state of the grid environment is obtained.

[0140] Optionally, the model training module 406 is further configured to:

[0141] Based on a greedy strategy, the sample reference cell pairing result for the current iteration is determined in the replica grid of the grid environment;

[0142] Based on the sample matching score values ​​corresponding to multiple cell groups, determine the total sample reference matching score value corresponding to each cell and the sample reference paired cell in the sample reference cell pairing results, and obtain the reference reward value for the current iteration;

[0143] The difference between the reward value and the reference reward value is determined to obtain the reward value difference. The reward value difference and the reward value of the current iteration are then determined as the updated reward value of the current iteration.

[0144] Exemplary embodiments of this disclosure also provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the electronic device to perform a method according to an embodiment of this disclosure.

[0145] Exemplary embodiments of this disclosure also provide a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to embodiments of this disclosure.

[0146] like Figure 5 As shown, an exemplary embodiment of this disclosure also provides a computer program product 500, including a computer program 501, wherein the computer program 501, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this disclosure.

[0147] refer to Figure 6The present invention describes a structural block diagram of an electronic device 600 that can serve as a server or client of the present disclosure, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0148] like Figure 6 As shown, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 may also store various programs and data required for the operation of the electronic device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0149] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, output unit 607, storage unit 608, and communication unit 609. Input unit 606 can be any type of device capable of inputting information to electronic device 600. Input unit 606 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 607 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 608 may include, but is not limited to, disks and optical discs. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth™ devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0150] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above. For example, in some embodiments, the methods of the exemplary embodiments of this disclosure can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 600 via ROM 602 and / or communication unit 609. In some embodiments, the computing unit 601 can be configured to perform the methods of the exemplary embodiments of this disclosure by any other suitable means (e.g., by means of firmware).

[0151] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0152] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, 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 devices, magnetic storage devices, or any suitable combination of the foregoing.

[0153] As used in this disclosure, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0154] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0155] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0156] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0157] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this disclosure are performed, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal, a user equipment, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center integrating one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video disc (DVD); or it can be a semiconductor medium, such as a solid-state drive (SSD).

[0158] Although this disclosure has been described in conjunction with specific features and embodiments, it will be apparent that various modifications and combinations can be made therein without departing from the spirit and scope of this disclosure. Accordingly, this specification and drawings are merely exemplary illustrations of the disclosure as defined by the appended claims and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this disclosure. It is obvious that those skilled in the art can make various alterations and modifications to this disclosure without departing from its spirit and scope. Thus, this disclosure is also intended to include any such modifications and modifications that fall within the scope of the claims of this disclosure and their equivalents.

Claims

1. A resource scheduling method, characterized in that, include: Obtain the historical load index values ​​of multiple cells within a historical time period, which are associated with multiple load indices respectively; The historical load index values ​​associated with each load index in each cell during the historical period are input into the load index prediction model to obtain the target load index value associated with each load index in each cell during the target period. Based on the target load index value and load index threshold associated with each load index in each cell during the target time period, the matching score value of two cells in each cell group in the multiple cell groups is determined during the target time period. The multiple cell groups are obtained by matching two cells in each of the multiple cells. The matching score value is used to characterize the degree of load difference between the two cells. Based on the matching score set composed of the matching score values ​​of two cells in each cell group during the target time period, and the cell pairing strategy prediction model, the cell pairing result is determined, and resource scheduling is performed during the target time period based on the cell pairing result. The cell pairing result includes the paired cells determined for resource scheduling for each cell. Specifically, based on the target load index value and load index threshold associated with each load index for each cell within the target time period, the matching score value of two cells within each cell group in the multiple cell groups during the target time period is determined, including: For each cell group, based on the load index threshold associated with each load index and the comparison result of the target load index value associated with each load index in the first cell of the cell group during the target time period, a first modified target load index value associated with each load index in the first cell of the cell group during the target time period is obtained. Based on the load index threshold associated with each load index, and the comparison result of the target load index value associated with each load index in the second cell of the cell group during the target time period, a second modified target load index value associated with each load index in the second cell of the cell group is obtained. Determine the first and second cells in each cell group, and the first and second modified target load index values ​​associated with each load index during the target time period, to obtain the load difference amount associated with each load index for the first and second cells in each cell group during the target time period; The load difference between the first cell and the second cell in each cell group and multiple load indicators during the target time period is weighted and summed to obtain the matching score value of the two cells in each cell group during the target time period.

2. The resource scheduling method as described in claim 1, characterized in that, Before performing resource scheduling during the target time period based on the cell pairing results, the method further includes: In the set of matching score values, find the target matching score value associated with each cell and the paired cell in the cell matching result, and determine the sum of multiple target matching score values ​​to obtain the cell matching result score value associated with the cell matching result; Based on the greedy strategy and the matching score set, a reference cell pairing result is determined, wherein the reference cell pairing result includes a reference paired cell determined for each cell; In the set of matching score values, find the reference matching score value associated with each cell in the reference cell pairing result and the reference paired cell, and determine the sum of multiple reference matching score values ​​to obtain the reference cell pairing result score value associated with the reference cell pairing result; If the reference cell pairing result score is greater than the cell pairing result score, then the reference cell pairing result is determined as the updated cell pairing result.

3. The resource scheduling method as described in claim 1, characterized in that, The method further includes: Obtain a sample dataset, wherein the sample dataset includes multiple historical load index values ​​of each cell associated with multiple load indices during the sample historical period, and load index label values ​​of each cell associated with multiple load indices during the sample target period. The time series network is trained using the sample dataset, and the load index prediction model is obtained after the time series network converges. The multiple historical load index values ​​associated with each load index for each cell during the sample historical period are input into the load index prediction model to obtain the predicted value of the sample load index associated with each load index for each cell during the sample target period. Based on the predicted value of the sample load index and the load index threshold associated with each load index for each cell in the target sample period, the sample matching score of the two cells in each cell group in the target sample period is determined, and a cell pairing strategy prediction model is trained based on the sample matching score of the two cells in each cell group in the target sample period.

4. The resource scheduling method as described in claim 3, characterized in that, The step of training a cell pairing strategy prediction model based on the sample matching scores of two cells within each cell group during the target time period includes: A sample matching score matrix is ​​constructed using the sample matching score values ​​of two cells within each cell group during the target time period. Multiple cell groups are identified as the action space, and a grid environment is constructed based on the sample matching score matrix. The grid environment includes the multiple cell groups and a target matching score corresponding to each cell group. The target matching score is determined based on the sample matching score. The process of repeatedly determining sample-paired cells for each cell in the action space from the current state of the grid environment, obtaining the sample cell pairing result of the current iteration, the updated state of the grid environment, and the reward value corresponding to the updated state of the grid environment, continues until the iteration stopping condition is met, resulting in a first target sample dataset and a second target sample dataset. The first target sample dataset includes the current state of the grid environment, the sample cell pairing result, the reward value obtained after determining the sample-paired cells for each cell, and the updated state of the grid environment in each iteration. The second target sample dataset includes the updated state of the grid environment in each iteration and the reward value corresponding to the updated state of the grid environment. The cell pairing strategy prediction model is jointly trained based on the first target sample dataset and the value network based on the second target sample dataset. After the cell pairing strategy prediction model and the value network converge, the cell pairing strategy prediction model is obtained.

5. The resource scheduling method as described in claim 4, characterized in that, The process of determining sample cell pairs for each cell in the action space from the current state of the grid environment to obtain the sample cell pairing result of the current iteration, the updated state of the grid environment, and the reward value corresponding to the updated state of the grid environment includes: Root node initialization is performed based on the current state of the mesh environment; The process of repeatedly searching for child nodes starting from the root node, expanding all child nodes of the current child node if it is determined that the current child node has not been expanded, and selecting the next child node to search among all child nodes of the current child node based on the confidence interval upper bound algorithm, until the search termination condition is met, yields multiple optional cell pairing results. The optional cell pairing result associated with the search path with the most visits among the search paths corresponding to the multiple optional cell pairing results is determined as the sample cell pairing result of the current iteration process; The updated state of the grid environment associated with the matching results of the sample selectable cells is determined as the updated state of the grid environment in the current iteration. The total value of the sample matching score values ​​corresponding to each cell and the sample paired cell in the sample cell pairing results is determined based on the sample matching score values ​​corresponding to multiple cell groups. The reward value corresponding to the updated state of the grid environment is obtained.

6. The resource scheduling method as described in claim 5, characterized in that, The method further includes: Based on a greedy strategy, the sample reference cell pairing result for the current iteration is determined in the replica grid of the grid environment; Based on the sample matching score values ​​corresponding to multiple cell groups, determine the total sample reference matching score value corresponding to each cell and the sample reference paired cell in the sample reference cell pairing results, and obtain the reference reward value for the current iteration; The difference between the reward value and the reference reward value is determined to obtain the reward value difference. The reward value difference and the reward value of the current iteration are then determined as the updated reward value of the current iteration.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the resource scheduling method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the resource scheduling method according to any one of claims 1 to 6.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the resource scheduling method according to any one of claims 1 to 6.

Images