Automatic computing power network arrangement system and method based on parameter dependence algorithm

Abstract

The invention provides an automatic computing power network arrangement system and method based on a parameter dependency algorithm. The automatic computing power network arrangement system comprises an interface document structured processing module, a parameter vectorization module, a dependency relationship identification module, a topological sorting module and an automatic process generation module. According to the invention, the automation and intelligence degrees of multi-element arrangement of the computing network are improved, the user operation experience is remarkably improved, the labor cost is reduced, the adaptability and flexibility of the system are enhanced, and the innovation and efficient delivery of computing network products are finally promoted.

Description

Technical Field

[0001] The present invention relates to the field of computing power network technology, and in particular to an automated computing power network orchestration system and method based on a parameter-dependent algorithm. Background Art

[0002] With the rapid development of computing power network technology, traditional cloud-network combination services can no longer meet the needs of complex business scenarios. In order to meet these challenges, computing power network integrated services have emerged. The core is to flexibly combine computing power, network, data, intelligence and other multi-factor capabilities, driven by business scenarios, and orchestrate computing network solutions that meet specific needs. This new service model aims to achieve a transition from traditional cloud-network combination services to integrated services with deep integration of multiple factors, thereby promoting computing network product innovation.

[0003] In the existing technology, customers usually select appropriate computing network multi-factor capabilities and perform orchestration according to their own business needs. However, this orchestration process is not just a simple stacking of factors, but requires in-depth integration of the capabilities of each factor, and completes the topology architecture, process design and parameter configuration according to the coordination and logical relationship between the factors. This complex orchestration task has a high professional threshold for most customers, especially when configuring the order of opening the factors and the parameter dependencies, which often requires the intervention of professional support personnel to complete.

[0004] Existing technologies also have the defect of complex process configuration, especially when multiple element capabilities are involved. The accurate arrangement of processes and parameters between various elements not only needs to ensure that the solution can be correctly activated, but also must take into account the timeliness of the activation. Since the orchestration work involves multi-level and multi-dimensional coordination and integration, the operation is difficult and the error rate increases accordingly. At the same time, since the process configuration relies on professional support personnel to complete manually, this high dependence on manpower directly increases the labor cost of the project, and is limited by the number and capabilities of support personnel, further limiting the innovation efficiency of computing network products.

[0005] In addition, under the existing orchestration model, the delivery time of the computing network solution is relatively long. The involvement of professional support personnel and complex process configuration will slow down the overall progress, resulting in the delay of the solution launch time, which seriously affects the innovation speed and market responsiveness of computing network products. Customers need to wait for a long time to use the orchestrated solution, which affects their business progress.

[0006] Therefore, the existing computing network multi-factor orchestration technology has problems such as poor customer operation experience, complex process configuration, large amount of manpower consumption, and extended delivery time. These problems directly affect the innovation efficiency, user satisfaction and market competitiveness of computing network products. Summary of the invention

[0007] In order to overcome the shortcomings of the prior art, the present invention proposes an automated computing power network orchestration system and method based on a parameter-dependent algorithm, which not only improves the automation and intelligence of the multi-factor orchestration of the computing network, but also significantly improves the user operation experience, reduces labor costs, and enhances the adaptability and flexibility of the system, ultimately promoting the innovation and efficient delivery of computing network products.

[0008] To achieve the above objectives, the present invention provides an automated computing power network orchestration system based on a parameter-dependent algorithm, comprising: The interface document structured processing module is used to read and preprocess the interface specification documents related to computing network products, and extract metadata such as interface name, interface code, interface service type, interface call address, interface protocol, input and output parameters; by cleaning irrelevant characters in the document (such as spaces, line breaks, invisible characters, etc.), and removing invalid lines (such as blank lines, contentless titles, duplicate lines, etc.), it ensures that the extracted metadata is accurate and complete, and provides accurate basic data for subsequent dependency judgment.

[0009] The parameter vectorization module is used to vectorize the interface input and output parameters extracted by structured extraction, and use the BERT model to convert the parameter description into a numerical vector. The vectorization processing includes word segmentation, semantic vectorization, and custom dictionary loading to improve the processing capabilities of professional terms in the computing network product interface and ensure that the semantic similarity calculation between parameters is more accurate.

[0010] The dependency identification module is used to calculate the dependencies between interfaces based on the vectorized parameters; by traversing the input parameters of each product interface and the output parameters of other product interfaces, their cosine similarity is calculated. If the similarity is higher than the preset threshold (such as 0.85), it is determined that a dependency exists, and an interface dependency graph is established to mark the calling order between each product interface.

[0011] The topological sorting module is used to generate a business process diagram based on the dependency graph through a topological sorting algorithm (such as the Kahn algorithm); this module ensures that the generated process has no circular dependencies, calls each product interface in sequence, and generates correct business processes such as activation and cancellation.

[0012] The automatic process generation module is used to automatically generate the business process of the computing power network solution after topological sorting, including the input and output parameter configuration of the business process; the system automatically associates the corresponding parameters according to the dependency of the interface, reduces manual intervention, quickly generates personalized business processes, and supports customized optimized solutions based on business scenarios.

[0013] Furthermore, it also includes: The parameter filtering module is used to pre-process the interface input parameters before vectorization, excluding parameters that do not change with the business scenario and can be assigned default values, as well as invalid historical parameters, to reduce the computing burden and improve system processing efficiency.

[0014] Furthermore, it also includes: The interface dependency library is used to store the generated interface dependencies, including the input and output parameter association information of each product interface. Through the dependency library, the system can quickly infer and generate processes for different business solutions, thereby improving the efficiency of business process generation.

[0015] Furthermore, it also includes: The parallel task optimization module is used to identify product interfaces that can be executed in parallel; by analyzing the dependencies between interfaces, interfaces without dependencies are grouped and executed in parallel, thus reducing the execution time of the solution and optimizing process performance.

[0016] Furthermore, it also includes: The process visualization interface is used to display the generated solution topology and business process diagram, and supports users to make real-time adjustments to the product interface sequence and parameter configuration in the process through a graphical interface to enhance the user operation experience.

[0017] Furthermore, it also includes: The process optimization module is used to optimize the generated business processes; the system simplifies the process, reduces repetitive operations and improves the overall execution efficiency of the system by eliminating redundant operations and merging the same or similar parameters.

[0018] An automated computing power network orchestration method based on a parameter-dependent algorithm, applicable to any of the automated computing power network orchestration systems based on a parameter-dependent algorithm described above, comprising: Step 1: Read the interface document of the computing network product, clean and structure the document, and extract the interface metadata and parameter metadata; Step 2: Use word embedding technology to semantically vectorize the extracted parameters and filter out unnecessary parameters; Step 3: Calculate the dependencies between different interfaces based on the vectorized parameters and build a dependency graph; Step 4: Sort the dependency graph using a topological sorting algorithm to generate the correct interface call sequence; Step 5: Generate a business process diagram based on the topological sorting results, identify interfaces that can be executed in parallel, and optimize the process execution order; Step 6: Merge similar parameters in the generated process, remove redundant parameters, and build an optimized parameter tree; Step 7: Optimize the execution efficiency of business processes by reducing redundant operations and parallelizing execution.

[0019] Furthermore, step 1 is as follows: Step 11: Read the interface specification documents related to the computing network product; Step 12: Clean up irrelevant characters (such as spaces, tabs, line breaks, illegal characters, etc.) in the document; Step 13: Remove invalid lines in the document (such as blank lines, comment lines, duplicate lines, etc.), and extract interface metadata and parameter metadata, including interface name, interface code, product to which the interface belongs, service type (activation, change, cancellation, etc.), input and output parameters and their attributes; Step 14: Use natural language processing tools (such as THULAC word segmentation tool) to segment the interface description and parameter description to extract descriptive key information; Step 2 is as follows: Step 21: Process the extracted parameter metadata using word embedding technology and use the BERT model to convert the parameter text into a semantic vector; Step 22: Filter out default parameters that do not need to be dynamically changed and invalid historical parameters to reduce unnecessary performance consumption; Step 23: Concatenate the parameter description information and other related metadata into a string and input it into the BERT model to generate a feature vector of the parameter for subsequent dependency calculation.

[0020] Furthermore, step 3 is as follows: Step 31: Construct the input and output parameter set of the product interface, and extract the input parameter (Pin) and output parameter (Pout) of each product interface; Step 32: Calculate the cosine similarity between the output parameters of the interfaces between different products and the input parameters of other products; Step 33: Determine whether there is a dependency relationship between the interfaces based on the similarity value. If the cosine similarity is greater than a preset threshold (such as 0.85), it is considered that the two interfaces have a dependency relationship, and the dependency relationship is recorded. Step 4 is as follows: Step 41: Build a dependency graph of the solution based on the identified interface dependencies; Step 42: Use a topological sorting algorithm (such as Kahn algorithm) to sort the dependency graph to ensure that the starting product interface of each directed edge in the graph is called before the end product interface; Step 43: If a loop (i.e., circular dependency) is found in the graph, the user is prompted to adjust the topology structure to ensure the correctness of the process.

[0021] Furthermore, step 5 is as follows: Step 51: Generate a business process diagram based on the result of topological sorting to clarify the calling sequence of each product interface; Step 52: After the flow chart is generated, the system identifies product interfaces that can be executed in parallel, and groups interfaces that have no dependency for parallel execution; Step 53: For interfaces with dependencies, call them in sequence to ensure the logic of process execution; Step 6 is as follows: Step 61: extract the input and output parameters of each interface in the generated flow chart, and construct the parameter tree structure of the input and output parameters; Step 62: according to the identified inter-interface dependency, remove the redundant parameters that have been relied upon in the parameter tree; Step 63: Calculate the semantic similarity of the remaining parameters, merge similar parameters, and process them uniformly in the process to reduce repeated operations; Step 7 is as follows: Step 71: After generating the flow chart, continue to identify redundant operations in the process to reduce unnecessary repeated calls; Step 72: Further improve the process execution efficiency by optimizing and merging similar parameters and parallelizing the execution strategy.

[0022] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention provides an automated computing network orchestration system and method based on a parameter-dependent algorithm. By introducing a topological sorting algorithm and a semantic vectorization algorithm, the system can automatically analyze the interface specification documents of computing network products, build an interface dependency library, and automatically generate the business process of the solution and its input and output parameters. This automated function greatly reduces the need for manual intervention, improves the accuracy and efficiency of orchestration, and makes the complex computing network orchestration process more intelligent.

[0023] 2. The present invention provides an automated computing network orchestration system and method based on a parameter-dependent algorithm. The system provides an intuitive visual orchestration interface. Users can complete the topology architecture orchestration of the solution by simply dragging and dropping. The system then automatically generates business processes and parameter configurations. This greatly simplifies the traditional manual intervention process, not only saving time, but also allowing users to complete complex orchestration tasks more flexibly and quickly, significantly improving the user experience.

[0024] 3. The present invention provides an automated computing network orchestration system and method based on parameter-dependent algorithms. The existing technology relies on a large amount of manual operations, while this system significantly reduces the reliance on professionals and reduces labor costs through intelligent automatic orchestration. At the same time, since the automated process improves the orchestration efficiency, the online delivery time of the solution is also shortened, which directly improves the overall benefits and market competitiveness of the computing network business.

[0025] 4. The present invention provides an automated computing network orchestration system and method based on a parameter-dependent algorithm. The system is highly flexible and adaptable, can quickly respond to different customer needs, and can be dynamically adjusted and customized according to changes in business scenarios. This technical solution can flexibly adapt to changing business needs without multiple manual adjustments, greatly improving the system's adaptability and meeting the needs of diverse application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 It is an example diagram of the overall process of the present invention; Figure 2 It is a schematic diagram of the topological architecture of the sunshine kitchen solution of the present invention; Figure 3 This is a schematic diagram of the interface dependency of the process of opening the present invention; Figure 4 This is a schematic diagram of the optimized business process of the present invention. DETAILED DESCRIPTION

[0028] The technical solution of the present invention will be more clearly and completely explained below through description of preferred embodiments of the present invention in combination with the accompanying drawings.

[0029] The present invention performs data preprocessing and structured extraction of interface key information and parameter key information from interface specification documents related to the product capabilities of each element of the computing network (hereinafter referred to as "products"), and then uses a semantic vectorization algorithm to identify the parameter dependencies between different product interfaces, thereby establishing dependencies between different products, that is, the order of activation. After the customer selects multiple products to complete the topological architecture arrangement of the solution, the system selects the interfaces of the corresponding products according to the topological architecture and product dependencies to build an interface dependency graph, calculates and generates the business process of the solution through a topological sorting algorithm, and automatically generates the input parameters and output parameters of the solution business process by using the parameter dependencies between different product interfaces that have been vectorized in the early stage. An example of the overall process of the solution is as follows: Figure 1 shown.

[0030] Main implementation logic: Structuring of interface documents The system reads the interface specification documents related to various computing network products in advance, and pre-processes the contents of the interface documents after reading, including cleaning irrelevant characters, invalid lines and other information in the documents. The cleaning contents include: Irrelevant characters: including space characters at the beginning and end of characters, invisible characters, tab characters, line breaks, illegal characters, repeated punctuation and other irrelevant characters.

[0031] Invalid lines: include blank lines, comment lines, short characters on a single line, repeated lines, empty tables, and titles with no content.

[0032] After the cleaning is completed, the title, table (interface input parameters, output parameters), interface description and other information at all levels in the document are extracted, and the corresponding key metadata are extracted according to the two dimensions of interface and parameter: Interface metadata: interface name, interface code, interface product, interface service type (activation, change, cancellation, etc.), interface call address, interface protocol, interface description and other data.

[0033] Parameter metadata: parameter name, parameter code, attribute interface, attribute product, attribute interface service type, parameter type, parameter description, parameter value range, parameter unit and other data.

[0034] Since some of the key metadata are text descriptions, such as interface descriptions, parameter descriptions, etc., in order to ensure accurate metadata extraction, the THULAC word segmentation tool is used to segment descriptive texts such as interface descriptions and parameter descriptions, and then the required key information is obtained based on the word segmentation results.

[0035] THULAC (THU Lexical Analyzer for Chinese) is a Chinese lexical analysis toolkit developed by the Natural Language Processing and Social Humanities Computing Laboratory of Tsinghua University. THULAC can accurately segment Chinese text into meaningful vocabulary units, supports the effective recognition of common words, new words and network terms, and performs part-of-speech tagging during word segmentation. At the same time, the interface specification document contains many professional categories of vocabulary due to its high degree of professionalism. In order to further improve the accuracy of word segmentation, a set of custom dictionaries specifically for computing network products is pre-configured and loaded into THULAC for word segmentation. For example, for the parameter description "IP address of the server used to access the camera", the key information "camera server IP address" can be obtained by extracting the nouns according to the word segmentation results according to the part of speech. After the interface metadata and parameter metadata are extracted, structured data is constructed in JSON format.

[0036] Parameter vectorization The system uses the word embedding technology Bert to vectorize the parameter structured data generated in the above steps. The vectorized data will be used to judge the parameter dependency in the next step.

[0037] Word embedding technology is a method of converting words in natural language into numerical vectors. By representing each word as a multi-dimensional numerical vector, these vectors can capture the semantic relationship between words. The core idea of ​​word embedding technology is based on distributed representation, that is, the meaning of a word can be inferred from its context in the text. By mapping words into a low-dimensional continuous vector space, the performance and accuracy of natural language processing (NLP) tasks can be improved. Word embedding technology is widely used in various NLP tasks, such as text classification, machine translation, etc. It enables computers to process and understand natural language text more efficiently, thereby achieving more complex language processing tasks.

[0038] As a pre-trained language model based on the Transformer architecture, Bert has learned rich language knowledge and deep semantic representations through pre-training on a large-scale corpus. When applied to specific tasks, Bert can quickly adapt through fine-tuning, providing more accurate and powerful support for text vectorization. Therefore, the system chooses Bert as the word embedding technology in order to achieve better performance in subsequent processing.

[0039] The specific steps are as follows: When vectorizing parameters, in order to reduce unnecessary performance consumption and reduce the complexity of subsequent parameter dependency judgment, the parameters need to be filtered first. The filtering rules are as follows: For interface input parameters, they will not change with business scenarios and can be directly assigned default values.

[0040] For interface input parameters, parameters that were added for historical reasons but are no longer in use The key metadata of the extracted parameters are concatenated into a string according to the parameter dimension as the input of BERT. Among these metadata, text description attributes such as parameter description have been segmented and sorted in the early stage, and the remaining attributes are already standardized in the interface specification document, ensuring the consistency and comparability of the data, so no additional special processing is required. For example, for the camera ID parameter in a certain interface, its key metadata is: { "Parameter name": "Camera ID", "Parameter encoding": "cameraId", "Service type of the home interface": "Service activation", "Parameter type": "String", "Parameter Description": "Camera ID", "Parameter value range": "", "Parameter Unit": "" } The concatenated string is: "camera ID [SEP] cameraId [SEP] AI analysis task creation [SEP] AI algorithm [SEP] service activation [SEP] String [SEP] camera id [SEP] [SEP] [SEP]" Among them, "[SEP]" is a special tag in BERT. When processing multiple text fragments, [SEP] is used to separate these text fragments to help the model understand the structure of the input sequence.

[0041] Load a Bert model tokenizer and convert the concatenated string into a digital sequence that the Bert model can process.

[0042] from transformers import BertModel #Load the Tokenizer from the BERT model tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-uncased') # Perform word segmentation on the specified texts inputs = tokenizer(texts, return_tensors='pt', padding=True,truncation=True, max_length=512) return_tensors: This parameter specifies the data type returned by the tokenizer. 'pt' means returning PyTorch tensors.

[0043] padding: When this parameter is set to True, it instructs the tokenizer to pad all processed texts to ensure that they have the same length.

[0044] truncation: When this parameter is set to True, it instructs the tokenizer to truncate texts when necessary to ensure that their length does not exceed a certain maximum value, preventing the processing of too long texts, which may consume too much computing resources or cause the model performance to degrade.

[0045] max_length: This parameter specifies the maximum length of the text. Currently it is set to 512. All processed text will be truncated or padded to this length.

[0046] The processed text and number sequence is input into the BERT model, and the hidden state of the last layer of BERT is extracted as a feature vector. The last hidden state of BERT contains the information of the input text after being processed by the multi-layer Transformer structure. This information includes both the semantic content of the text and the relationship of the context, which is very suitable as a feature vector for subsequent tasks.

[0047] from transformers import BertModel #Load the Bert model model = BertModel.from_pretrained('bert-base-multilingual-uncased') #Use the loaded BERT model to forward propagate the input data and obtain the output of the model outputs = model(**inputs) #Extract the representation of the last hidden layer from the output of the model, take the representation of the first token, and then add a batch dimension embeddings1 = outputs.last_hidden_state[:, 0, :].unsqueeze(0) Build an interface dependency library The system builds an interface dependency library based on the interface, parameter metadata, and vectorized data. The construction process includes the following steps: Parameters are grouped in order according to the interface service type and the product in the interface and parameter metadata. For example, first extract the input and output parameters from the product C service activation interface. , It is the input parameter of product C in the service activation interface. It is the output parameter of product C in the service activation interface. The input and output parameters of the service activation interfaces of other products are extracted in sequence to finally form the product dimension parameter set of the service activation interface. . Each element in this set Contains all input and output parameters of a specific product in the service activation interface.

[0048] The dependency of interfaces is established by associating the output parameters and input parameters between interfaces. Specifically, if product A opens an output parameter in interface I1 , used as the input parameter in the B product activation interface I2 , then it means that you need to call interface I1 first, and then call interface I2, which is recorded as I1→I2.

[0049] In order to systematically identify and construct the dependencies between interfaces of different products under a certain business type, we need to During the traversal, we calculate Each output parameter below is consistent with the other elements in the collection The cosine similarity between each input parameter under (m ≠ n). With this method, we can accurately determine the dependency between interfaces, that is, if the output parameter of one interface is highly similar to the input parameter of another interface, then it can be determined that the latter depends on the former. The feature vector of a certain output parameter obtained in the previous step is embeddings1, The feature vector of the output parameter obtained in the previous step is embeddings2, and the cosine similarity calculation code is as follows: from torch.nn.functional import cosine_similarity # Calculate cosine similarity similarity = cosine_similarity(embeddings1, embeddings2) # Get the average cosine similarity between two parameters (tensors) cosine_similarity = similarity.mean().item() For each element in the set P After all traversals are completed, the cosine similarity between each input and output parameter can be calculated. The closer the cosine similarity between two parameters is to 1, the higher the degree of similarity between the two parameters in direction. Based on the current interface specifications, we set the preset threshold to 0.85. In practical applications, the setting of the specific threshold needs to be judged and adjusted according to the actual situation. When the calculated similarity exceeds this preset threshold, it can be determined that there is a correlation between the two parameters, and then it can be inferred that there is a dependency relationship between the interfaces to which they belong, and the interface dependency library is constructed accordingly. For example, in the output parameters of the video cloud activation interface, there is parameter A, and its metadata is: { "Parameter name": "Camera ID", "Parameter encoding": "cameraId", "Service type of the home interface": "Service activation", "Parameter type": "String", "Parameter Description": "Platform Camera Id", "Parameter value range": "", "Parameter Unit": "" } Among the input parameters in the AI ​​algorithm analysis task creation interface, there is a B parameter, whose metadata is: { "Parameter name": "Camera ID", "Parameter encoding": "cameraId", "Service type of the home interface": "Service activation", "Parameter type": "String", "Parameter Description": "Camera ID", "Parameter value range": "", "Parameter Unit": "" } After the cosine similarity calculation, the cosine similarity of the vectors between the two parameters = 0.9036458134651184, which shows that the two parameters are similar, and the output parameter A of the video cloud activation interface is used as the input parameter B in the AI ​​algorithm analysis task creation interface. From this, it can be concluded that if the video cloud and AI algorithm products are activated at the same time, the video cloud activation interface is required before calling the AI ​​algorithm analysis task creation interface.

[0050] Build the solution process interface dependency graph Customers create computing network solutions based on their own business needs. Through the visual interface, they select multiple suitable computing network product capabilities to orchestrate the topological architecture of the computing network solution. Based on this, the system identifies the products involved and the data relationships between products, and uses the interface relationship dependency library established in the early stage to screen out qualified interfaces and their dependencies, and respectively construct the solution's activation and cancellation process interface dependency graphs, which are presented in the form of DAG (directed acyclic graph). Taking the Sunshine Kitchen solution as an example, the detailed construction process of its activation process interface dependency graph is as follows: The topology of the Sunshine Kitchen solution is as follows: Figure 2 As shown, According to the topology, the products involved in the Sunshine Kitchen solution include: camera terminal, cloud dedicated line, video cloud, cloud networking, and AI algorithm. The data relationship between products includes: camera terminal <-> cloud dedicated line, cloud dedicated line <-> video cloud, video cloud <-> cloud networking, cloud networking <-> AI algorithm.

[0051] According to the products involved, the interfaces and their dependencies selected from the interface relationship dependency library include: cloud dedicated line→camera terminal, video cloud→camera terminal, video cloud→cloud dedicated line, video cloud→cloud networking, AI algorithm→cloud networking, AI algorithm→cloud dedicated line, video cloud→AI algorithm.

[0052] After filtering according to the data relationship between products, the interfaces and their dependencies involved in the business process of the solution are finally obtained, including: cloud dedicated line→camera terminal, video cloud→camera terminal, video cloud→cloud dedicated line, video cloud→cloud networking, AI algorithm→cloud networking, video cloud→AI algorithm.

[0053] The above relationship is converted into directed edges: (video cloud, camera terminal), (video cloud, cloud networking), (video cloud, cloud dedicated line), (video cloud, AI algorithm), (cloud dedicated line, camera terminal), (AI algorithm, cloud networking).

[0054] Directed edge is a basic concept in graph theory, which refers to an edge with direction. In a directed graph, an edge connects two vertices and has a clear starting point and an end point. Directed edges are usually represented by arrows, and the direction of the arrow is the direction of the edge. In a directed graph, the direction of the edge is very important, which determines the reachability between the vertices in the graph.

[0055] The directed edges are used to construct the open process interface dependency graph, such as Figure 3 As shown, Automatically generate solution flow chart The system uses the Kahn algorithm to topologically sort the interface dependency graph generated in the previous step, so that for each directed edge (u, v) in the graph, u (in the sorted record) appears before v, and then automatically compiles a solution flow chart based on the result of the topological sorting. The Kahn algorithm is a topological sorting algorithm for directed acyclic graphs (DAGs). It continuously removes vertices with an in-degree of 0 until all vertices are visited, thereby obtaining a topological sorting of the graph. The detailed steps are as follows: Calculate the in-degree of each vertex in the graph. In-degree refers to the number of edges pointed to by other vertices in a vertex in the graph. For example, if a→b, the in-degree of b is 1. The final in-degrees of the vertices in the graph are shown in Table 1, which are: Table 1

[0056] Create two empty queues S and Q respectively, where S is used to store the sorted vertex order; Q is used to store vertices with in-degree 0. Currently, all vertices with in-degree 0 are added to queue Q, and queue Q = [Video Cloud].

[0057] Take a vertex v from queue Q and add it to queue S. After traversing all adjacent vertices w of vertex v, for each adjacent vertex w, reduce its in-degree by 1 (that is, remove an edge from v to w). After completing this operation, if the in-degree of a vertex w becomes 0, add the vertex w to queue Q. At this point, queue S = [Video Cloud], queue Q = [Cloud Dedicated Line, AI Algorithm].

[0058] Continue to repeat the previous step until the queue Q is empty. At this time, the queue S is the sorted vertex order, that is, the linear product interface call order. Queue S = [video cloud, cloud dedicated line, AI algorithm, camera terminal, cloud networking].

[0059] If the number of vertices in queue S is less than the total number of vertices in the graph after the algorithm is executed, it means that there is a loop in the graph. Because the vertices in the loop cannot be completely removed through the above process, their in-degree will never become 0, so they cannot be added to the result list of topological sorting. In actual applications, if this happens, it is usually because the topology architecture of the solution is configured with incorrect data relationships, and the customer needs to adjust the topology architecture according to the prompts.

[0060] In order to improve the efficiency of business processes, it is necessary to continue to find product interfaces that can be called in parallel. Traverse the queue S, and for each product interface, check whether there is a dependency relationship with its subsequent product interface. If there is no dependency, it means that these product interfaces can be called and executed in parallel. These product interfaces are grouped, and the product interfaces in each group can be processed in parallel. For product interfaces with dependencies, continue to call them in the order of topological sorting. Taking the Sunshine Kitchen solution as an example, its sorted queue S = [video cloud, cloud dedicated line, AI algorithm, camera terminal, cloud networking], and the queue is judged cyclically: Take the first product interface "Video Cloud", which has a dependency relationship with the subsequent product interface "Cloud Dedicated Line" and cannot be processed in parallel, so create the first parallel group S1=[Video Cloud].

[0061] Continue to take the second product interface "Cloud Dedicated Line". There is no dependency between it and the subsequent product interface "AI Algorithm" and they can be processed in parallel. Therefore, create a second parallel group. =[Cloud dedicated line, AI algorithm].

[0062] use The product interface in the continues to judge with the subsequent interface, because There is a dependency between the "Cloud Dedicated Line" in the following product interface "Camera Terminal", so End of parallel grouping.

[0063] Continue to judge from the "camera terminal", it has no dependency with the subsequent product interface "cloud networking", and can be processed in parallel, so create a third parallel group =[Camera terminal, cloud networking]. At this time, the interface in queue S has been judged to be completed. End of parallel grouping.

[0064] After the parallel grouping is completed, the parallel groups are processed sequentially. In the parallel group, there is only the product interface "Video Cloud", so it is placed in the first link of the business process; There are two product interfaces "Cloud Dedicated Line" and "AI Algorithm" in the parallel group, which are placed in parallel in the second link of the business process. Since these two interfaces depend on the product interface "Video Cloud" in the first link, a relationship is set from "Video Cloud" to these two interfaces; There are also two product interfaces "Camera Terminal" and "Cloud Network" in the parallel grouping, which are placed in parallel in the third link of the business process. Since these two interfaces rely on the product interfaces "Cloud Dedicated Line" and "AI Algorithm" in the second link respectively, the relationship pointing to the corresponding interface in the third link is set from the product interface in the second link. At this point, the optimized business process is automatically generated. The generated optimized business process is as follows Figure 4 As shown; Automatically generate solution process input and output parameters After the business process is generated, the input and output parameters of each interface in the process are processed, and the parameters with dependencies are excluded. For the remaining parameters, they are vectorized and then cosine similarity is judged. The same parameters are merged, and finally the input and output parameters of the process are generated according to the path of the parameters. The detailed steps are as follows: Extract the input and output parameters of each interface in the process, and construct the input and output parameter trees and output parameter trees respectively. Each parameter tree has a first-level node containing a common node and several interface nodes, and the parameters are the leaf nodes on this tree. Taking the automatic construction of the input parameters of the service activation process of the Sunshine Kitchen Solution as an example, the structure of the input parameter tree is shown in Table 2: Table 2

[0065] According to the inter-interface parameter dependency relationship obtained in step C, the corresponding parameter leaf nodes in the parameter tree are identified and removed. After removing the dependent parameters, the parameter tree structure is shown in Table 3: Table 3

[0066] According to the parameter vector obtained in step B, starting from the parameters under the first interface node in the parameter tree, the cosine similarity of the parameters of each leaf node under it and the parameters of the leaf nodes under the remaining interface nodes in the parameter tree is calculated in turn. If the cosine similarity of two parameters exceeds the preset threshold (preset threshold = 0.85), it can be determined that the two parameters have the same meaning and are merged into a new parameter. Subsequently, the two original parameters are removed from the interface nodes to which they originally belonged, and the merged new parameters are added to the common node. When adding to the common node, for parameter nodes with the same meaning, the hierarchical information of the first parameter node identified as having the same meaning is used.

[0067] Remove it from the corresponding interface node and add the merged new parameter to the public node. When adding to the public node, for parameter nodes with the same meaning, take the level of the first parameter node and add it to the public node.

[0068] For example, starting from the first parameter "Order Number" under the "Video Cloud Service Activation Interface", we first calculate the cosine similarity between it and the first parameter "Group Number" under the "Cloud Dedicated Line Service Activation Interface", and the calculation result is 0.8125, which are different parameters; continue to calculate the cosine similarity with the next parameter "Customer Order Number", and the calculation result is 0.9479, which is greater than the preset threshold. Therefore, it is judged that the two parameters have the same meaning and can be merged into one parameter.

[0069] After the parameters under the first interface are traversed, continue to start from the parameters under the second interface node in the parameter tree and repeat the above steps until the parameters under all interface nodes in the parameter tree are traversed. Finally, the input parameter tree structure of the Sunshine Kitchen Solution Service Activation Process is shown in Table 4: Table 4

[0070] At this point, the parameter tree is the input or output parameter of the streamlined, merged and optimized business process.

[0071] As a specific implementation method: Application of an automated computing power network orchestration system based on parameter-dependent algorithms System deployment environment The system can be deployed on a distributed cloud platform to process a large amount of interface data of computing network products and complex business process orchestration. The system includes the following main modules: Interface document parsing module: supports interface specification documents in different formats (such as JSON, XML, and text files); BERT vectorization module: vectorizes parameters through the BERT model; Dependency identification module: calculates parameter dependencies and builds a dependency graph; Topological sorting module: sorts the dependency graph and generates an executable business process graph; Visualization module: Provides users with an intuitive interface for adjusting and viewing the generated flowchart.

[0072] The system supports containerized deployment and uses Docker to package, deploy and expand applications.

[0073] The implementation steps are as follows: Reading and preprocessing of interface documents: The system first reads the interface specification documents of the computing network products from the database or file storage. These documents may be stored in different formats, including XML, JSON, etc. The documents are stored in a file system, and each document contains the interface information of the product, such as the input parameters, output parameters, interface description, etc. of the interface.

[0074] Preprocessing process: The system uses regular expressions to identify and clean irrelevant characters (such as spaces, line breaks, etc.) in documents; Remove invalid lines (such as blank lines, duplicate lines, etc.); Extract valid data into structured data (JSON format), including interface name, input parameters, output parameters, etc.

[0075] For example, an interface document might contain the following data: { "Interface Name": "Camera Initialization", "Input parameters": [ {"Parameter name": "Camera ID", "Type": "String"}, {"Parameter name": "Camera position", "Type": "String"} ], "Output Parameters": [ {"Parameter name": "Initialization result", "Type": "Boolean"} ] } Parameter vectorization: After the interface document is preprocessed, the system uses the BERT model to vectorize the parameter description text. The system loads the pre-trained BERT model and converts the description information of each parameter into a high-dimensional vector.

[0076] For example, for the "camera ID" and "camera position" parameters in the above interface, the system converts them into vector representations of specific dimensions for subsequent similarity calculations.

[0077] The vectorization process uses the following steps: Perform word segmentation on each parameter, for example, "camera ID" will be segmented into ["camera", "ID"]; The word segmentation results are input into the BERT model to generate vector representations, such as: from transformers import BertTokenizer, BertModel tokenizer = BertTokenizer.from_pretrained('bert-base-chinese') model = BertModel.from_pretrained('bert-base-chinese') inputs = tokenizer("Camera ID", return_tensors="pt") outputs = model(**inputs) The output is a high-dimensional vector for subsequent identification of dependencies.

[0078] Dependency identification: The system calculates the similarity of the input parameters and output parameters of all interfaces, and determines whether there is a dependency between different product interfaces by calculating the cosine similarity. The specific steps for dependency identification are as follows: The output parameters of each interface are compared with the input parameters of other interfaces one by one, and the cosine similarity is calculated. Assume that the vector of output parameter A is represented as vector_A, and the vector of input parameter B is represented as vector_B, and the similarity between them is calculated: from numpy import dot from numpy.linalg import norm cosine_similarity = dot(vector_A, vector_B) / (norm(vector_A) * norm(vector_B)) When the similarity is greater than 0.85, the system considers that these parameters have a dependency relationship. For example, the output parameter of "Camera ID" may be used as the input parameter of the next interface, thus forming a dependency relationship between interfaces.

[0079] The recognition result generates a dependency graph (DAG) that records the dependency order between interfaces. The nodes of the graph represent interfaces, and the edges represent dependency relationships.

[0080] Topological sorting and flow chart generation: After dependency identification is completed, the system topologically sorts the dependency graph to determine the execution order of each product interface. The system uses the Kahn algorithm to sort the DAG to ensure that there is no circular dependency.

[0081] The system prioritizes all nodes with in-degree 0 (interfaces without dependencies), gradually removing the nodes and recording their execution order.

[0082] For each node removed, update the in-degree of the adjacent nodes until all nodes in the graph have been processed.

[0083] The generated sorting results are used to construct a business process graph, where each node represents an interface call and the edges represent the order of the calls.

[0084] Automatic process generation and parameter configuration: The system automatically generates a business process diagram based on the sorting results and configures the input and output parameters of each interface.

[0085] The input parameters of each interface are automatically inherited from the output parameters of the previous interface. For example, the output "Initialization Result" of "Camera Initialization" will be automatically configured into the flowchart as the input of the next interface.

[0086] If the input parameters of two or more interfaces are found to be similar, the system automatically merges these parameters to reduce duplicate configuration.

[0087] Parallel task optimization: After generating the flowchart, the system will analyze which interfaces can be executed in parallel. For example, some interfaces have no dependencies and can be executed at the same time. The system will group these interfaces to optimize the process execution time.

[0088] The parallelization optimization strategy is to analyze the nodes without direct connections in the dependency graph and allow them to be executed simultaneously in different threads or processes, thereby improving the execution efficiency of the entire solution.

[0089] Visual interface display: The system provides a graphical interface to display the generated business process diagram through the front end. Users can adjust the nodes in the process by dragging and dropping, and the system will automatically regenerate the dependency diagram and update the topological sorting results.

[0090] In the interface, users can view detailed information of each node, including parameter configuration, dependencies, execution order, etc.

[0091] The system supports users to adjust parameters, display the adjusted flow chart in real time, and dynamically generate new processes.

[0092] Implementation case: Computing orchestration of camera intelligent monitoring system Assume that the user wants to generate a business process for a camera intelligent monitoring system through this system. The monitoring system needs: Initialize the camera; Connect to cloud server; Perform video analysis.

[0093] Through the system's automatic arrangement, the process is as follows: Camera initialization interface: Input: camera ID, camera position; Output: Initialization result.

[0094] Cloud server connection interface: Input: camera ID, server address; Output: Connection status.

[0095] Video analysis interface: Input: camera ID, connection status; Output: Analysis results.

[0096] The system automatically identifies the dependency of the camera IDs between the three interfaces and generates an executable flow chart. The input and output parameters of each interface are automatically configured in the process, and users can view the entire process through a visual interface and make adjustments as needed.

[0097] This implementation demonstrates how to implement complex computing network product orchestration in actual scenarios and generate complete business processes through an automated computing network orchestration system based on parameter dependency algorithms. Through this system, users do not need to manually configure the dependencies and parameters of each interface, which greatly improves the orchestration efficiency.

[0098] The above specific implementations are only descriptions of the preferred implementations of the present invention, and do not limit the protection scope of the present invention. Without departing from the design concept and spirit of the present invention, various modifications, substitutions and improvements made by ordinary technicians in this field to the technical solution of the present invention based on the text description and drawings provided by the present invention should all fall within the protection scope of the present invention. The protection scope of the present invention is determined by the claims.

Claims

1. An automated computing power network orchestration system based on a parameter-dependent algorithm, characterized in that: include: The interface document structured processing module is used to read and extract the interface metadata and parameter metadata in the computing network product interface document, perform structured processing, and provide input data for dependency identification; The parameter vectorization module is used to perform semantic vectorization processing on the extracted metadata and convert the parameters in the document into semantic vectors; A dependency identification module is used to identify the dependency between product interfaces by calculating the similarity between different product interface parameters and to establish a product interface dependency graph; The topological sorting module is used to determine the execution order of product interfaces based on the dependency graph using the topological sorting algorithm and generate a business process diagram with acyclic dependencies; The automatic process generation module is used to automatically generate business processes based on the topological sorting results, and automatically configure the input and output parameters in the process through parameter dependencies to achieve automatic orchestration and customized processing of business processes.

2. According to claim 1, an automated computing power network orchestration system based on a parameter-dependent algorithm is characterized in that: Also includes: The parameter filtering module is used to filter the input parameters in the metadata before parameter vectorization, exclude default parameters that do not change with business scenarios, and historical parameters that are no longer used, thereby improving system performance and ensuring the accuracy of dependencies.

3. According to claim 1, an automated computing power network orchestration system based on a parameter-dependent algorithm is characterized in that: Also includes: The interface dependency library is used to store the dependency information of interfaces of different computing power network products; The interface dependency library dynamically updates and maintains the dependencies between interfaces, providing a reference for subsequent process generation.

4. According to claim 1, an automated computing power network orchestration system based on a parameter-dependent algorithm is characterized in that: Also includes: The parallel task optimization module is used to analyze the product interface dependencies in the process after the business process is generated, identify the product interfaces that can be executed in parallel, and group them for parallel execution.

5. The automatic computing power network orchestration system based on parameter-dependent algorithm according to claim 1 is characterized in that: Also includes: The process visualization interface is used to provide a visual display of business processes, display the generated topology structure and business process diagram, and support users to adjust the product interface in the process and update the business process diagram in real time.

6. The automatic computing power network orchestration system based on parameter-dependent algorithm according to claim 1 is characterized in that: Also includes: The process optimization module is used to optimize the generated business processes and reduce repeated operations by merging similar or redundant parameters.

7. An automated computing power network orchestration method based on a parameter-dependent algorithm, applicable to an automated computing power network orchestration system based on a parameter-dependent algorithm as claimed in any one of claims 1 to 6, characterized in that: include: Step 1: Read the interface document of the computing network product, clean and structure the document, and extract the interface metadata and parameter metadata; Step 2: Use word embedding technology to semantically vectorize the extracted parameters and filter out the parameters; Step 3: Calculate the dependencies between different interfaces based on the vectorized parameters and build a dependency graph; Step 4: Sort the dependency graph using a topological sorting algorithm to generate the correct interface call sequence; Step 5: Generate a business process diagram based on the topological sorting results, identify interfaces that can be executed in parallel, and optimize the process execution order; Step 6: Merge similar parameters in the generated process, remove redundant parameters, and build an optimized parameter tree; Step 7: Optimize the execution efficiency of business processes by reducing redundant operations and parallelizing execution.

8. The method for automatic computing power network orchestration based on parameter-dependent algorithm according to claim 7, characterized in that: Step 1 is as follows: Step 11: Read the interface specification documents related to the computing network product; Step 12: Clean up irrelevant characters in the document; Step 13: Remove invalid lines from the document and extract interface metadata and parameter metadata, including interface name, interface code, product to which the interface belongs, business type, input and output parameters and their attributes; Step 14: Use natural language processing tools to segment the interface description and parameter description to extract descriptive key information; Step 2 is as follows: Step 21: Process the extracted parameter metadata using word embedding technology and use the BERT model to convert the parameter text into a semantic vector; Step 22: Filter out default parameters that do not need to be dynamically changed and invalid historical parameters to reduce performance consumption; Step 23: Concatenate the parameter description information and metadata into a string and input it into the BERT model to generate a feature vector of the parameter for subsequent dependency calculation.

9. The method for automatic computing power network orchestration based on parameter-dependent algorithm according to claim 7, characterized in that: Step 3 is as follows: Step 31: construct the input and output parameter set of the product interface, and extract the input parameters and output parameters of each product interface; Step 32: Calculate the cosine similarity between the output parameters of the interfaces between different products and the product input parameters; Step 33: Determine whether there is a dependency relationship between the interfaces based on the cosine similarity value. If the cosine similarity is greater than a preset threshold, it is considered that the two interfaces have a dependency relationship, and the dependency relationship is recorded. Step 4 is as follows: Step 41: Build a dependency graph of the solution based on the identified interface dependencies; Step 42: Use a topological sorting algorithm to sort the dependency graph to ensure that the starting product interface of each directed edge in the graph is called before the end product interface; Step 43: If a loop, i.e., a circular dependency, is found in the graph, the user is prompted to adjust the topology structure to ensure the correctness of the process.

10. The method for automatic computing power network orchestration based on parameter-dependent algorithm according to claim 7, characterized in that: Step 5 is as follows: Step 51: Generate a business process diagram based on the result of topological sorting to clarify the calling sequence of product interfaces; Step 52: After the flow chart is generated, the system identifies product interfaces that can be executed in parallel, and groups interfaces that have no dependency for parallel execution; Step 53: For interfaces with dependencies, call them in sequence to ensure the logic of process execution; Step 6 is as follows: Step 61: extract the input and output parameters of each interface in the generated flow chart, and construct the parameter tree structure of the input and output parameters; Step 62: according to the identified inter-interface dependency, remove the redundant parameters that have been relied upon in the parameter tree; Step 63: Calculate the semantic similarity of the remaining parameters, merge similar parameters, and process them uniformly in the process to reduce repeated operations; Step 7 is as follows: Step 71: After the flow chart is generated, continue to identify redundant operations in the process to reduce repeated calls; Step 72: Improve the process execution efficiency by optimizing and merging similar parameters and parallelizing the execution strategy.

Images