A method for selecting an algorithm for a network digital model and a corresponding device

Abstract

The present invention discloses an algorithm selection method for a network digital model and a corresponding device, comprising: obtaining performance measurement indicators and accuracy requirements, and obtaining at least one candidate algorithm according to the performance measurement indicators and accuracy requirements; obtaining the algorithm capability of each candidate algorithm, sorting all the candidate algorithms according to the algorithm capability, and prioritizing the candidate algorithm with the largest algorithm capability as the candidate algorithm; obtaining at least one associated parameter on which the candidate algorithm depends; if the accuracy of the associated parameters can all meet the accuracy requirements, then the candidate algorithm is used as the target algorithm of the digital model; if the accuracy of at least one of the associated parameters cannot meet the accuracy requirements, then reselecting the candidate algorithm with the largest algorithm capability from the remaining candidate algorithms as the candidate algorithm.

Description

Technical Field

[0001] The present invention belongs to the field of digital models, and more specifically, relates to an algorithm selection method for a network digital model and a corresponding device. Background Art

[0002] Currently, leveraging digital technology to support network design, deployment, and operations is a key technical approach. Evolving towards network digitalization is an inevitable trend accompanying the development of full-factor digitalization. Network digitalization involves digitally modeling each network component, incorporating data from production processes and real-time changes in physical facilities to accurately depict network changes and status.

[0003] During the process of network digitization, significant progress has been made in modeling the behavior and state of various network components. On the one hand, with the help of simulation technology, white-box algorithms based on physical mechanism descriptions have emerged, achieving promising results in simulation and experimental environments. On the other hand, with the development of computing and big data technologies, artificial intelligence (AI) algorithms (black-box algorithms) based on machine learning, deep learning, and reinforcement learning have been widely proposed and verified, providing new approaches for constructing digital models. However, overall, white-box algorithms based on mechanism descriptions and black-box AI algorithms based on big data remain insufficient when applied in production environments, especially in communication network scenarios that require high real-time performance, high-precision control, and high reliability.

[0004] Furthermore, in current modeling and application practices, digital models are tightly coupled with algorithms and cannot be autonomously selected, updated, or replaced in specific application environments, resulting in weak adaptability and insufficient generalization of digital models. There is a lack of measurement methods for algorithm capabilities, especially for the real-time application capabilities of algorithms, making it impossible to evaluate and select better algorithms. There is no algorithm update mechanism, and adaptive updates are impossible when abnormalities or failures occur in sensors / probes on which the algorithm relies. Summary of the Invention

[0005] In response to the above-mentioned defects or improvement needs of the prior art, the present invention provides an algorithm selection method and corresponding device for a network digital model, which aims to decouple the algorithm from the digital model, select the algorithm based on the algorithm capability and the accuracy that the algorithm can achieve, and improve the generalization ability and semi-autonomous behavior ability of the digital model, thereby solving the technical problem that the digital model is tightly coupled with the algorithm and cannot be autonomously selected, updated or replaced in a specific application environment, resulting in weak adaptability and insufficient generalization ability of the digital model.

[0006] To achieve the above object, according to one aspect of the present invention, a method for selecting an algorithm for a network digital model is provided, comprising:

[0007] Obtaining a performance metric and an accuracy requirement, and obtaining at least one candidate algorithm based on the performance metric and the accuracy requirement;

[0008] Obtaining the algorithm capability of each candidate algorithm, sorting all candidate algorithms according to the algorithm capability, and prioritizing the candidate algorithm with the greatest algorithm capability as the candidate algorithm;

[0009] Obtaining at least one associated parameter on which the alternative algorithm depends;

[0010] If the accuracy of the associated parameters can meet the accuracy requirement, the alternative algorithm is used as the target algorithm of the digital model;

[0011] If the accuracy of at least one of the associated parameters cannot meet the accuracy requirement, a candidate algorithm with the greatest algorithm capability is reselected from the remaining candidate algorithms as an alternative algorithm.

[0012] Furthermore, the associated parameters include probe parameters and / or index parameters, wherein the probe parameters are parameters supported by the probe component of the digital model, and the index parameters are output parameters from other models; the algorithm selection method includes:

[0013] If the associated parameter on which the alternative algorithm depends is a probe parameter, then obtaining the detection accuracy of the probe component and determining whether the accuracy of the probe component can meet the accuracy requirement;

[0014] If the accuracy requirement can be met, determining whether the associated parameters on which the alternative algorithm depends are indicative parameters;

[0015] If there is an indicative parameter, a request message carrying the accuracy requirement is sent to the association model corresponding to the indicative parameter, and whether the indicative parameter meets the accuracy requirement is determined according to a response message of the association model.

[0016] Furthermore, the obtaining of the algorithm capability of each of the to-be-selected algorithms includes:

[0017] Dynamically obtain the algorithm characterization degree M, perception S, average computing resource consumption rate R and algorithm efficiency E;

[0018] Update the algorithm capability in real time according to the following formula;

[0019] Ability = log(M × S × R ÷ E)

[0020] Among them, Ability refers to algorithm capability.

[0021] Furthermore, the algorithm efficiency E is obtained as follows: algorithm efficiency E = time complexity × space complexity;

[0022] The algorithm characterization degree is obtained as follows: algorithm characterization degree M = average index precision ÷ (error degree × nested level dependency);

[0023] The perception S is obtained as follows:

[0024]

[0025] Where n is the number of probes;

[0026] The method for obtaining the average computing resource consumption rate R is as follows:

[0027]

[0028] Where n represents the number of statistics. Each time the algorithm is run, the time consumption rate and space utilization rate are counted. When the algorithm is not used, the average computing resource consumption rate R is set to 1.

[0029] Furthermore, the algorithm selection method further includes:

[0030] Dynamically updating the algorithm index accuracy of the algorithm corresponding to the digital model according to the sensing accuracy of the probe parameters, the accuracy of the index parameters and / or the fitting parameters of the algorithm;

[0031] Dynamically update algorithm capabilities based on probe accuracy, algorithm execution, time complexity, space complexity, and / or nesting level dependencies.

[0032] Furthermore, each algorithm is equipped with an algorithm description, and the format of the algorithm description is: [parameter list, algorithm identifier, performance measurement indicators, performance requirements, [dynamic description]], where the format of [dynamic description] is [algorithm indicator accuracy, [parameter identifier 1, parameter accuracy, whether it can be finely characterized], [parameter identifier 2, parameter accuracy, whether it can be finely characterized], ..., [parameter identifier n, parameter accuracy, whether it can be finely characterized], fitting parameter 1, fitting parameter 2, ..., fitting parameter m].

[0033] Furthermore, the obtaining of performance metrics and accuracy requirements, and obtaining at least one candidate algorithm according to the performance metrics and accuracy requirements includes:

[0034] The algorithm description of the algorithm to be selected is parsed, and an algorithm having the performance measurement index and algorithm index accuracy that meets the algorithm requirements is selected as the algorithm to be selected.

[0035] Furthermore, the algorithm selection method further includes:

[0036] When updating a new algorithm, add algorithm information according to the algorithm description;

[0037] According to the parameter list in the algorithm information, the parameter set in the current digital model is updated and merged;

[0038] After the parameter set of the digital model is updated, determine whether the probe component of the digital model supports data collection of the newly added parameter type. If not, set the algorithm capability of the algorithm to 0; if supported, obtain the algorithm capability of the algorithm based on the algorithm characterization degree M, perception S, average computing resource consumption rate R, and algorithm efficiency E.

[0039] Furthermore, the algorithm selection method further includes:

[0040] After selecting an alternative algorithm, obtaining resource usage of the alternative algorithm;

[0041] Determining whether resource usage of the alternative algorithm will cause the model efficiency of the digital model to be lower than a set efficiency threshold;

[0042] If so, the algorithm with the greatest priority capability is reselected from the remaining candidate algorithms as the alternative algorithm;

[0043] If not, execute the steps after "obtaining at least one associated parameter on which the alternative algorithm depends".

[0044] To achieve the above-mentioned purpose, according to one aspect of the present invention, there is provided an algorithm selection device for a network digital model, comprising at least one processor and a memory, wherein the at least one processor and the memory are connected via a data bus, and the memory stores instructions that can be executed by the at least one processor, and after being executed by the processor, the instructions are used to complete the algorithm selection method described in the first aspect.

[0045] In general, the above technical solutions conceived by the present invention have the following beneficial effects compared with the existing technology: the embodiment of the present invention proposes an algorithm selection method and corresponding device for a network digital model, decoupling the algorithm from the digital model, selecting the algorithm based on the algorithm capability and the accuracy that the algorithm can achieve, and improving the generalization ability and semi-autonomous behavior ability of the digital model.

[0046] Furthermore, the algorithms of digital models can support real-time updates, and the algorithm capabilities can also support real-time updates, thereby enabling the digital model to flexibly select algorithms, support flexible selection of digital models, and flexible negotiation and organization between digital models, thereby improving the generalization ability and semi-autonomous behavior ability of the digital model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of an algorithm selection method for a network digital model provided by an embodiment of the present invention;

[0048] Figure 2 This is a general framework diagram of an algorithm selection for a network digital model provided by the present invention;

[0049] Figure 3 This is a schematic diagram of the relationship between a digital model, an algorithm, and a sensor provided by the present invention;

[0050] Figure 4 This is a schematic diagram of a process for updating the accuracy of an algorithm provided by the present invention;

[0051] Figure 5 It is a flow chart of an update algorithm capability provided by the present invention;

[0052] Figure 6 This is a schematic diagram of dependencies between network digital models provided by the present invention;

[0053] Figure 7 This is a schematic diagram of an algorithm selection for a network digital model provided by the present invention;

[0054] Figure 8 This is a schematic diagram of an example of algorithm selection for a network digital model provided by the present invention;

[0055] Figure 9 The present invention provides a schematic structural diagram of an algorithm selection device for a network digital model. DETAILED DESCRIPTION

[0056] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0057] Example 1:

[0058] Current digital models are tightly coupled to algorithms, preventing flexible algorithm selection and making it difficult to improve their adaptability to scenarios and environments. Furthermore, they lack the ability to measure algorithm capabilities, particularly their real-time application capabilities, effectively and flexibly supporting the evaluation and selection of superior algorithms. There is also no algorithm update mechanism, making adaptive updates impossible or updates orchestrated by the digital model when, for example, an algorithm-dependent sensor experiences an anomaly or malfunction.

[0059] To solve the above problems, this embodiment provides an algorithm selection method for a network digital model. The algorithm selection method supports the measurement and update of algorithm capabilities, provides a basis for algorithm selection, and flexibly selects algorithms based on the algorithm update mechanism and algorithm capability measurement method. Figure 1 and Figure 2 , the algorithm selection algorithm specifically includes:

[0060] Step 10: Obtain performance metrics and accuracy requirements, and obtain at least one candidate algorithm based on the performance metrics and accuracy requirements;

[0061] Each algorithm is equipped with an algorithm description, based on which you can select the appropriate algorithm for a specific scenario and the accuracy requested by the algorithm.

[0062] The format of the algorithm description is: [parameter list, algorithm identifier, performance measurement indicators, performance requirements, [dynamic description]], where the format of [dynamic description] is [algorithm indicator accuracy, [parameter identifier 1, parameter accuracy, whether it can be finely characterized], [parameter identifier 2, parameter accuracy, whether it can be finely characterized], ..., [parameter identifier n, parameter accuracy, whether it can be finely characterized], fitting parameter 1, fitting parameter 2, ..., fitting parameter m].

[0063] The parameter list includes the parameter types that the algorithm relies on, and the algorithm identifier includes the version number, algorithm source, etc. Performance requirements include computing performance requirements, algorithm characterization capability requirements, and perception performance requirements.

[0064] Dynamic description of algorithms that can be dynamically updated: used to describe whether parameters are performance metrics of other digital models, and can be independently selected in algorithm selection based on accuracy requirements, etc.

[0065] For example, the performance metric S of a digital model (such as digital model A) is generally a measurable output. Therefore, when another digital model (such as digital model B) relies on this performance metric S, it can be added as an input parameter to the parameter list of digital model B. Furthermore, when digital model B utilizes this type of indicative parameter, it must meet certain accuracy requirements. A dynamic description can be understood as follows: if the performance metric S output by digital model A fails to meet the accuracy requirements of digital model B, digital model B will select the performance metric S output by another digital model as its input parameter.

[0066] Furthermore, when a parameter can be finely characterized, it can be expanded into several nested sub-digital models using the finely characterized markers, further enhancing the fine characterization of the digital model. This enhanced fine characterization capability also facilitates fault tracing and location. Fine characterization means that some input parameters of the current model become output parameters of other models, reflected in the other model, enabling more detailed description and characterization.

[0067] In this embodiment, the algorithm description of the algorithm to be selected is parsed, and an algorithm having the performance measurement index and algorithm index accuracy that meets the algorithm requirements is selected as the algorithm to be selected.

[0068] The relationship between digital models and algorithms is as follows: digital models can be supported by a variety of algorithms. A running digital model can select one of these algorithms for operation. The algorithm can then calculate corresponding performance metrics. In actual application scenarios, it is necessary to combine various data such as production processes and real-time changes in physical facilities to select the appropriate algorithm to accurately depict network changes and status.

[0069] Furthermore, the algorithm is strongly related to the parameter list, which includes static parameters and dynamic parameters. Both static parameters and dynamic parameters are input data of the digital model, but static parameters generally do not change significantly, such as the type description, calibration nominal value and other data of the EDFA amplifier.

[0070] An algorithm's dynamic parameters fall into two main categories: probe parameter data provided by the probe components supporting the algorithm, and indicative parameters derived from other models. These indicative parameters can be understood as the values ​​of performance metrics calculated by other models, which are the outputs of these models. For example, the EDFA's pump power and signal input power are dynamic parameters. The pump power is derived from the indicative parameters of the pump source's digital model, while the signal input power can be derived from the probe parameter data provided by the OPM. The accuracy of the algorithm depends not only on its own fitting parameters but also on the accuracy of the corresponding probe components and the indicative parameters output by other models.

[0071] In this embodiment, algorithm update is a prerequisite for algorithm selection. Algorithm update includes algorithm capability update and algorithm indicator accuracy update. Appropriate algorithm update method helps to better select the algorithm. Detailed description of algorithm update is provided below.

[0072] Step 20: Obtain the algorithm capability of each algorithm, sort all the candidate algorithms according to the algorithm capability, and prioritize the algorithm with the greatest algorithm capability as the candidate algorithm;

[0073] In actual application scenarios, obtaining the algorithm capability of each algorithm includes:

[0074] Dynamically obtain the algorithm characterization degree M, perception S, average computing resource consumption rate R and algorithm efficiency E;

[0075] Update the algorithm capability in real time according to the following formula;

[0076] Ability = log(M × S × R ÷ E)

[0077] Among them, Ability refers to algorithm capability.

[0078] The algorithm efficiency E is obtained as follows: algorithm efficiency E = time complexity × space complexity;

[0079] The algorithm characterization degree is obtained as follows: algorithm characterization degree M = average index precision ÷ (error degree × nested level dependency);

[0080] Among them, nested hierarchical dependencies can be requested and responded layer by layer and accumulated according to the interactive capabilities and recursive methods of the digital model.

[0081] For example, models OMS1 and OMS2 form the OCH model. Models OMS1 and OMS2 establish message channels with OCH1 to enable interaction. OMS1 is composed of OTS1 and OTS2. Models OTS1 and OTS2 establish message channels with OMS1 to enable interaction. When limited to this layer, the model's nesting level is set to 1. If the model cannot immediately provide the nesting level, a request must be made to the nested submodels.

[0082] OCH1 can send request messages to OMS1 and OMS2; OMS1 initiates request messages to OTS1 and OTS2. OTS1 and OTS2 each send a response message with a nesting level of 1 to OMS1. OMS1's nesting level is 1 + 1 = 2, which is the sum of its own and its sub-models. OMS2's nesting level is 1. OMS1 sends a response message with a nesting level of 2 to OCH1, and OMS2 sends a response message with a nesting level of 1 to OCH1. OCH selects the maximum value of 2 as the maximum nesting level of its nested sub-models, and the final calculated nesting level of OCH1 is: 2 + 1 = 3.

[0083] The perception S is obtained as follows:

[0084]

[0085] Where n is the number of probes; the perception accuracy is dimensionless and is a percentage or decimal, and its value can be obtained according to different probe parameters.

[0086] The method for obtaining the average computing resource consumption rate R is as follows:

[0087]

[0088] Where n represents the number of statistics. Each time the algorithm is run, the time consumption rate and space utilization rate are counted. When the algorithm is not used, the average computing resource consumption rate R is set to 1.

[0089] The algorithm obtained in this way has monotonicity and the ability to be statically described and dynamically updated, making it suitable for calculation updates in matrices.

[0090] In this embodiment, the algorithm capability of each algorithm is dynamically updated, and the algorithm indicator accuracy of each algorithm is also dynamically updated. Therefore, the algorithms to be selected confirmed in step 10 and the ranking results of the algorithms to be selected based on the algorithm capability in step 20 are also dynamically updated. Therefore, the appropriate algorithm can be dynamically selected based on the description of the algorithm characterization capability of the digital model itself to achieve accurate characterization of network changes and status.

[0091] Step 30: Obtain at least one associated parameter on which the alternative algorithm depends;

[0092] The associated parameters include probe parameters and / or index parameters, wherein the probe parameters are parameters supported by the probe component of the digital model, and the index parameters are output parameters from other models.

[0093] Step 40: If the accuracy of the associated parameters can meet the accuracy requirement, the alternative algorithm is used as the target algorithm of the digital model;

[0094] Step 50: If the accuracy of at least one of the associated parameters cannot meet the accuracy requirement, then reselect the algorithm with the greatest priority capability from the remaining algorithms to be selected as the alternative algorithm.

[0095] Steps 40 and 50 specifically include: if the associated parameter on which the alternative algorithm depends is a probe parameter, obtaining the detection accuracy of the probe component, and determining whether the accuracy of the probe component can meet the accuracy requirement;

[0096] If the accuracy requirement can be met, determine whether the associated parameters on which the alternative algorithm depends are indicative parameters; if the accuracy requirement cannot be met, reselect the algorithm with the greatest priority capability from the remaining candidate algorithms as the alternative algorithm, and reconfirm whether the associated parameters on which the alternative algorithm depends all meet the accuracy requirement.

[0097] If there is an indicative parameter, a request message carrying the accuracy requirement is sent to the association model corresponding to the indicative parameter, and whether the indicative parameter meets the accuracy requirement is determined according to a response message of the association model.

[0098] Specifically, after receiving the accuracy requirement, the association model selects an algorithm in the aforementioned manner until it determines whether a suitable algorithm can meet the accuracy requirement. If so, the accuracy of the selected algorithm (target algorithm) is used as the accuracy of the indicator parameter output by the association model, and a corresponding response message carrying the accuracy is sent to the requesting party. If the accuracy requirement cannot be met, a response message indicating that the accuracy requirement cannot be met is sent to the requesting party.

[0099] In this embodiment, after step 20 and before step 30, the following steps are also included: after selecting an alternative algorithm, obtaining the resource occupancy of the alternative algorithm; judging whether the resource occupancy of the alternative algorithm will cause the model efficiency of the digital model to be lower than the set efficiency threshold; if so, reselecting the algorithm with the greatest priority algorithm capability from the remaining algorithms to be selected as the alternative algorithm; if not, executing the steps after "obtaining at least one associated parameter on which the alternative algorithm depends".

[0100] In a specific application scenario, when the digital model contains 1 to K algorithms, an algorithm K is selected as the target algorithm. Then the algorithm indicator accuracy of algorithm K will be used as the committed accuracy of the digital model and can be sent to the associated model through interactive messages to facilitate the associated model to select the algorithm.

[0101] Associated models refer to other digital models registered with this digital model. For example, for digital model A, its associated models are digital models B, C, and D, etc., which are registered with digital model A. When digital model A undergoes an algorithm update, its promised accuracy also needs to be updated to digital models B, C, and D, etc., to facilitate the selection of a more optimal digital model based on promised accuracy when organizing digital models.

[0102] The algorithm selection method provided in this embodiment has at least the following beneficial effects: it can support the description of the algorithm characterization capabilities of the digital model itself, and when it serves as a dependent digital model, it can support the reference for updating the algorithm characterization capabilities of other digital models; it solves the problem of measuring the algorithm characterization capabilities and can be updated during actual algorithm application; it can use the negotiation capabilities between digital models to select a better algorithm and enable complex digital models to meet the accuracy requirements to the greatest extent possible.

[0103] Example 2:

[0104] In this embodiment, when the algorithm is used by the digital model, the accuracy of the description of the physical entity's operating mechanism needs to be considered. Therefore, the algorithm itself needs to be adaptively adjusted according to the operating environment. In the application of digital models, the algorithm is dynamically and statistically updated. Figures 4 and 5 , the dynamic update process of the algorithm is:

[0105] Dynamically update the algorithm index accuracy of the algorithm corresponding to the digital model according to the accuracy of the probe parameters, the accuracy of the index parameters and / or the fitting parameters of the algorithm; specifically, dynamically obtain the accuracy of the algorithm's index parameters and the perceived accuracy of the algorithm's probe parameters, update the algorithm index accuracy according to the accuracy of the index parameters and the perceived accuracy of the probe parameters, and update the algorithm index accuracy to the corresponding algorithm description.

[0106] Among them, the indicative parameter is an external influencing factor. The accuracy of the indicative parameter is obtained as follows: when the parameter of an algorithm is used as an indicator of an external digital model (that is, an indicative parameter), the promised accuracy of the external digital model for the indicator is obtained, and the promised accuracy is used as the parameter accuracy of the indicative parameter.

[0107] In this embodiment, the algorithm capability is also dynamically updated based on probe accuracy, algorithm execution conditions, time complexity, space complexity, and / or nesting level dependency.

[0108] The relationship between digital models, algorithms and sensors (probes) is shown in the figure Figure 3 As shown. Probe parameters (corresponding to Figure 3 The sensor's actual accuracy and drift are monitored, and the sensor's perception accuracy is determined based on the drift and actual accuracy. Furthermore, when the sensor's perception accuracy changes, the perception power S also changes to update the algorithm's capabilities.

[0109] Dynamically obtain time complexity and space complexity, and dynamically update the algorithm efficiency based on the time complexity and space complexity to update the algorithm capability; each time the algorithm is run, the time consumption rate and space utilization rate are calculated once, and the average computing resource consumption rate R is updated according to the calculation formula in the previous article to update the algorithm capability.

[0110] During the use of the digital model, the algorithm capabilities and algorithm indicator accuracy change dynamically, which can support the model to dynamically select the appropriate algorithm to accurately depict network changes and status.

[0111] Furthermore, in actual use, the algorithm's fitting parameters can be adjusted according to the application scenario to update the algorithm's index accuracy. If a certain algorithm requires the use of fitting parameters {a1, a2, ...am}, then by adjusting the value of one or more fitting parameters, the final algorithm index accuracy can be changed. Therefore, each algorithm index accuracy can correspond to multiple sets of adjusted fitting parameter values. For example, for the function equation y = a1·x1+a2·x2+b, x1 and x2 are parameters, which are updated through input methods such as dynamic data obtained by probe sampling, and y is updated through function solution; a1 and a2 are fitting parameters, which generally do not change under specific algorithm index accuracy and application environment. However, when changing the application scenario and algorithm index accuracy, it is necessary to adjust the fitting parameters a1 and a2 to achieve the purpose of fine-tuning the function equation to adapt to the new environment and accuracy requirements.

[0112] The updating process of the digital model algorithm is as follows:

[0113] When a new algorithm is updated, the algorithm information is added according to the algorithm description; the digital model updates and merges the parameter set in the current digital model according to the parameter list in the algorithm information. Specifically, the parameter types in the algorithm's parameter list are compared with the parameter types in the digital model's parameter set one by one to determine which parameter types need to be added to the digital model's parameter set. After the digital model's parameter set is updated, it is determined whether the probe component of the digital model supports data collection of the newly added parameter types. If not, the algorithm cannot be run and cannot be selected in subsequent algorithm updates. Specifically, the algorithm capability of the algorithm can be set to 0 to prevent the algorithm from being selected. If supported, the algorithm capability of the algorithm is obtained based on the algorithm characterization degree M, perception S, average computing resource consumption rate R, and algorithm efficiency E.

[0114] When the algorithm is not a brand new algorithm, the fitting parameters of the algorithm are updated according to the algorithm request accuracy through the correction module to ensure that the algorithm indicator accuracy of the algorithm can be higher than the algorithm request accuracy.

[0115] Among them, the fitting parameters are generally strongly correlated with the algorithm. By adjusting the fitting parameters, the accuracy of the algorithm indicators can be fine-tuned. In actual use, you can directly create a new list and add the modifiable fitting parameters of all algorithms to the list. The modifiable fitting parameters can be adjusted to update the accuracy of the algorithm indicators.

[0116] The calibration module verifies probe accuracy and updates probe status. Dynamic data sampled by the probe serves as an input parameter for algorithms or digital models and is generally reliable. However, as the probe ages due to environmental factors, time, and material degradation, its accuracy may decrease, failing to meet the basic input requirements of the algorithm or digital model. In this case, the sampled data must be corrected based on probe drift to verify and update probe accuracy.

[0117] The parameters of the judgment algorithm can be finely characterized. When a parameter can be finely characterized, the parameter is recorded as an indicative parameter, the indicative parameter is marked, and the indicative parameter is associated with the corresponding model. The fact that it can be finely characterized means that the parameter can be used as an indicator and can be finely characterized by a certain digital model. Specifically, the indicative parameters of the digital model can be coded Code_1, Code_2, ..., Code_n, and Code_x (x = 1, 2...n) is used to search for digital models with Code_x in all digital models, and the digital model is associated with the indicative parameter.

[0118] During the algorithm update process, interaction messages need to be sent to related models. The formats of the interaction messages are as follows: algorithm update request message format: [source identifier: target identifier: algorithm update request], algorithm update response message format: [source identifier: target identifier: algorithm update response], accuracy requirement commitment notification message format: [source identifier: measurement indicator: accuracy requirement commitment].

[0119] In practice, algorithm updates must be supported both offline and online. Offline updates generally do not require the publication of guaranteed accuracy via notification messages. However, in online scenarios, when the algorithm of a digital model A is updated, its algorithm, accuracy, and other information must be updated within the model itself. Furthermore, the associated accuracy must be published to other associated digital models through the interactive module.

[0120] When a digital model performs an algorithm update, it is also necessary to inform the digital model to prepare for the algorithm update. The source is generally not the digital model, but the algorithm update party, which can be a dedicated update agent. Then a digital model A is requested to perform the algorithm update; after the digital model A completes the algorithm update, it responds to the update agent and the algorithm update is completed.

[0121] The following describes the algorithm update process:

[0122] In actual use, the target algorithm selected by the digital model itself is used to calculate the sampled data and / or the input data input by the upstream model to obtain the calculation results; the calculation results are identified by the reference interval;

[0123] If the calculation result is abnormal data, the fitting parameters of the target algorithm are adjusted. If at least one normal value cannot be obtained after a preset number of adjustments, the self-digital model is triggered to reselect the target algorithm.

[0124] In this embodiment, the algorithm is adjusted in two main cases, as follows:

[0125] (1) First, adjust the fitting parameters of the target algorithm. If the calculation result is normal data after a preset number of adjustments, the corresponding fitting parameters are updated to the target algorithm.

[0126] In this embodiment, an adjustment step of the fitting parameters is selected based on the exploration efficiency, and the fitting parameters are adjusted based on the adjustment step to update the target algorithm. A trial calculation is performed using the updated target algorithm to obtain a trial calculation result until the trial calculation result is opposite to the result before the trial calculation.

[0127] The reverse dichotomy method is used to adjust the fitting parameters corresponding to the target algorithm in the previous trial calculation until the trial calculation result is opposite to the result of the previous trial calculation; the fitting parameter combination at this time is updated to the target algorithm to achieve the purpose of updating the target algorithm.

[0128] (2) If after the preset number of adjustments, at least one normal value cannot be obtained, it means that the self-digital model is triggered to reselect the target algorithm. This process is as follows Figure 1 Just reselect the method shown.

[0129] The following describes how to adjust the algorithm's fitting parameters to update the algorithm, specifically including the following steps:

[0130] (1) selecting an adjustment step of the fitting parameters according to the exploration efficiency, and adjusting the fitting parameters according to the adjustment step to obtain a first fitting parameter combination;

[0131] Specifically, the number of explorations corresponding to the change of the trial calculation result from abnormal data to normal data is obtained. If the number of explorations is less than a set threshold, the adjustment step is reduced; if the number of explorations is greater than the set threshold, the adjustment step is increased.

[0132] (2) The first fitting parameter combination is updated to the target algorithm, and a trial calculation is performed using the updated target algorithm to obtain a trial calculation result B.

[0133] (3) If the trial calculation result B is in the same state as the result A before the trial calculation, continue to adjust the fitting parameters according to the selected adjustment step until the trial calculation result B is in the opposite state to the result A before the trial calculation.

[0134] If the trial calculation result B is the same as the result A before the trial calculation, the process returns to step (1) and updates the first fitting parameter combination. If the trial calculation result B is the opposite of the result A before the trial calculation, the process jumps to step (4).

[0135] (4) Use the reverse dichotomy method to adjust the fitting parameters corresponding to the target algorithm in the previous trial calculation to obtain a second fitting parameter combination, update the second fitting parameter combination to the target algorithm, and use the updated target algorithm to perform a trial calculation to obtain the trial calculation result C.

[0136] Among them, if the adjustment step corresponding to the previous trial calculation is greater than 1 (enlargement), the second fitting parameter combination is half of the fitting parameters corresponding to the target algorithm in the previous trial calculation; if the adjustment step corresponding to the previous trial calculation is less than 1 (reduction), the second fitting parameter combination is twice the fitting parameters corresponding to the target algorithm in the previous trial calculation.

[0137] (5) If the trial result C is in the same state as the trial result B before the trial calculation, the fitting parameters are adjusted according to the reverse dichotomy method until the trial result C and the trial result B are in opposite states, and the fitting parameters are updated to the fitting parameter list in the algorithm description. In other words, in this process, "continue to adjust the fitting parameters according to the reverse dichotomy method" means to execute step (4) again, and the fitting parameter combination corresponding to the opposite state of the trial result C and the trial result B is updated to the target algorithm, thereby achieving the purpose of updating the target algorithm.

[0138] If the trial calculation result C is the same as the trial calculation result B before the trial calculation, the process proceeds to step 404. The same trial calculation result status before and after the trial calculation means that if the data before the trial calculation is abnormal, then the data after the trial calculation is also abnormal; if the data before the trial calculation is normal, then the data after the trial calculation is also normal. The opposite status of the calculation result before and after the trial calculation means that if the data before the trial calculation is abnormal, then the data after the trial calculation is normal; if the data before the trial calculation is normal, then the data after the trial calculation is abnormal.

[0139] Steps (1) to (3) are forward fine-tuning, and steps (4) to (5) are reverse fine-tuning. Forward fine-tuning and reverse fine-tuning are understood as follows: forward fine-tuning is to solve from the initial normal result to the abnormal result; or from the initial abnormal result to the normal result; reverse fine-tuning is to solve from the abnormal result of forward fine-tuning to the normal result; or from the normal result of forward fine-tuning to the abnormal result.

[0140] In this embodiment, "adjusting the fitting parameters of the target algorithm" has two functions: 1. If the calculation result is abnormal data, the target algorithm is updated by "adjusting the fitting parameters of the target algorithm"; 2. Active exploration is performed using the fitting parameter combinations generated in this process to enrich the data set and obtain a better new reference interval based on the mean and variance.

[0141] Example 3:

[0142] Based on the above implementation 1, combined Figure 7 This embodiment provides a specific algorithm selection method, and the specific process is as follows:

[0143] 1. Digital model M receives performance metrics and accuracy requests from digital model N;

[0144] 2. The digital model M determines the set of selectable algorithms [A1, A2, ...Ak, ..., An] based on the performance metrics. Algorithms with a characterization ability of 0 are directly eliminated and sorted: Sort({AbilityA1, ...AbilityAn}).

[0145] It should be noted here that the update of the characterization capability of the algorithm is continuously carried out based on statistical data, so that the characterization capability of different algorithms can be continuously refreshed, the algorithms can be sorted, and the optimal algorithm after sorting can be further selected.

[0146] 3. Traverse each algorithm Ak in sequence from large to small.

[0147] 4. Determine whether the algorithm Ak has available computing resources based on the efficiency of the digital model. If so, proceed to step 5. Otherwise, jump to step 3.

[0148] 5. Combination Figure 6 , the Ak algorithm is preliminarily selected as the digital model algorithm, and an accuracy request is initiated to the nested sub-models M.SubM1 and M.SubM2.

[0149] 6. After receiving the accuracy request, M.SubM1 and M.SubM2 similarly select the algorithm and respond with the accuracy commitment to the digital model M.

[0150] 7. When the digital model M receives the accuracy commitments from all nested sub-digital models M.SubM1 and M.SubM2, the digital model determines to select the algorithm.

[0151] 8. If the accuracy of at least one of the sub-digital models received by the digital model M cannot be satisfied, jump to step 3.

[0152] 9. Send an algorithm selection success response message to the digital model N. If a suitable algorithm cannot be selected to meet the accuracy requirement, a precision unsatisfactory response message is sent to the requesting party, and the digital model N needs to reselect a digital model.

[0153] Example 4:

[0154] Combine Figure 8Based on actual application scenarios, this embodiment provides a digital model algorithm selection method. The application scenario is: an optical channel (OCH) consists of optical devices such as a transmitter, a combiner, optical fibers, amplifiers, splitters, and receivers. The fiber segments from the combiner to the amplifier, from the amplifier to the splitter, and the connecting fiber segments are called an optical multiplexing section (OMS). This is the most typical transmission section in the optical channel and is representative of the transmission path.

[0155] Since the algorithm selection process is similar, this article will only describe a specific embodiment of an OMS. An OMS consists of an optical transmission segment 1 (OTS1) and an optical transmission segment 2 (OTS2); an OTS consists of a long-distance optical fiber (Fiber) and an optical amplifier (OA), see the schematic diagram. Figure 8 shown.

[0156] Specifically, OCH consists of OMS1, OMS2, and OMS3. The model instance corresponding to OCH is O1. The digital model instances corresponding to OMS1, OMS2, and OMS3 are M1, M2, and M3, respectively. M1's algorithm needs to be updated. M1, M2, and M3 have a sequential dependency relationship; M1 and O1, M2 and O1, and M3 and O1 have nested dependencies.

[0157] Model M1 already has four algorithms: A1, A2, A3, and A4. The algorithm update agent initiates an algorithm update request to model M1. This request includes algorithm A4, its parameter descriptions, and accuracy requirements. This information is sent to model M1 via a request message. For example, the algorithm update agent includes information such as algorithm A4, A4's parameter list, and accuracy requirements. After receiving the algorithm update request, model M1 updates algorithm A4 using the algorithm update method. A successful update is then returned to the algorithm update agent via an algorithm update response message. Model M1 then sends notification messages to models M2 and O1.

[0158] The OMS model receives OSNR and accuracy requests. The OMS model currently supports four algorithms, A1, A2, A3, and A4, to characterize the OSNR metric. The parameters in A1, A2, and A4 depend on the OTS model's characterization. A3 is a black-box algorithm and does not rely on OTS.

[0159] A1, A2, and A3 require the support of sensor S1. The current S1 accuracy is 99.9% after normalization. A4 requires the support of sensors S1 and S2. The current S2 accuracy is 60% after normalization (less than the set threshold), which means that S2 may not be able to be used normally. Therefore, the capability value of A4 is 0.

[0160] After the algorithm update, if A1, A2, A3, and A4 are not used, the default values ​​are used for the algorithm characterization capability calculation. After repeated use, A1, A2, A3, and A4 are updated based on their respective application statistics and the algorithm characterization capability calculation method. A1's ability is calculated to be a1; A2's ability is calculated to be a2; A3's ability is a3; and A4's ability is a4 = 0. After sorting, it is found that a2 > a1 > a3 > a4 = 0. Therefore, the algorithm's ability ranking order is A2, A1, A3. A4 is directly eliminated.

[0161] The OMS model initially selected A2. After calculating the model's efficiency, it was determined that A2 would cause the model to be in a state of unavailable computing resources for a long time, so this algorithm was abandoned.

[0162] Next, the OMS model selects A1. Based on the computational resources required by the algorithm, it is able to meet the requirements. Furthermore, the OMS associates the OTS with the parameters involved in the A1 algorithm and initiates an OSNR accuracy commitment request (±0.5dB in all cases) to OTS1 and OTS2. The OTS1 model also determines, based on the algorithm selection method, that it cannot meet the accuracy commitment requested by the OMS model. Therefore, it sends a response message to the OMS model stating that the accuracy cannot be met.

[0163] After receiving a response from OTS1 stating that the accuracy requirements cannot be met, the OMS model reselects A3. A3's computing resources are determined to meet the requirements. Because A3 is a black-box algorithm and does not rely on OTS1 or OTS2, the OMS model selects A3 and returns A3's accuracy as a commitment in a response message to the request source.

[0164] Example 5:

[0165] Based on the algorithm selection method provided in the above embodiments 1 to 4, the present invention also provides a device that can be used to implement the above method, such as Figure 9 FIG2 is a schematic diagram of the device architecture of an embodiment of the present invention. The device of this embodiment includes one or more processors 21 and a memory 22. Figure 9 A processor 21 is taken as an example.

[0166] The processor 21 and the memory 22 may be connected via a bus or other means. Figure 9 The bus connection is taken as an example.

[0167] The memory 22, as a non-volatile computer-readable storage medium for the algorithm selection method, can be used to store non-volatile software programs, non-volatile computer executable programs, and modules, such as the algorithm selection method in Example 1. The processor 21 executes the non-volatile software programs, instructions, and modules stored in the memory 22 to execute various functional applications and data processing of the device, thereby implementing the algorithm selection methods of Examples 1 to 4.

[0168] The memory 22 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 22 may optionally include a memory remotely located relative to the processor 21, and such remote memory may be connected to the processor 21 via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0169] The program instructions / modules are stored in the memory 22 and, when executed by the one or more processors 21, perform the algorithm selection method in the above embodiment, for example, perform the algorithm selection method described above. Figure 1 The steps shown.

[0170] Those skilled in the art will understand that all or part of the steps in the various methods of the embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a disk or an optical disk, etc.

[0171] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for selecting an algorithm for a network digital model, characterized in that: include: Obtaining a performance metric and an accuracy requirement, and obtaining at least one candidate algorithm based on the performance metric and the accuracy requirement; Obtaining the algorithm capability of each candidate algorithm, sorting all candidate algorithms according to the algorithm capability, and prioritizing the candidate algorithm with the greatest algorithm capability as the candidate algorithm; Obtaining at least one associated parameter on which the alternative algorithm depends; If the accuracy of the associated parameters can meet the accuracy requirement, the alternative algorithm is used as the target algorithm of the digital model; If the accuracy of at least one of the associated parameters cannot meet the accuracy requirement, a candidate algorithm with the greatest algorithm capability is reselected from the remaining candidate algorithms as an alternative algorithm.

2. The algorithm selection method according to claim 1, wherein: The associated parameters include probe parameters and / or index parameters, wherein the probe parameters are parameters supported by the probe component of the digital model, and the index parameters are output parameters from other models; the algorithm selection method includes: If the associated parameter on which the alternative algorithm depends is a probe parameter, then obtaining the detection accuracy of the probe component and determining whether the accuracy of the probe component can meet the accuracy requirement; If the accuracy requirement can be met, determining whether the associated parameters on which the alternative algorithm depends are indicative parameters; If there is an indicative parameter, a request message carrying the accuracy requirement is sent to the association model corresponding to the indicative parameter, and whether the indicative parameter meets the accuracy requirement is determined according to a response message of the association model.

3. The algorithm selection method according to claim 1, wherein: The obtaining of the algorithm capability of each of the to-be-selected algorithms includes: Dynamically obtain the algorithm characterization degree M, perception S, average computing resource consumption rate R and algorithm efficiency E; Update the algorithm capability in real time according to the following formula; Abilty=log(M×S×R÷E) Among them, Ability refers to algorithm capability.

4. The algorithm selection method according to claim 3, wherein: The algorithm efficiency E is obtained as follows: algorithm efficiency E = time complexity × space complexity; The algorithm characterization degree is obtained as follows: algorithm characterization degree M = average index precision ÷ (error degree × nested level dependency); The perception S is obtained as follows: Where n is the number of probes; The method for obtaining the average computing resource consumption rate R is as follows: Where n represents the number of statistics. Each time the algorithm is run, the time consumption rate and space utilization rate are counted. When the algorithm is not used, the average computing resource consumption rate R is set to 1.

5. The algorithm selection method according to claim 1, wherein: The algorithm selection method further includes: Dynamically updating the algorithm index accuracy of the algorithm corresponding to the digital model according to the sensing accuracy of the probe parameters, the accuracy of the index parameters and / or the fitting parameters of the algorithm; Dynamically update algorithm capabilities based on probe accuracy, algorithm execution, time complexity, space complexity, and / or nesting level dependencies.

6. The algorithm selection method according to claim 1, wherein: Each algorithm is equipped with an algorithm description, and the format of the algorithm description is: [parameter list, algorithm identifier, performance measurement indicators, performance requirements, [dynamic description]], where the format of [dynamic description] is [algorithm indicator accuracy, [parameter identifier 1, parameter accuracy, whether it can be finely characterized], [parameter identifier 2, parameter accuracy, whether it can be finely characterized], ..., [parameter identifier n, parameter accuracy, whether it can be finely characterized], fitting parameter 1, fitting parameter 2, ..., fitting parameter m].

7. The algorithm selection method according to claim 6, wherein: The obtaining of the performance metric and the accuracy requirement, and obtaining at least one candidate algorithm according to the performance metric and the accuracy requirement includes: The algorithm description of the algorithm to be selected is parsed, and an algorithm having the performance measurement index and algorithm index accuracy that meets the algorithm requirements is selected as the algorithm to be selected.

8. The algorithm selection method according to claim 6, wherein: The algorithm selection method further includes: When updating a new algorithm, add algorithm information according to the algorithm description; According to the parameter list in the algorithm information, the parameter set in the current digital model is updated and merged; After the parameter set of the digital model is updated, determine whether the probe component of the digital model supports data collection of the newly added parameter type. If not, set the algorithm capability of the algorithm to 0; if supported, obtain the algorithm capability of the algorithm based on the algorithm characterization degree M, perception S, average computing resource consumption rate R, and algorithm efficiency E.

9. The algorithm selection method according to any one of claims 1 to 8, wherein: The algorithm selection method further includes: After selecting an alternative algorithm, obtaining resource usage of the alternative algorithm; Determining whether resource usage of the alternative algorithm will cause the model efficiency of the digital model to be lower than a set efficiency threshold; If so, the algorithm with the greatest priority capability is reselected from the remaining candidate algorithms as the alternative algorithm; If not, execute the steps after "obtaining at least one associated parameter on which the alternative algorithm depends".

10. An algorithm selection device for a network digital model, characterized in that: The method comprises at least one processor and a memory, wherein the at least one processor and the memory are connected via a data bus, and the memory stores instructions that can be executed by the at least one processor, and after the instructions are executed by the processor, they are used to complete the algorithm selection method described in any one of claims 1 to 9.

Images