Feature data processing method, apparatus, device, medium, and product

Abstract

The present disclosure provides a feature data processing method, device, equipment, medium and product, and relates to the technical field of quantum computing. The feature data processing method comprises: determining a feature sequence corresponding to collected feature data, the feature sequence comprising at least one missing value, the missing value being configured to be assigned a filling value; constructing a target function of the feature sequence based on a preset constraint condition of the filling value; converting the target function into a target Hamiltonian function, and inputting the target Hamiltonian function into a quantum annealing device supporting a quantum annealing algorithm for optimization calculation; and assigning the missing value according to an output result of the quantum annealing device. Through the embodiment of the present disclosure, the reliability and accuracy of filling missing data when pre-processing feature data can be improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, and in particular, to a feature data processing method, device, equipment, medium and product. BACKGROUND

[0002] At present, various time series may be collected in the data processing scene based on the Internet of Things technology, including 4 / 5G wireless communication data, financial data, sensor data, medical monitoring data, etc. Before data analysis and data training, the time series are usually preprocessed, which often involves the processes of supplementing data missing values and correcting abnormal data. The data filling technology of the related technology has defects such as overfitting, local extreme value limitation and low reliability in the processing scene of non-linear mode and complex correlation data.

[0003] It is explained that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0004] The purpose of the present disclosure is to provide a feature data processing method, device, equipment, medium and product, which at least partly overcomes the problems caused by the limitations and defects of the related technology.

[0005] According to a first aspect of an embodiment of the present disclosure, a feature data processing method is provided, comprising: determining a feature sequence corresponding to collected feature data, the feature sequence comprising at least one missing value, the missing value being configured to be assigned by a filling value; constructing a target function of the feature sequence based on a preset constraint condition of the filling value; converting the target function into a target Hamiltonian function, and inputting the target Hamiltonian function into a quantum annealing device supporting a quantum annealing algorithm for optimization calculation; and assigning the missing value according to an output result of the quantum annealing device.

[0006] In an exemplary embodiment of the present disclosure, the preset constraint condition includes at least one of global smoothness, local consistency and statistical constraint, and the construction of the target function of the feature sequence based on the preset constraint condition of the filling value includes:

[0007] determining a global difference function corresponding to the feature sequence after the filling value is assigned according to the global smoothness;

[0008] determining a local consistency function corresponding to the missing value according to the local consistency;

[0009] determining a statistical constraint function corresponding to the feature sequence after the filling value is assigned according to the statistical constraint;

[0010] The global difference function, the local consistency function and the statistical constraint function are calculated according to preset weights, and the preset weights are configured to be dynamically adjusted to make the target function converge.

[0011] The target function of the feature sequence is constructed according to the result of the weighted calculation.

[0012] In an exemplary embodiment of the present disclosure, the feature sequence is a time sequence, and the global difference function corresponding to the feature sequence after the imputed values are determined according to the global smoothness includes:

[0013] determining a difference value between the imputed value and a neighboring value of the imputed value;

[0014] The difference value is accumulated according to a set of time points in the time sequence to determine the global difference function.

[0015] In an exemplary embodiment of the present disclosure, the local consistency function corresponding to the missing value is determined according to the local consistency includes:

[0016] determining a set of positions corresponding to the missing value;

[0017] determining a set of neighborhood indexes of the missing value, and the set of neighborhood indexes is configured to be determined by a sampling frequency and / or a spatial density of the feature sequence;

[0018] traversing and calculating a difference value between any numerical value in the set of positions and any numerical value in the set of neighborhood indexes, and summing all the difference values to determine the local consistency function.

[0019] In an exemplary embodiment of the present disclosure, the statistical constraint function corresponding to the feature sequence after the imputed values are determined according to the statistical constraint includes:

[0020] determining a distribution fitting error of the feature sequence after the imputed values are determined;

[0021] determining the distribution fitting error as the statistical constraint function, wherein the distribution fitting error includes at least one of a KL divergence, a mean and a standard deviation of the feature sequence, and the KL divergence is determined according to the feature sequence without the missing value and the feature sequence after the imputed values are determined.

[0022] In an exemplary embodiment of the present disclosure, the target function is converted into a target Hamiltonian function, and the target Hamiltonian function is input into a quantum annealing device supporting a quantum annealing algorithm for optimization calculation.

[0023] transforming the objective function into an objective Hamiltonian function;

[0024] transforming the objective Hamiltonian function into a QUBO matrix;

[0025] inputting the QUBO matrix into a quantum annealing device supporting a quantum annealing algorithm for optimization calculation.

[0026] In an example embodiment of the present disclosure, transforming the objective Hamiltonian function into a QUBO matrix comprises:

[0027] transforming a value range of a variable corresponding to the padding value in the objective Hamiltonian function;

[0028] discretizing the value range into a finite candidate set, and any value in the candidate set being represented by a binary variable;

[0029] replacing a continuous variable in the objective Hamiltonian function with a discrete variable;

[0030] constructing the QUBO matrix according to the binary variable and the discrete variable.

[0031] In an example embodiment of the present disclosure, assigning the missing value according to the output result of the quantum annealing device comprises:

[0032] the output result of the quantum annealing device performing optimization calculation on the QUBO matrix;

[0033] decoding the binary variable in the output result into the padding value, and assigning the missing value according to the padding value.

[0034] In an example embodiment of the present disclosure, assigning the missing value according to the output result of the quantum annealing device further comprises:

[0035] performing noise reduction processing on the output result of the quantum annealing device performing optimization calculation on the QUBO matrix based on repeated sampling and / or filtering algorithms.

[0036] In an example embodiment of the present disclosure, further comprising:

[0037] in response to completing the assignment of the missing value according to the padding value, performing data verification on the assigned feature sequence, the data verification comprising at least one of smoothness verification, local consistency verification, distribution matching verification, and business rule verification;

[0038] in response to the feature sequence satisfying the constraint condition of the data verification, determining that the assignment of the feature sequence is completed.

[0039] In an example embodiment of the present disclosure, further comprising:

[0040] In response to the feature sequence not satisfying the constraint condition of the data verification, adjusting a preset weight of the objective function based on a preset optimization algorithm and / or adjusting a configuration parameter of the quantum annealing algorithm, the preset optimization algorithm including at least one of a gradient descent algorithm, a Bayesian optimization algorithm, and an adaptive adjustment optimization algorithm, and the configuration parameter including at least one of annealing time, annealing path, and chain strength.

[0041] According to a second aspect of the embodiments of the present disclosure, a feature data processing method of a communication network is provided, comprising:

[0042] Collecting cell data indicators of the communication network;

[0043] Converting the cell data indicators into feature data;

[0044] Processing the feature data by using the steps of the feature data processing method according to any one of the above technical solutions.

[0045] According to a third aspect of the embodiments of the present disclosure, a feature data processing apparatus is provided, comprising:

[0046] A determination module configured to determine a feature sequence corresponding to the collected feature data, the feature sequence including at least one missing value, and the missing value being configured to be assigned a fill value;

[0047] A construction module configured to construct an objective function of the feature sequence based on a preset constraint condition of the fill value;

[0048] A conversion module configured to convert the objective function into a target Hamiltonian function, and input the target Hamiltonian function into a quantum annealing device supporting a quantum annealing algorithm for optimization calculation;

[0049] A calculation module configured to assign the missing value according to an output result of the quantum annealing device.

[0050] According to a fourth aspect of the embodiments of the present disclosure, a feature data processing apparatus of a communication network is provided, comprising:

[0051] A data collection module configured to collect cell data indicators of the communication network;

[0052] A data conversion module configured to convert the cell data indicators into feature data;

[0053] A data processing module configured to process the feature data by using the steps of the feature data processing method according to any one of the above technical solutions.

[0054] According to a fifth aspect of the present disclosure, an electronic device is provided, comprising a memory; and a processor coupled to the memory, the processor is configured to execute a method according to any one of the preceding aspects based on instructions stored in the memory.

[0055] According to a sixth aspect of the present disclosure, a computer readable storage medium is provided, having stored thereon a program which, when executed by a processor, implements a feature data processing method according to any one of the preceding aspects.

[0056] According to a seventh aspect of the present disclosure, a computer program product is provided, comprising a computer program which, when executed by a processor, implements a feature data processing method according to any one of the preceding aspects.

[0057] In the embodiments of the present disclosure, by determining a feature sequence corresponding to the collected feature data, the feature sequence includes at least one missing value, the missing value is configured to be assigned by a filling value, and a target function of the feature sequence is constructed based on a preset constraint condition of the filling value. Then, the target function is converted into a target Hamiltonian function, and the target Hamiltonian function is input into a quantum annealing device supporting a quantum annealing algorithm for optimization calculation. Finally, the missing value is assigned according to the output result of the quantum annealing device. The quantum annealing algorithm performs optimization calculation on the target function by quantum tunneling effect to avoid the local extremum problem commonly existing in classical optimization algorithms, and can find the optimal solution of the missing value in a wider search space, overcome the local extremum problem, and effectively balance the global trend and local features in the missing value filling scheme. Compared with the classical algorithm, the calculation efficiency has obvious advantages, and the global trend and local features of the feature data can be more accurately restored.

[0058] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure. It is apparent that the accompanying drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor based on these drawings.

[0060] Figure 1 A schematic diagram of an exemplary system architecture of a feature data processing scheme to which embodiments of the present application can be applied is shown;

[0061] Figure 2is a flowchart of a feature data processing method in an example embodiment of the present disclosure;

[0062] Figure 3 is a flowchart of another feature data processing method in an example embodiment of the present disclosure;

[0063] Figure 4 is a flowchart of another feature data processing method in an example embodiment of the present disclosure;

[0064] Figure 5 is a block diagram of a feature data processing apparatus in an example embodiment of the present disclosure;

[0065] Figure 6 is a block diagram of another feature data processing apparatus in an example embodiment of the present disclosure;

[0066] Figure 7 is a block diagram of an electronic device in an example embodiment of the present disclosure. DETAILED DESCRIPTION

[0067] Example implementations are now described with reference to the following drawings. The example implementations, however, can be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the example implementations to those skilled in the art. The features, structures, or characteristics described can be combined in one or more implementations. In the following description, numerous specific details are provided, such as examples of programming, software modules, user selections, etc., to provide a thorough understanding of the example implementations. One skilled in the relevant art will recognize, however, that the implementations can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. Some features, structures, or characteristics can be described as being implemented over one or more entities, portions of which can be implemented as hardware, software, or firmware. In such examples, a given portion can be implemented to include one or more components, which can be hardware, software, or firmware components. Additionally, a given portion can be implemented to include one or more processors, which can be hardware processors or hardware processor components.

[0068] Furthermore, the accompanying drawings are only schematic and are non-limiting examples, in which like references denote like features in various drawings, and in which only those parts essential to an understanding of the present disclosure have been shown or described, and in which:

[0069] In the related art, the missing value filling value assignment scheme includes linear interpolation, mean filling, KNN filling, etc., and usually relies on local information and is easily affected by initial values or parameter settings, resulting in that the filling value assignment result may not be optimal.

[0070] In view of the technical problems existing in the data processing scene in the related art, the present application proposes a new feature data processing method, device, equipment, medium and product, which will be described below in combination with Figures 1 to 7 The embodiments of the present disclosure are described in detail.

[0071] Figure 1 A schematic diagram of an exemplary system architecture to which the feature data processing scheme of the embodiments of the present application can be applied is shown.

[0072] As Figure 1 shown, the system architecture 100 can include one or more of terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 serves as a medium to provide a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 can include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0073] It should be understood that Figure 1 the number of terminal devices, networks and servers in

[0074] The user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, etc. The terminal devices 101, 102, 103 can be various electronic devices with display screens, including but not limited to smartphones, tablet computers, portable computers and desktop computers, etc.

[0075] In some embodiments, the feature data processing method provided by the embodiments of the present application is generally executed by the server 105, and accordingly, the feature data processing device is generally provided in the terminal device 103 (which can also be the terminal device 101 or 102). In other embodiments, some terminals can have similar functions as the server device to execute the present method.

[0076] The following will be described in detail with reference to the accompanying Figures 2 to 7 The example embodiments of the present disclosure are described in detail.

[0077] Figure 2 is a flowchart of the feature data processing method in the exemplary embodiments of the present disclosure.

[0078] Referring to Figure 2 , the feature data processing method can include:

[0079] Step S202, determining a feature sequence corresponding to the collected feature data, the feature sequence including at least one missing value, the missing value being configured to be assigned by a filling value;

[0080] Step S204: Construct the objective function of the feature sequence based on the preset constraints of the above-mentioned fill values;

[0081] Step S206: The objective function is transformed into an objective Hamiltonian function, and the objective Hamiltonian function is input into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation;

[0082] Step S208: Assign values ​​to missing values ​​based on the output of the quantum annealing device.

[0083] In the above embodiments of this disclosure, by determining the feature sequence corresponding to the collected feature data, the feature sequence includes at least one missing value, the missing value is configured to be filled with a value, and a target function of the feature sequence is constructed based on the preset constraints of the filling value. The target function is then transformed into a target Hamiltonian function, and the target Hamiltonian function is input to a quantum annealing device that supports the quantum annealing algorithm for optimization calculation. Finally, the missing value is filled with a value based on the output of the quantum annealing device. The quantum annealing algorithm optimizes the target function by crossing the energy barrier through the quantum tunneling effect, thereby avoiding the local extremum problem common in classical optimization algorithms. It can find the optimal solution for the missing value in a wider search space, overcome the local extremum problem, and effectively balance the global trend and local features in the missing value filling scheme. Compared with classical algorithms, it has a significant advantage in computational efficiency and can more accurately recover the global trend and local features of the feature data.

[0084] The following section provides a detailed explanation of each step in the feature data processing method.

[0085] In an exemplary embodiment of this disclosure, the aforementioned preset constraints include at least one of global smoothness, local consistency, and statistical constraints. The objective function for constructing the feature sequence based on the preset constraints of the filled values ​​includes:

[0086] Based on the global smoothness described above, determine the global difference function corresponding to the feature sequence after assigning the fill value;

[0087] Based on the aforementioned local consistency, determine the local consistency function corresponding to the missing values;

[0088] Based on the above statistical constraints, determine the statistical constraint function corresponding to the above feature sequence after assigning the above fill values;

[0089] The global difference function, the local consistency function, and the statistical constraint function are weighted according to preset weights. The preset weights are configured to be dynamically adjusted to make the objective function converge.

[0090] The objective function for the above feature sequence is constructed based on the weighted calculation results.

[0091] In the embodiments described above, the constructed objective function (Loss) includes three optimization objectives: global smoothness, local consistency, and statistical constraints. This ensures that the feature data after being filled with values ​​not only satisfies the continuity of the global trend but also maintains local consistency with feature data adjacent to the missing values, and preserves the statistical distribution characteristics of the data as much as possible. Furthermore, the objective function can balance the influence of each objective by adjusting the weighting coefficients.

[0092] In related technologies, imputation methods typically consider only a single objective. For example, mean imputation only considers the mean of the data, KNN imputation focuses on local data similarity, while interpolation methods primarily consider data continuity. Compared to related technologies, the embodiments of this disclosure comprehensively consider multiple factors and design an objective function with multi-level optimization objectives. This ensures data smoothness while also taking into account the consistency between the imputed feature data and the feature data adjacent to the missing values, as well as the statistical distribution characteristics of the data. After imputation based on the embodiments of this disclosure, the quality and robustness of the overall feature data are significantly superior to existing technologies.

[0093] In the above embodiments of this disclosure, the optimization objective corresponding to global smoothness can be denoted as: With global smoothness as one of the optimization objectives, the filled time series can be made to be continuous in the overall trend, and the overall smoothness of the series can be measured by the difference operation.

[0094] In the above embodiments of this disclosure, the optimization objective corresponding to local consistency can be denoted as: One of the optimization objectives is to ensure that the filled value is consistent with the surrounding neighborhood data.

[0095] In the above embodiments of this disclosure, the optimization objective corresponding to the statistical constraints can be denoted as: With statistical constraints as one of the optimization objectives, the goal is to make the filled feature sequence as close as possible to the distribution of data values. This can be defined by the distribution fitting error. The objective function is... ,in These are weighting coefficients used to balance the influence of various optimization objectives.

[0096] In an exemplary embodiment of this disclosure, the aforementioned feature sequence is a time series, and determining the global difference function corresponding to the aforementioned feature sequence after the padding value assignment based on the aforementioned global smoothness includes:

[0097] Determine the difference between the above-mentioned fill value and its nearest neighbor;

[0098] The difference values ​​are accumulated based on the set of time points in the time series to determine the global difference function.

[0099] In the above embodiments of this disclosure, the optimization objective S corresponding to global smoothness is the aforementioned global difference function, whose expression can be written as: , where T is the set of time points in the feature sequence.

[0100] in, and One of them is the fill value. and The other one is the value adjacent to the missing value, or and For two adjacent missing values. Additionally, if and If all values ​​are missing, an iterative filling strategy can be used to fill isolated missing values ​​first, and then process consecutive missing segments.

[0101] In one exemplary embodiment of this disclosure, determining the local consistency function corresponding to the missing value based on the aforementioned local consistency includes:

[0102] Determine the set of locations corresponding to the above missing values;

[0103] A neighborhood index set for the aforementioned missing values ​​is determined, wherein the neighborhood index set is configured to be determined by the sampling frequency and / or spatial density of the aforementioned feature sequences;

[0104] The difference between any value in the above location set and any value in the above neighborhood index set is calculated by iterating through the data and summing all the difference values ​​to determine the above local consistency function.

[0105] In the above embodiments of this disclosure, the optimization objective D of local consistency is the aforementioned local consistency function, the expression of which can be written as: ,in, The filler value is the value corresponding to the i-th missing value in the feature sequence. Let M be the j-th value in the feature sequence, M be the set of missing value locations, and N(i) be the... The set of neighborhood indexes.

[0106] In the above embodiments of this disclosure, if the neighborhood index set N(i) contains only missing values, the local constraint is ignored, or pre-filled using interpolation. Alternatively, if the neighborhood index set N(i) contains several or more missing values, the missing values ​​are filled using an iterative optimization method, such as filling isolated missing values ​​in a finite manner, and then gradually processing consecutive missing segments.

[0107] In one exemplary embodiment of this disclosure, the statistical constraint function corresponding to the feature sequence after the fill value assignment is determined according to the above statistical constraints includes:

[0108] Determine the distribution fitting error of the above feature sequence after assigning the above filler values;

[0109] The above distribution fitting error is defined as the above statistical constraint function, wherein the above distribution fitting error includes at least one of the above feature sequence's KL divergence, mean, and standard deviation, and the above KL divergence is determined based on the feature sequence that does not contain the above missing values ​​and the feature sequence after being filled with the above values.

[0110] In the above embodiments of this disclosure, the statistical constraint function corresponding to the optimization objective P of the statistical constraint can be the Kullback-Leibler divergence, also known as relative entropy, and its expression is denoted as... Data and Filled represent the original data after removing missing values ​​and the data after filling or processing, respectively, i.e., by minimizing... The goal is to make the filled distribution approximate the data value distribution in the feature sequence as closely as possible.

[0111] In the above embodiments of this disclosure, the statistical constraint function corresponding to the optimization objective P of the statistical constraint can be the error of the mean and variance, and its expression is denoted as... ,in, and These are the mean and standard deviation of the feature sequence after the filler values ​​are assigned, respectively. and These are the mean and standard deviation of the feature sequence before the fill values ​​are assigned, respectively.

[0112] In one exemplary embodiment of this disclosure, transforming the objective function into a target Hamiltonian function and inputting the target Hamiltonian function into a quantum annealing device supporting the quantum annealing algorithm for optimization calculation includes:

[0113] Transform the above objective function into an objective Hamiltonian function;

[0114] Transform the above objective Hamiltonian function into a QUBO matrix;

[0115] The QUBO matrix described above is input into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation.

[0116] In the above embodiments of this disclosure, during the process of transforming the objective function into a QUBO matrix, firstly, the objective function Loss is transformed into the objective Hamiltonian H of the quantum annealing algorithm, and then H is expressed in QUBO matrix form, the expression of which is as follows: ,in, The diagonal elements of the QUBO matrix represent individual variables. Independent energy contribution, The off-diagonal elements in the QUBO matrix represent variables. Coupling effect with other variables, Indicates whether a candidate imputation value is selected for the i-th missing value. , The k-th candidate padding value is represented by k, where k is the number of candidate values ​​for each discrete padding value.

[0117] In one exemplary embodiment of this disclosure, assigning values ​​to missing values ​​based on the output of the quantum annealing device further includes:

[0118] The output result of the above-mentioned QUBO matrix optimization calculation is denoised based on repeated sampling and / or filtering algorithms.

[0119] In the above embodiments of this disclosure, by performing noise reduction processing on the output result of the above-mentioned QUBO matrix optimization calculation based on repeated sampling and / or filtering algorithms, the noise in the output result can be effectively reduced, thereby improving the reliability and accuracy of the objective function optimization result.

[0120] In one exemplary embodiment of this disclosure, transforming the aforementioned target Hamiltonian function into a QUBO matrix includes:

[0121] Define the range of values ​​for the variables corresponding to the aforementioned filler values ​​in the target Hamiltonian function.

[0122] The above range of values ​​is discretized into a finite candidate set, and any value in the candidate set is represented by a binary variable.

[0123] Replace the continuous variables in the above objective Hamiltonian function with discrete variables;

[0124] Construct the QUBO matrix based on the binary variables and discrete variables described above.

[0125] In the above embodiments of this disclosure, the processing steps for constructing the QUBO matrix based on the binary variables and discrete variables include:

[0126] (1) Variable discretization: filling in values Since the variables are continuous, the QUBO matrix requires the variables to be binary. Therefore, each... The range of values ​​is discretized into a finite set of candidates, and each value is represented by a binary variable, the expression of which is: ,in This is a binary variable representing whether the i-th fill value is selected as a candidate value. And satisfy the constraints .

[0127] (2) Objective function discretization: Replace the continuous variables involved in the objective function Loss with discrete variables. Taking global smoothness as an example, Similarly, local consistency and statistical constraints are also transformed into discrete forms;

[0128] (3) Constructing the QUBO matrix: Construct the QUBO matrix Q based on the discretized objective function, with diagonal elements Representing the individual effect of each variable, off-diagonal elements. This represents the interaction between variables, such as smoothness constraints between adjacent time points.

[0129] In one exemplary embodiment of this disclosure, assigning values ​​to missing values ​​based on the output of the quantum annealing device includes:

[0130] Obtain the output results of the above-mentioned quantum annealing device for optimizing the above-mentioned QUBO matrix;

[0131] The binary variables in the above output are decoded into the above padding values, and the above missing values ​​are assigned values ​​based on the above padding values.

[0132] In the embodiments described above, the constructed QUBO matrix Q is input into a quantum annealing device, and a global optimal solution is searched using a quantum annealing algorithm. The dimension of the QUBO matrix is ​​determined by the discretization range of the filler values ​​and the number of actual positions. For example, if each filler value has... K If there are *m* candidate values ​​and *m* missing values, then the dimension of the QUBO matrix is... .

[0133] In the above embodiments of this disclosure, the binary result is mapped to the specific value of the missing value according to the output of the quantum annealing algorithm, and the output of the quantum annealing algorithm is a binary variable configuration. Decoding yields the actual fill value; the decoding process can be based on... The value of is used to determine the candidate value corresponding to the i-th missing value. If multiple If both are 1, then the configuration with the highest probability or the optimal objective function value is selected, and the result is decoded into a set of filler values. .

[0134] In one exemplary embodiment of this disclosure, it further includes:

[0135] In response to the completion of assigning values ​​to the missing values ​​based on the above-mentioned fill values, data verification is performed on the assigned feature sequence. The data verification includes at least one of smoothness verification, local consistency verification, distribution matching verification, and business rule verification.

[0136] In response to the fact that the above feature sequence satisfies the constraints of the above data verification, the assignment of the above feature sequence is completed.

[0137] In the above embodiments of this disclosure, verifying the rationality of the feature sequence filling result after completing the filling value assignment includes smoothness test (checking whether the overall trend of the feature sequence after the filling value assignment is smooth, visualizing the data trend, and ensuring that there are no abnormal jumps or noise), local consistency check (calculating the error between the filling value and its neighborhood data value to ensure that the filling value is reasonable) and distribution matching (comparing the distribution of the filled feature sequence with the original data distribution, mean and variance, etc., to evaluate the consistency of the distribution), but is not limited to these.

[0138] In the above embodiments of this disclosure, business rule verification can be based on an empirical value range determined by historically stored business data, that is, the filling value assignment should meet the constraints of the empirical value range.

[0139] In the embodiments described above, when the filling result meets the verification conditions, the output filled value is assigned to the feature sequence. Smoothness testing, local consistency verification, distribution matching verification, and business rule verification are then performed to ensure that the feature data after the filling value assignment matches the original data in terms of overall trend, local consistency, and statistical characteristics. This effectively avoids unreasonable or overfitting filling value assignment results. Furthermore, iterative optimization capabilities are added. By adjusting the weights of the objective function and the parameters of the quantum annealing device, it is further ensured that the filling result meets the verification requirements.

[0140] In one exemplary embodiment of this disclosure, it further includes:

[0141] In response to the fact that the above feature sequence does not meet the constraints of the above data verification, the preset weights of the above objective function are adjusted based on the preset optimization algorithm and / or the configuration parameters of the above quantum annealing algorithm are adjusted. The preset optimization algorithm includes at least one of gradient descent algorithm, Bayesian optimization algorithm and adaptive adjustment optimization algorithm, and the configuration parameters include at least one of annealing time, annealing path and chain strength.

[0142] In the above embodiments of this disclosure, if it is determined that the feature sequence after padding does not meet the constraints of the above data verification, then adjusting the parameters of the objective function includes: the weights of the objective function Loss. These are the weights for controlling smoothness, local consistency, and distribution matching, respectively; the parameter adjustments for the quantum annealing algorithm include annealing time, annealing path, and chain strength.

[0143] Figure 3 This is a flowchart of a feature data processing method for a communication network in an exemplary embodiment of this disclosure.

[0144] refer to Figure 3 The characteristic data processing method of a communication network may include:

[0145] Step S302: Collect cell data indicators of the communication network;

[0146] Step S304: Convert the above-mentioned cell data indicators into feature data;

[0147] Step S306: Process the feature data using the feature data processing method described in any of the above technical solutions.

[0148] exist Figure 3 Based on what is shown, as Figure 4 As shown, the feature data processing method in step S306 specifically includes the following steps:

[0149] Step S402: Obtain the time series and model the missing value imputation problem as finding the optimal set of imputation values.

[0150] Step S404: Construct an objective function to evaluate the quality of the data filling scheme.

[0151] Step S406: Transform the objective function into a QUBO function and construct the QUBO matrix.

[0152] Step S408: Optimize the QUBO matrix using quantum annealing.

[0153] Step S410: Decode the output of the quantum annealing device into the actual fill value.

[0154] Step S412: Verify the rationality of the filling result. If the verification passes, proceed to step S414; otherwise, proceed to step S404.

[0155] Step S414: Fill the output padding values ​​into the time series data.

[0156] With the construction and operation of 5G networks, a large amount of cell data generated in the network is used to evaluate and optimize wireless network performance. Cell data indicators include uplink PRB occupancy rate, PDCCH channel occupancy rate, RRC connection count, traffic, transmit power, etc., but are not limited to these.

[0157] exist Figure 3and Figure 4 Based on the illustrated embodiments, the aforementioned cell data metrics are crucial for monitoring network health, evaluating performance, and troubleshooting. However, due to equipment failures, network fluctuations, transmission problems, and other reasons, missing values ​​are inevitable in the data. To ensure subsequent data analysis and optimization, these missing values ​​must be appropriately filled in. Existing filling methods, such as linear interpolation, mean imputation, and KNN methods, often struggle to capture the complex temporal sequence and global trends of 5G network data.

[0158] Therefore, the embodiments of this disclosure take the preprocessing and missing value imputation of existing 5G cell data as the object, and use the missing value imputation assignment method based on the quantum annealing QUBO model. Through global optimization, it ensures that the feature data after the imputation assignment performs well in terms of global smoothness, local consistency and statistical matching.

[0159] Step 1, Data Modeling: Obtain the operator's current network indicator data for wireless cells in City A in January 2024. Historical data is recorded hourly.

[0160] The average uplink PRB occupancy rate of cell B at different time points is shown in Table 1 (the data has been blurred).

[0161] As shown in Table 1, this time series dataset contains a total of 10,000 rows, denoted as... , This represents the average uplink PRB occupancy rate at the i-th time point.

[0162] In Table 1, rows 22, 1008, 6238, 6239, and 9184 contain missing data, denoted as NaN. To fill these missing values, we model this as an optimization problem: finding an optimal set of filler values. This ensures that the filled data conforms to the objective function.

[0163] Table 1

[0164]

[0165] Step 2, Define the objective function: Construct the objective function Loss, which includes:

[0166] (1) Global smoothness The filled data should maintain the continuity of the overall trend as much as possible, that is, the difference between adjacent data points should be small. The calculation method is as follows: , where T is the set of data time points in Table 1.

[0167] in, and One of them is the fill value. and The other one is the value adjacent to the missing value, or and For two adjacent missing values. Additionally, if and If all values ​​are missing, an iterative filling strategy can be used to fill isolated missing values ​​first, and then process consecutive missing segments.

[0168] (2) Local consistency The filled data should be as consistent as possible with the neighboring data. For example, after filling the missing values ​​in row 22, it should be close to the data in rows 21 and 23. The calculation method is as follows: ,in, The filler value is the value corresponding to the i-th missing value in the feature sequence. Let M be the j-th value in the feature sequence, M be the set of missing value locations, and N(i) be the... The set of neighborhood indexes;

[0169] (3) Statistical constraints The filled data should be as consistent as possible with the statistical characteristics of the other data in Table 1, such as the mean and variance. The mean of the average PRB occupancy rate data in Table 1 is 0.80, and the standard deviation is 0.05. The filled data should maintain these statistical characteristics. The calculation method is as follows: ,in, and These are the mean and standard deviation of the feature sequence after the filler values ​​are assigned, respectively. and They are 0.80 and 0.05 respectively;

[0170] The constructed objective function is ,in These are the weighting coefficients.

[0171] Step 3: Transform the objective function into a QUBO matrix: First, fill in the values... Discretize into a finite set of candidate values, for example, imputation values ​​for each missing value. Five candidate values ​​can be selected (0.78, 0.80, 0.82, 0.84, 0.86). Next, to discretize the continuous variable into binary form, each of the other mesenchymal molecules is represented as a binary variable. ,in This indicates that the i-th missing value selected the k-th candidate value. Then, a binary variable is constructed for each missing value, and the continuous part of the objective function is transformed into a discrete variable. Based on the discretized objective function, a QUBO matrix is ​​constructed.

[0172] Step 4: Optimize the QUBO matrix using quantum annealing: Input the QUBO matrix into the quantum annealing device and optimize it using the quantum annealing algorithm. Table 1 contains 5 missing values, and each missing value is filled by 5 candidate values. Therefore, the dimension of the QUBO matrix is ​​25×25. Quantum annealing finds the optimal filling value configuration through global optimization. Since the data in Table 1 has good continuity, the result returned by the quantum annealing device is the best solution, and the binary output shown below can be obtained.

[0173] .

[0174] Step 5: Decode the actual fill value: Based on the output of the quantum annealing device, map the binary result to the specific value of the missing value. The result in Table 1 above is decoded as z. 22 =0.78 (b) 22 ,1=1), z 1008 =0.80 (b) 1008 ,2=1),z 6238 =0.82 (b) 6238 ,3=1), z 6239 =0.84 (b) 6239 ,4=1), z 9184 =0.86 (b) 9184 ,5=1).

[0175] Step 6: Validate the infill results: Use real time series data to validate the reasonableness of the infill results, specifically including:

[0176] (1) Smoothness test: The filled data is smooth with no obvious fluctuations;

[0177] (2) Local consistency check: Calculate the error between the padding value and its neighboring data. Taking row 22 as an example, the padding value of row 22 is 0.78, and the neighboring data is row 21 with a padding value of 0.85 and row 23 with a padding value of 0.84. The error is calculated as follows: The error is very small, indicating that the filled value is consistent with its neighborhood data;

[0178] (3) Distribution matching verification: Comparing the mean and variance of the filled values ​​with the original data, before the missing values ​​were filled, the mean of the average occupancy rate of the upstream PRB in Table 1 was 0.80 and the standard deviation was 0.05. After the missing values ​​were filled, the mean of the average occupancy rate of the upstream PRB in Table 1 was 0.80 and the standard deviation was 0.048, which showed a good match.

[0179] If the filled value is determined to meet the verification conditions, the output filled value will be assigned to Table 1 for subsequent data processing and data training.

[0180] Corresponding to the above method embodiments, this disclosure also provides a feature data processing apparatus, which can be used to execute the above method embodiments.

[0181] Figure 5 This is a block diagram of a feature data processing apparatus according to an exemplary embodiment of the present disclosure.

[0182] refer to Figure 5 The feature data processing device 500 may include:

[0183] The determination module 502 is configured to determine the feature sequence corresponding to the collected feature data, wherein the feature sequence includes at least one missing value, and the missing value is configured to be assigned a filler value.

[0184] Construction module 504 is configured to construct the objective function of the feature sequence based on the preset constraints of the above-mentioned fill values;

[0185] The conversion module 506 is configured to convert the above objective function into an objective Hamiltonian function, and input the above objective Hamiltonian function into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation;

[0186] The calculation module 508 is configured to assign values ​​to missing values ​​based on the output of the quantum annealing device.

[0187] In one exemplary embodiment of this disclosure, the construction module 504 is further configured to:

[0188] Based on the global smoothness described above, determine the global difference function corresponding to the feature sequence after assigning the fill value;

[0189] Based on the aforementioned local consistency, determine the local consistency function corresponding to the missing values;

[0190] Based on the above statistical constraints, determine the statistical constraint function corresponding to the above feature sequence after assigning the above fill values;

[0191] The global difference function, the local consistency function, and the statistical constraint function are weighted according to preset weights. The preset weights are configured to be dynamically adjusted to make the objective function converge.

[0192] The objective function for the above feature sequence is constructed based on the weighted calculation results.

[0193] In one exemplary embodiment of this disclosure, the construction module 504 is further configured to:

[0194] Determine the difference between the above-mentioned fill value and its nearest neighbor;

[0195] The difference values ​​are accumulated based on the set of time points in the time series to determine the global difference function.

[0196] In one exemplary embodiment of this disclosure, the construction module 504 is further configured to:

[0197] Determine the set of locations corresponding to the above missing values;

[0198] A neighborhood index set for the aforementioned missing values ​​is determined, wherein the neighborhood index set is configured to be determined by the sampling frequency and / or spatial density of the aforementioned feature sequences;

[0199] The difference between any value in the above location set and any value in the above neighborhood index set is calculated by iterating through the data and summing all the difference values ​​to determine the above local consistency function.

[0200] In one exemplary embodiment of this disclosure, the construction module 504 is further configured to:

[0201] Determine the distribution fitting error of the above feature sequence after assigning the above filler values;

[0202] The above distribution fitting error is defined as the above statistical constraint function, wherein the above distribution fitting error includes at least one of the above feature sequence's KL divergence, mean, and standard deviation, and the above KL divergence is determined based on the feature sequence that does not contain the above missing values ​​and the feature sequence after being filled with the above values.

[0203] In one exemplary embodiment of this disclosure, the conversion module 506 is further configured to:

[0204] Transform the above objective function into an objective Hamiltonian function;

[0205] Transform the above objective Hamiltonian function into a QUBO matrix;

[0206] The QUBO matrix described above is input into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation.

[0207] In one exemplary embodiment of this disclosure, the conversion module 506 is further configured to:

[0208] Define the range of values ​​for the variables corresponding to the aforementioned filler values ​​in the target Hamiltonian function.

[0209] The above range of values ​​is discretized into a finite candidate set, and any value in the candidate set is represented by a binary variable.

[0210] Replace the continuous variables in the above objective Hamiltonian function with discrete variables;

[0211] Construct the QUBO matrix based on the binary variables and discrete variables described above.

[0212] In one exemplary embodiment of this disclosure, the computing module 508 is further configured to:

[0213] Obtain the output results of the above-mentioned quantum annealing device for optimizing the above-mentioned QUBO matrix;

[0214] The binary variables in the above output are decoded into the above padding values, and the above missing values ​​are assigned values ​​based on the above padding values.

[0215] In one exemplary embodiment of this disclosure, the computing module 508 is further configured to:

[0216] The output result of the above-mentioned QUBO matrix optimization calculation is denoised based on repeated sampling and / or filtering algorithms.

[0217] In one exemplary embodiment of this disclosure, the feature data processing apparatus 500 is further configured to:

[0218] In response to the completion of assigning values ​​to the missing values ​​based on the above-mentioned fill values, data verification is performed on the assigned feature sequence. The data verification includes at least one of smoothness verification, local consistency verification, distribution matching verification, and business rule verification.

[0219] In response to the fact that the above feature sequence satisfies the constraints of the above data verification, the assignment of the above feature sequence is completed.

[0220] In one exemplary embodiment of this disclosure, the feature data processing apparatus 500 is further configured to:

[0221] In response to the fact that the feature sequence after the above-mentioned filling value assignment does not meet the above-mentioned data verification constraints, the preset weights of the above-mentioned objective function are adjusted and / or the configuration parameters of the above-mentioned quantum annealing algorithm are adjusted. The preset optimization algorithm includes at least one of gradient descent algorithm, Bayesian optimization algorithm and adaptive adjustment optimization algorithm, and the configuration parameters include at least one of annealing time, annealing path and chain strength.

[0222] refer to Figure 6 The feature data processing device 600 of the communication network may include:

[0223] Data acquisition module 602 is configured to collect cell data indicators of the communication network;

[0224] The data transformation module 604 is configured to convert the above-mentioned cell data indicators into feature data.

[0225] The data processing module 606 is configured to process the feature data using the steps of the feature data processing method described in any of the above technical solutions.

[0226] Since the functions of the feature data processing device 500 and the feature data processing device 600 of the communication network have been described in detail in their respective method embodiments, they will not be repeated here.

[0227] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0228] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0229] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented in the following forms: entirely in hardware, entirely in software (including firmware, microcode, etc.), or in a combination of hardware and software, collectively referred to herein as “circuit,” “module,” or “system.”

[0230] The following reference Figure 7 To describe an electronic device 700 according to this embodiment of the present invention. Figure 7 The electronic device 700 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0231] like Figure 7 As shown, the electronic device 700 is manifested in the form of a general-purpose computing device. The components of the electronic device 700 may include, but are not limited to: at least one processing unit 710, at least one storage unit 720, and a bus 730 connecting different system components (including storage unit 720 and processing unit 710).

[0232] The storage unit stores program code that can be executed by the processing unit 710, causing the processing unit 710 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 710 can perform the method shown in the embodiments of this disclosure.

[0233] Storage unit 720 may include readable media in the form of volatile storage units, such as random access memory (RAM) 7201 and / or cache memory 7202, and may further include read-only memory (ROM) 7203.

[0234] The storage unit 720 may also include a program / utility 7204 having a set (at least one) program module 7205, such program module 7205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0235] Bus 730 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0236] Electronic device 700 can also communicate with one or more external devices 740 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 700, and / or with any device that enables electronic device 700 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 750. Furthermore, electronic device 700 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 760. As shown, network adapter 760 communicates with other modules of electronic device 700 via bus 730. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0237] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0238] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible embodiments, various aspects of the invention may also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of the invention described in the "Exemplary Methods" section of this specification.

[0239] The program product for implementing the above-described method according to embodiments of the present invention may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0240] The readable medium can be a readable signal medium or a readable storage medium. A readable storage medium can 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 of readable storage media (a non-exhaustive list) 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 devices, magnetic storage devices, or any suitable combination thereof.

[0241] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0242] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0243] In exemplary embodiments of this disclosure, a computer program product is also provided. This computer program product can be loaded or stored on any combination of one or more readable media. The program code for performing operations of the present invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0244] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0245] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and concept of this disclosure are indicated by the claims.

Claims

1. A feature data processing method, characterized in that, include: Collect cell data metrics from the communication network; The cell data indicators are converted into feature data; Determine the feature sequence corresponding to the feature data, wherein the feature sequence includes at least one missing value, and the missing value is configured to be assigned a filler value; Construct the objective function of the feature sequence based on the preset constraints of the filled values; The objective function is transformed into an objective Hamiltonian function, and the objective Hamiltonian function is input into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation, including: Transform the objective function into an objective Hamiltonian function; Transforming the target Hamiltonian function into a QUBO matrix includes: The range of values ​​for the variable corresponding to the fill value in the target Hamiltonian function; The range of values ​​is discretized into a finite candidate set, and any value in the candidate set is represented by a binary variable. Replace the continuous variables in the target Hamiltonian function with discrete variables; Construct the QUBO matrix based on the binary variables and the discrete variables; The QUBO matrix is ​​input into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation; The missing values ​​are assigned based on the output of the quantum annealing device, including: Obtain the output result of the quantum annealing device performing optimization calculations on the QUBO matrix; The binary variables in the output are decoded into the padding values, and the missing values ​​are assigned values ​​based on the padding values.

2. The feature data processing method as described in claim 1, characterized in that, The preset constraints include global smoothness, local consistency, and statistical constraints. The objective function for constructing the feature sequence based on the preset constraints of the filled values ​​includes: Based on the global smoothness, determine the global difference function corresponding to the feature sequence after the padding value is assigned; Determine the local consistency function corresponding to the missing value based on the local consistency; The statistical constraint function corresponding to the feature sequence after the filling value is assigned is determined based on the statistical constraints. The global difference function, the local consistency function, and the statistical constraint function are weighted according to preset weights, and the preset weights are configured to be dynamically adjusted to make the objective function converge. The objective function of the feature sequence is constructed based on the weighted calculation results.

3. The feature data processing method as described in claim 2, characterized in that, The feature sequence is a time series, and the global difference function corresponding to the feature sequence after the padding value assignment is determined based on the global smoothness, including: Determine the difference between the fill value and its nearest neighbors; The difference values ​​are accumulated based on the set of time points in the time series to determine the global difference function.

4. The feature data processing method as described in claim 2, characterized in that, The local consistency function corresponding to the missing value is determined based on the local consistency, including: Determine the set of locations corresponding to the missing values; A neighborhood index set for the missing value is determined, the neighborhood index set being configured to be determined by the sampling frequency and / or spatial density of the feature sequence; The difference between any value in the location set and any value in the neighborhood index set is calculated by iterating through the data, and all difference values ​​are summed to determine the local consistency function.

5. The feature data processing method as described in claim 2, characterized in that, The statistical constraint function corresponding to the feature sequence after the filling value assignment is determined based on the statistical constraints, including: Determine the distribution fitting error of the feature sequence after assigning the filler values; The distribution fitting error is determined as the statistical constraint function, wherein the distribution fitting error includes at least one of the KL divergence, mean, and standard deviation of the feature sequence, and the KL divergence is determined based on the feature sequence without the missing values ​​and the feature sequence after the filling values ​​are assigned.

6. The feature data processing method as described in claim 1, characterized in that, Assigning values ​​to missing values ​​based on the output of the quantum annealing device also includes: The output result of the optimized calculation of the QUBO matrix is ​​subjected to noise reduction based on repeated sampling and / or filtering algorithms.

7. The feature data processing method as described in claim 1, characterized in that, Also includes: In response to the completion of assigning values ​​to the missing values ​​according to the filling values, data verification is performed on the assigned feature sequence, including smoothness verification, local consistency verification, distribution matching verification and business rule verification; In response to the feature sequence satisfying the constraints of the data verification, the assignment of the feature sequence is determined to be complete.

8. The feature data processing method as described in claim 7, characterized in that, Also includes: In response to the feature sequence not satisfying the constraints of the data verification, the preset weights of the objective function are adjusted and / or the configuration parameters of the quantum annealing algorithm are adjusted based on a preset optimization algorithm. The preset optimization algorithm includes at least one of gradient descent algorithm, Bayesian optimization algorithm and adaptive adjustment optimization algorithm, and the configuration parameters include at least one of annealing time, annealing path and chain strength.

9. A feature data processing apparatus, characterized in that, include: The module is defined as a collection module for cell data metrics from the communication network. The cell data indicators are converted into feature data; Determine the feature sequence corresponding to the collected feature data, wherein the feature sequence includes at least one missing value, and the missing value is configured to be assigned a filler value; The construction module is configured to construct the objective function of the feature sequence based on preset constraints of the filled values; The transformation module is configured to transform the objective function into a target Hamiltonian function, and input the target Hamiltonian function into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation, including: Transform the objective function into an objective Hamiltonian function; Transforming the target Hamiltonian function into a QUBO matrix includes: The range of values ​​for the variable corresponding to the fill value in the target Hamiltonian function; The range of values ​​is discretized into a finite candidate set, and any value in the candidate set is represented by a binary variable. Replace the continuous variables in the target Hamiltonian function with discrete variables; Construct the QUBO matrix based on the binary variables and the discrete variables; The QUBO matrix is ​​input into a quantum annealing device that supports the quantum annealing algorithm for optimization calculation; The calculation module is configured to assign values ​​to missing values ​​based on the output of the quantum annealing device, including: Obtain the output result of the quantum annealing device performing optimization calculations on the QUBO matrix; The binary variables in the output are decoded into the padding values, and the missing values ​​are assigned values ​​based on the padding values.

10. An electronic device, characterized in that, include: Memory; as well as A processor coupled to the memory, the processor being configured to execute the feature data processing method as described in any one of claims 1-8 based on instructions stored in the memory.

11. A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the feature data processing method as described in any one of claims 1-8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the feature data processing method as described in any one of claims 1-8.

Images