A project design and optimization method, system, and storage medium

Abstract

This invention provides a project design and optimization method, system, and storage medium. The method includes: acquiring initial project information; cleaning, normalizing, and performing correlation processing on the initial project information to obtain processed project information, wherein the processed project information includes processed historical project information and processed current project information; constructing a knowledge graph by processing the processed historical project information using a natural language processing algorithm and a fusion model; constructing a project performance prediction model using machine learning methods based on the processed historical project information and the knowledge graph; acquiring project requirements; determining constraints based on the project requirements; obtaining candidate projects based on the constraints, the project performance prediction model, and the processed current project information; and determining the final project based on the candidate projects. This application can improve design efficiency and reduce resource consumption.

Description

Technical Field

[0001] This application relates to the technical fields of e-government and low-voltage intelligent projects, and in particular to a project design and optimization method, system and storage medium. Background Technology

[0002] With the development of digitalization, e-government and intelligent low-voltage electrical systems can improve governance efficiency and service levels. E-government, by integrating information resources and optimizing business processes, can achieve one-stop online services, improving the convenience of government services. Intelligent low-voltage electrical systems, by integrating intelligent devices and systems, can realize functions such as security monitoring, energy management, and equipment linkage, ensuring building safety and reducing energy consumption. Scientific planning and scheme optimization can avoid redundant construction, improve system compatibility, ensure that projects meet long-term development needs, and promote the modernization of governance capabilities.

[0003] Currently, the design and optimization of such projects often rely on manual experience and limited data analysis. However, these projects typically involve multi-system integration, cross-departmental collaboration, long-term technology iteration, and massive user demands. Without systematic planning, issues such as functional redundancy, data silos, and insufficient compatibility can easily arise, leading to resource waste and operational difficulties. Therefore, relying solely on manual artifacts and limited data analysis at this stage results in low project design efficiency and increased resource consumption. Summary of the Invention

[0004] To improve design efficiency and reduce resource consumption, embodiments of this application provide a project design and optimization method, system, and storage medium.

[0005] Firstly, a project design and optimization method is provided, the method comprising:

[0006] Initial project information is obtained, and the initial project information is cleaned, normalized, and correlated to obtain processed project information, wherein the processed project information includes processed historical project information and processed current project information;

[0007] A knowledge graph is constructed by processing the processed historical project information using natural language processing algorithms and fusion models.

[0008] Based on the processed historical project information and knowledge graph, a project performance prediction model is constructed using machine learning methods.

[0009] Obtain project requirements, determine constraints based on the project requirements, and obtain candidate projects based on the constraints, the project performance prediction model, and the processed project information.

[0010] The final project is determined based on the candidate projects.

[0011] In some embodiments, the process of cleaning, normalizing, and correlating the initial project information to obtain processed project information includes:

[0012] The initial project information is cleaned by performing missing value processing, outlier identification and correction, and redundant data merging operations to obtain cleaned project information.

[0013] The cleaned project information is subjected to format standardization, encoding mapping, and semantic consistency operations to complete the normalization process and obtain normalized project information;

[0014] The normalized project information is subjected to multi-source data association and business rule association operations to complete the association processing and obtain the processed project information.

[0015] In some embodiments, the process of performing multi-source data association and business rule association operations on the normalized project information to obtain processed project information includes:

[0016] Obtain the data source and rules, and use the data source and rules to construct the row headers and column headers of the table to obtain a blank table;

[0017] The normalized project information is filled into the blank table according to the row and column headings to obtain the processed project information.

[0018] In some embodiments, the process of constructing a knowledge graph by using natural language processing algorithms and fusion models to process the processed historical project information includes:

[0019] Semantic item information is obtained by semantic parsing the processed historical item information using natural semantic algorithms, wherein the semantic item information includes structured semantic item information and unstructured semantic item information;

[0020] Historical constraints are extracted based on the structured and unstructured semantic item information to obtain a quantifiable logical rule base.

[0021] The integrated project information is obtained by processing the logical rule base using a fusion model.

[0022] The integrated project information is converted into a graph structure to obtain a knowledge graph.

[0023] In some embodiments, the performance prediction model includes a base layer and an algorithm layer, and the construction of the project performance prediction model based on the processed historical project information and knowledge graph using machine learning methods includes:

[0024] A foundational layer is constructed based on the processed historical project information and the knowledge graph.

[0025] The base layer is trained using the random forest algorithm to obtain the daily prediction pattern of the algorithm layer in the project performance prediction model;

[0026] The base layer is trained using the LSTM algorithm to obtain the adjustment period prediction pattern of the algorithm layer in the project performance prediction model.

[0027] In some embodiments, determining the constraints based on the project requirements includes:

[0028] Determine whether there is a conflict between the sub-requirements of the project requirements. If not, generate corresponding constraint sub-conditions based on each sub-requirement of the project. The preset conditions include several constraint sub-conditions.

[0029] If they exist, obtain conflicting project sub-requirements and non-conflicting project sub-requirements, and determine whether the correct project sub-requirement in the conflicting project sub-requirements can be determined based on the non-conflicting project sub-requirements. If so, generate corresponding constraint sub-conditions based on the non-conflicting project sub-requirements and the correct project sub-requirements.

[0030] If not, obtain the correction requirement, and generate corresponding constraint sub-conditions based on the non-conflicting project sub-requirement and the correction requirement.

[0031] In some embodiments, the process of obtaining candidate projects based on the constraints, the project performance prediction model, and the processed project information includes:

[0032] Obtain the penalty function, and adjust the crossover and mutation in the preset multi-objective genetic algorithm based on the constraints, penalty function and project performance prediction model to obtain the adjusted multi-objective genetic algorithm;

[0033] The adjusted multi-objective genetic algorithm is used to process the processed project information to obtain candidate projects.

[0034] In some embodiments, determining the final project based on the candidate projects includes:

[0035] The fuzzy comprehensive evaluation method was used to determine the scheme scores of the candidate projects;

[0036] The scheme with the highest score among the scheme scores is determined as the final project.

[0037] Secondly, a project design and optimization system is provided, comprising: a data acquisition module, a preprocessing module, a knowledge graph module, a modeling module, and a solution generation module; wherein,

[0038] The acquisition module is used to obtain initial project information;

[0039] The preprocessing module is used to clean, normalize, and perform correlation processing on the initial project information to obtain processed project information, wherein the processed project information includes processed historical project information and processed current project information.

[0040] The knowledge graph module is used to construct a knowledge graph by processing the processed historical project information using natural language processing algorithms and fusion models;

[0041] The modeling module is used to construct a project performance prediction model based on the processed historical project information and knowledge graph using machine learning methods.

[0042] The solution generation module is used to obtain project requirements, determine constraints based on the project requirements, obtain candidate projects based on the constraints, the project performance prediction model, and the processed project information, and determine the final project based on the candidate projects.

[0043] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon that can run on a processor, wherein when executed by the processor, the computer program implements a project design and optimization method as described in the first aspect.

[0044] By employing the above method, this application first obtains initial project information, then cleans, normalizes, and performs correlation processing on this initial project information to obtain processed project information. This processed project information includes processed historical project information and processed current project information. By performing cleaning, normalization, and correlation processing on the obtained initial project information, standardized and interconnected information can be accurately obtained, facilitating subsequent work and indirectly improving design efficiency.

[0045] Then, natural language processing algorithms and a fusion model are used to process the processed historical project information to construct a knowledge graph. Based on the processed historical project information and the knowledge graph, machine learning methods are used to construct a project performance prediction model. Since project information often comes from multiple sources and may contain conflicts, the fusion model can solve the problems of entity alignment, conflict resolution, and heterogeneous data integration, facilitating the standardization and strong correlation of information. Then, by leveraging the aforementioned strongly correlated and standardized information, algorithm layers for different types of periods are obtained. On the one hand, because the information has undergone the same processing beforehand, the project performance prediction model can be more suitable for processing the processed project information in subsequent iterations. On the other hand, the constructed project performance prediction model has corresponding algorithm layers for different types of periods, thus enabling the project performance prediction model to obtain more accurate output information when processing the processed project information in subsequent iterations.

[0046] The process involves obtaining project requirements, determining constraints based on these requirements, and generating candidate projects based on these constraints, the project performance prediction model, and the processed project information. The final project is then selected from these candidate projects. Since project information often comes from multiple sources and may contain conflicts, a fusion model can address issues such as entity alignment, conflict resolution, and heterogeneous data integration, facilitating information standardization, consistency, and strong correlation. Attached Figure Description

[0047] Figure 1 This is a flowchart of a project design and optimization method provided in an embodiment of this application.

[0048] Figure 2 This is a flowchart illustrating the method provided in this application for cleaning, normalizing, and performing correlation processing on initial project information to obtain processed project information.

[0049] Figure 3 This is a flowchart of a method for constructing a knowledge graph by processing historical project information based on natural semantic algorithms and fusion models, as provided in this application.

[0050] Figure 4 This is a flowchart of a method for obtaining candidate projects based on constraints, a project performance prediction model, and processed project information, as provided in this application.

[0051] Figure 5 This is a schematic diagram of a project design and optimization system connection provided in an embodiment of this application. Detailed Implementation

[0052] To better understand the purpose, technical solutions, and advantages of this application, it has been described and illustrated below with reference to the accompanying drawings and embodiments. However, those skilled in the art should understand that this application can be implemented without these details. It will be apparent to those skilled in the art that various modifications can be made to the embodiments disclosed in this application, and the general principles defined in this application can be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the illustrated embodiments, but is consistent with the broadest scope claimed in this application.

[0053] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0054] Figure 1 This is a block diagram of a project design and optimization method provided in an embodiment of this application. For example... Figure 1 As shown, a project design and optimization method includes the following steps:

[0055] Step S100: Obtain initial project information, and perform cleaning, normalization and correlation processing on the initial project information to obtain processed project information. The processed project information includes processed historical project information and processed current project information.

[0056] This application describes the process from the processing end. The aforementioned initial project information specifically refers to historical project data, industry standards, and environmental data obtained from databases, IoT devices, and industry knowledge bases. This initial project information includes initial historical project information and initial current project information. Initial historical project information specifically refers to historical information within the aforementioned initial project information, while initial current project information specifically refers to information about the current project within the aforementioned initial project information. When cleaning, normalizing, and correlating the initial project information, the same cleaning, normalizing, and correlating processes are performed on both the initial historical project information and the initial current project information. Figure 2 This is a flowchart illustrating the method provided in this application for cleaning, normalizing, and performing correlation processing on initial project information to obtain processed project information. For example... Figure 2 As shown, the process of cleaning, normalizing, and correlating the initial project information to obtain processed project information includes the following steps:

[0057] Step S101 involves performing missing value processing, outlier identification and correction, and redundant data merging operations on the initial project information to complete the cleaning process and obtain the cleaned project information.

[0058] Step S102: Perform format standardization, encoding mapping, and semantic consistency operations on the cleaned project information to complete the normalization process and obtain normalized project information.

[0059] Step S103: Perform multi-source data association and business rule association operations on the normalized project information to complete the association processing and obtain the processed project information.

[0060] Upon receiving initial project information, the processing unit first performs a cleaning process. This involves thoroughly reviewing the initial project information to determine if any data is missing. If no missing data is found, the initial project information is directly identified as the missing value processed version. If missing data is found, the associated database is used first to process the missing values, resulting in the processed information. If the associated database cannot process the missing values, manual processing is then employed. For example, missing fields in basic information such as population or legal entity information can be automatically filled in using associated departmental databases (public security household registration system, business registration system). For information that cannot be automatically filled in (e.g., missing company registered capital), an alert is triggered, and the information is pushed to the relevant business department for manual verification and completion.

[0061] Then, the information obtained after missing value processing, along with the information without missing value processing, undergoes outlier identification and correction. This involves establishing a business rule base, using automated scripts to screen abnormal records for outlier identification, and comparing information across departments for correction.

[0062] Specifically, the aforementioned business rule base is a collection of rules used to determine the legality, rationality, and logical consistency of information. These rules are logical judgment conditions extracted from historically verified real-world business knowledge, policies, regulations, and industry standards. They can be understood as a digital, computer-automated electronic information quality inspection manual. This business rule base is characterized by business orientation, configurability, and centralized management. That is, rules are typically designed as configurable scripts, allowing business personnel or data analysts to modify and extend them without rewriting program code. Furthermore, all rules are managed in a unified library, avoiding rule fragmentation, duplication, or conflict.

[0063] Furthermore, building a business rule base is an iterative process that originates from business needs and extends to technology, including the following steps: First, rule discovery and collection. This involves discovering and collecting rules from existing policies and regulations, business manuals and standards, domain expert opinions, and historical data and projects. Then, the discovered and collected rules are converted into structured rules that computers can understand, which can adopt an IF-THEN structure. Next, the formalized rules are stored in a rule base, which can be a simple database table, an Excel configuration file, or a professional business rule management system. Finally, rule engine scripts are written or existing rule engine tools, such as Drools, are used. By specifying this rule database, the initial cleansing of the acquired project information can be efficiently and accurately guided.

[0064] The processing end uses the aforementioned business rule base as a filter to scan the acquired information one by one and mark the information that does not conform to the business rule base, thereby realizing the identification of outliers by screening abnormal records through automated scripts.

[0065] Furthermore, to ensure the smooth completion of project design and optimization, the information obtained above needs to be corrected after identifying outliers. When specifying a business rule base, cross-departmental information can be incorporated to define various existing correction rules, allowing this application to make corrections by reading the business rule base. This makes the entire process automated, improving the quality and efficiency of information processing.

[0066] Next, the information obtained after outlier identification and correction will be used for redundant data merging. Specifically, the cleaned project information will be obtained by merging redundant data using entity alignment.

[0067] Specifically, entity alignment refers to the process of identifying and pointing to the same entity from information from different sources and in different formats. The same entity may have multiple records in different systems, and these records may be inconsistent. For example, for the same person, in system A, it might be Zhang San (ID number 123…), address Z, Y Road, B District, City A; while in system B, it might be Zhang Lao San (ID number 123…), with social security registration at Company C. The goal is to identify that Zhang San and Zhang Lao San are the same person, thereby eliminating redundancy, breaking down information silos, and achieving entity alignment.

[0068] The cleaned project information is obtained by merging redundant data using entity alignment. First, a coarse screening is performed to group potentially similar records together, and then a fine comparison is performed only internally.

[0069] Then, a selection process is performed, which involves recording the similarity between two records. This includes partial similarity calculation, selecting a similarity algorithm for each piece of information, such as using edit distance or pinyin similarity to calculate the similarity of the name part, or using direct comparison of differences to calculate the similarity of dates and numbers.

[0070] This includes overall similarity calculation, which can be performed using machine learning methods, such as using labeled data on whether they are the same entity as a training set to train a classification model, thereby obtaining the corresponding model, and then using the trained model to perform overall similarity calculation.

[0071] Finally, alignment is performed based on the obtained similarity scores. This involves determining which information needs alignment based on the similarity values, and then merging the information to complete the alignment. This merging is not a simple overwrite, but rather a union calculation. For example, consider two data sources: one from the Public Security Bureau (e.g., name Zhang San, ID number A, address B); and the other from the Social Security Bureau (e.g., name Zhang Laosan, ID number A, insurance unit C). First, these two data sources are grouped together based on the ID number, identifying them as the same entity. Then, entity merging is performed, resulting in the cleaned information: name Zhang San, former name Zhang Laosan, ID number A, address B, insurance unit C. This merging of redundant data yields more closely related information.

[0072] After receiving the cleaned project information, the processing end performs normalization processing on the cleaned project information. The normalization processing includes format standardization, encoding mapping, and semantic consistency operations.

[0073] Specifically, format standardization involves converting information with attributes such as time and units into a unified format. For example, it unifies the heterogeneous time formats of various departments to the ISO 8601 standard. It also performs unit conversions on economic data to eliminate the impact of dimensional differences on subsequent work. Coding mapping maps departmental-defined coding information to a unified coding system, ensuring data comparability. Semantic consistency unifies the names of various departments. For example, it uses a government terminology database to unify the descriptions of enterprises, companies, and legal entities as "market entities," avoiding semantic ambiguity. The processing end obtains normalized project information after normalizing the cleaned project information.

[0074] After receiving the normalized project information, the processing end performs correlation processing on the normalized project information. This correlation processing includes multi-source data correlation and business rule correlation operations.

[0075] Specifically, multi-source data association involves linking information from multiple sources. Business rule association involves linking rules from different departments. The processing end obtains processed project information after performing correlation processing on the normalized project information. Thus, by processing the initial project information through cleaning, normalization, and correlation, accurate and standardized information with interrelationships can be obtained, facilitating subsequent work and indirectly improving design efficiency.

[0076] The process of performing multi-source data association and business rule association operations on normalized project information to obtain processed project information includes: acquiring data sources and rules; using the data sources and rules to construct row headers and column headers of a table to obtain a blank table; and then filling the blank table with normalized project information according to the row headers and column headers of the blank table to obtain the processed project information.

[0077] One approach is to use the data source as the row header and the rules as the column header to create a blank table, or vice versa. Then, based on this blank table, the relevant information is stored in the cells to easily obtain a clear and concise overview of the processed items, facilitating quick and clear identification of highly relevant processed items.

[0078] Step S200: Use natural language processing algorithms and fusion models to process the processed historical project information and construct a knowledge graph.

[0079] Figure 3 This is a flowchart illustrating the method provided in this application for constructing a knowledge graph using natural language processing algorithms and fusion models to process processed historical project information. For example... Figure 3 As shown, the process of constructing a knowledge graph using natural language processing algorithms and fusion models to process historical project information includes the following steps:

[0080] Step S201: Semantic item information is obtained by performing semantic analysis on the processed historical item information using a natural semantic algorithm. The semantic item information includes structured semantic item information and unstructured semantic item information.

[0081] Step S202: Extract historical constraints based on structured and unstructured semantic item information to obtain a quantifiable logical rule base.

[0082] Step S203: Use the fusion model to process the logical rule base to obtain integrated project information.

[0083] Step S204: The integrated project information is converted into a graph structure to obtain a knowledge graph.

[0084] Since the initial project information has already been cleaned, normalized, and correlated, the information unification operation can be simplified when constructing a knowledge graph based on natural semantic algorithms and fusion models, thereby indirectly improving the efficiency of knowledge graph construction.

[0085] Firstly, key elements such as budget amounts, technology selections, and risk assessments can be extracted from processed historical project information, including project objectives and technical routes, using regular expressions and XML parsers to obtain structured semantic project information. Secondly, the BERT model can be used to perform semantic analysis on policy text information from processed historical project information, extracting entity relationships such as legal clauses, applicable scenarios, and responsible parties to obtain unstructured semantic project information. This semantic project information includes both structured and unstructured semantic project information.

[0086] Furthermore, for information other than project objectives, technical routes, and policy texts in processed historical project information, this other information can be further broken down into fixed sub-information with industry-standard formats and non-fixed sub-information without industry-standard formats. Then, on one hand, regular expressions and XML parsers are used to extract key elements from the fixed sub-information to obtain structured semantic item sub-information. On the other hand, the BERT model is used for semantic analysis of the non-fixed sub-information to obtain unstructured semantic item sub-information. Finally, the obtained structured and unstructured semantic item sub-information are combined to form the semantic item information for this part of the other information. This yields more accurate semantic item information, thereby indirectly improving the accuracy of the subsequent knowledge graph.

[0087] The processing end can then use the semantic item information obtained above as input, and use the BERT+Attention model to obtain historical constraints. These historical constraints are stored in the same database to obtain a quantifiable logical rule base. Alternatively, a Bidirectional Long Short-Term Memory Network-Conditional Random Field model (BiLSTM-CRF model) can be used to identify entity information such as project subjects, technical indicators, and implementation cycles from both structured and unstructured semantic item information. Then, graph embedding techniques, such as TransE, are used to obtain historical constraints, which are also stored in the same database to obtain a quantifiable logical rule base. This approach, using natural language processing algorithms to overcome the limitations of information matching, understands text at the semantic level, facilitates more accurate extraction of project entities and their relationships, and indirectly improves the accuracy of project design and optimization.

[0088] Next, the fusion model is used to perform entity alignment and disambiguation operations on the information in the logical rule base obtained above to obtain integrated project information. Since project information often comes from multiple sources and may have conflicts, the fusion model can solve the problems of entity alignment, conflict resolution, and heterogeneous data integration. This facilitates the standardization, neatness, and strong correlation of information, making it easier to efficiently and accurately implement project design and optimization using only a small amount of data. After all, invalid information has been removed, and the remaining information has strong correlations among itself.

[0089] Finally, the integrated project information is stored as a graph structure to obtain a knowledge graph, which facilitates subsequent visualization and complex queries of project design and optimization. This joint processing of structured and unstructured data improves the accuracy of the resulting knowledge graph.

[0090] Step S300: Based on the processed historical project information and knowledge graph, a project performance prediction model is constructed using machine learning methods.

[0091] The performance prediction model consists of a foundation layer and an algorithm layer. The process of constructing the project performance prediction model using machine learning methods based on processed historical project information and a knowledge graph includes the following steps:

[0092] Step S301: Construct a basic layer based on the processed historical project information and knowledge graph.

[0093] Step S302: Train the base layer using the random forest algorithm to obtain the daily prediction pattern of the algorithm layer in the project performance prediction model.

[0094] Step S303: Train the base layer using the LSTM algorithm to obtain the adjustment prediction mode of the algorithm layer in the project performance prediction model.

[0095] The project performance prediction model to be built is trained from an untrained initial learning model. The processing end first deploys processed historical information and a knowledge graph on a data interaction platform, and establishes an indirect connection between this platform and the input side of the initial learning model to obtain the base layer of the project performance prediction model. Then, on one hand, information representing daily life (i.e., information outside of policy adjustment periods) is selected from the base layer and used to train the initial learning model using a random forest algorithm to obtain the algorithm layer for the daily prediction mode of the project performance prediction model. On the other hand, information from policy adjustment periods is selected from the base layer and used to train the initial learning model using an LSTM algorithm to obtain the algorithm layer for the adjustment period prediction mode of the project performance prediction model. Essentially, the project performance prediction model is a model that reflects the mapping relationship between design information and performance indicators. By using the aforementioned strongly correlated and well-organized information to obtain algorithm layers for different types of periods, and because the information has already undergone the same processing, the project performance prediction model can be more suitable for processing the information already processed in the current project. On the other hand, the constructed project performance prediction model has made corresponding algorithm layers for different types of periods, so that the subsequent project performance prediction model can obtain more accurate output information when processing the information of this project.

[0096] Step S400: Obtain project requirements, determine constraints based on project requirements, and obtain candidate projects based on constraints, project performance prediction model, and processed project information.

[0097] To improve user satisfaction with the final project, it's necessary to further understand user requirements for the project and incorporate these requirements into the project design and optimization process. The processing end can passively obtain project requirements through user input or proactively by asking users; no further limitations are placed on the methods used to obtain project requirements here.

[0098] The purpose of obtaining project requirements from the processing end is to determine the constraints encountered during project design and optimization, i.e., to identify the constraints. Determining constraints based on project requirements includes the following steps:

[0099] Step S401: Determine whether there is a conflict between the sub-requirements of the project requirements. If not, generate corresponding constraint sub-conditions based on each sub-requirement of the project. The preset conditions include several constraint sub-conditions.

[0100] Step S402: If they exist, obtain the conflicting project sub-requirements and the non-conflicting project sub-requirements, and determine whether the correct project sub-requirement in the conflicting project sub-requirements can be determined based on the non-conflicting project sub-requirements. If so, generate the corresponding constraint sub-conditions based on the non-conflicting project sub-requirements and the correct project sub-requirements.

[0101] Step S403: If not, obtain the correction requirements and generate corresponding constraint sub-conditions based on the non-conflicting project sub-requirements and correction requirements.

[0102] Each project requirement includes at least one sub-requirement. First, a semantic algorithm is used to determine the meaning of each sub-requirement. Then, the meanings of all sub-requirements are compared. If they differ, there is a conflict between the sub-requirements included in the overall project requirements. If they are the same, there is no conflict.

[0103] When it is determined that there are no conflicts among the sub-requirements included in the project requirements, a constraint sub-condition can be generated directly based on the meaning of a sub-requirement obtained above, so that one sub-requirement corresponds to one constraint sub-condition. The constraint sub-condition specifically represents a mathematical expression.

[0104] When determining if there are conflicts among the sub-requirements of a project, the first step is to compare the meanings of all sub-requirements to identify conflicting sub-requirements and non-conflicting sub-requirements. Then, the meanings of each non-conflicting sub-requirement are combined to obtain a reference meaning. Next, it is determined whether this reference meaning can infer the theoretical meanings of all conflicting sub-requirements. Specifically, the reference meaning determines its primary attribute, and all conflicting sub-requirements determine their secondary attributes. If the primary attribute contains all secondary attributes, then the reference meaning can infer the theoretical meanings of all conflicting sub-requirements. If the primary attribute does not contain all secondary attributes, then the reference meaning cannot infer the theoretical meanings of all conflicting sub-requirements.

[0105] If not, then the correct sub-requirement within a conflicting project cannot be determined based on non-conflicting sub-requirements. If so, then the correct sub-requirement within a conflicting project can be determined based on non-conflicting sub-requirements. Here, the correct sub-requirement represents the information within the conflicting project requirements that correctly reflects the user's needs.

[0106] When it's determined that the correct sub-requirement from the conflicting sub-requirements can be identified based on the non-conflicting sub-requirements, then each non-conflicting sub-requirement and each correct sub-requirement is assigned a corresponding mathematical expression-type constraint sub-condition. Conversely, when it's determined that the correct sub-requirement from the conflicting sub-requirements cannot be identified based on the non-conflicting sub-requirements, the unidentifiable conflicting sub-requirements are sent to the user for feedback and corrections. Then, each non-conflicting sub-requirement and each correction request is assigned a corresponding mathematical expression-type constraint sub-condition. This process allows the processing end to automatically obtain accurate constraints based on existing information, improving the efficiency and accuracy of project design and optimization.

[0107] Subsequently, the project design work was carried out using the constraints obtained above, the project performance prediction model, and the processed project information. Figure 4 This is a flowchart illustrating the method provided in this application for obtaining candidate projects based on constraints, a project performance prediction model, and processed project information. (For example...) Figure 4 As shown, the process of obtaining candidate projects based on constraints, project performance prediction models, and processed project information includes the following steps:

[0108] Step S404: Obtain the penalty function, and adjust the crossover and mutation in the preset multi-objective genetic algorithm based on the constraints, penalty function and project performance prediction model to obtain the adjusted multi-objective genetic algorithm.

[0109] Step S405: Use the adjusted multi-objective genetic algorithm to process the information of the current project to obtain candidate projects.

[0110] Specifically, after determining the constraints, a penalty function can be obtained using pre-defined constraint and penalty relationships. This penalty function includes hard constraint penalty terms and soft constraint penalty terms. Hard constraint penalty terms correspond to higher penalties for violations; they are quadratic penalty functions, meaning the penalty increases quadratically with the degree of violation. Which of the aforementioned constraints corresponds to the hard constraint penalty terms can be set according to the actual situation. The hard constraint penalty term is the sum of the squares of max(0, the degree of violation of the hard constraint).

[0111] Soft constraint penalty terms correspond to lower penalties for violations, with penalties quantitatively determined by the severity of the violation. This is a linear penalty function, meaning the penalty is directly proportional to the degree of violation. Which of the constraints mentioned above correspond to the soft constraint penalty terms can be set according to the actual situation. The constraints corresponding to the soft constraint penalty terms and the hard constraint penalty terms constitute the aforementioned preset conditions. The soft constraint penalty term = the sum of the soft constraint weight coefficients * max(0, the degree of violation of the hard constraint).

[0112] Based on the constraints, penalty functions, and project performance prediction models obtained above, the mapping relationships reflected by these constraints, penalty functions, and project performance prediction models are grouped, and targeted mutation operators are designed to determine the constraint grouping strategy. Then, the crossover rate and mutation rate are dynamically adjusted according to the number of iterations to balance global search and layout optimization, thereby adjusting the crossover and mutation in the existing multi-objective genetic algorithm to obtain an adjusted multi-objective genetic algorithm.

[0113] Next, the adjusted multi-objective genetic algorithm is used to process the processed project information obtained in step S100 above to obtain candidate projects. Each candidate project includes several candidate sub-projects, resulting in at least one project. By incorporating user requirements and corresponding penalty functions into the genetic algorithm, and considering the mapping relationship between design information and performance indicators, the existing genetic algorithm is adjusted to obtain an adjusted multi-objective genetic algorithm. Then, using this adjusted multi-objective genetic algorithm to determine the project to be designed improves the feasibility of the obtained candidate projects, reduces the probability of repetitive project design, and thus indirectly improves work efficiency.

[0114] Step S500: Determine the final project based on the candidate projects.

[0115] After initially obtaining the candidate projects, the final project needs to be selected from them, thus optimizing the project. This process involves: using fuzzy comprehensive evaluation to determine the scheme scores for each candidate project; and identifying the scheme with the highest score as the final project. Specifically, the fuzzy comprehensive evaluation method is used to calculate the scheme sub-score for each candidate sub-project, and all sub-scores constitute the scheme score. Then, all sub-scores are compared, and the scheme with the highest sub-score is selected as the final project. This ensures that the final project better meets user needs, achieving efficient and resource-efficient project selection.

[0116] Preferably, after determining the final project, the process further includes: receiving user feedback information; determining whether the constraint performance prediction model needs to be adjusted based on the user feedback information; if not, generating a termination instruction; if so, cleaning, normalizing, and performing correlation processing on the user feedback information to obtain processed feedback information; and adding the processed feedback information to the processed project information to update the processed project information.

[0117] Specifically, the aforementioned user feedback information represents user feedback on the final project. If the feedback information includes user modifications to the final project, it indicates that the constraint performance prediction model needs adjustment. If the feedback information does not include user modifications to the final project, it indicates that the constraint performance prediction model does not need adjustment.

[0118] When it is determined that no adjustment to the constraint performance prediction model is needed, it indicates that the final project has been approved by the user, the project design and optimization work is completed, and the end instruction is generated.

[0119] When it is determined that the constraint performance prediction model needs adjustment, it indicates that the final project has been approved by the user. Processed feedback information can be obtained by cleaning, normalizing, and correlating the user feedback. This processed feedback information is then added to the processed project information to update the processed project information, and the project is redesigned and optimized. This ensures the accuracy of the project.

[0120] The essence of updating a knowledge graph is to integrate newly generated project information into the existing knowledge graph. Specifically, the first step is to obtain new processed historical project information. The data source for updating is the newly generated processed historical project information, such as the processed information of the current project, or supplementary information provided by the user. The operation in step S100 is performed on the obtained new processed historical project information to ensure that the new processed historical project information is consistent with the format and semantics of the processed historical project information in the existing knowledge graph, avoiding heterogeneous data conflicts.

[0121] Then, semantic parsing and rule base updates are performed. That is, natural language processing algorithms are used to perform semantic parsing on the preprocessed new historical project information to extract new semantic project information. From the new semantic project information, new historical constraints are extracted and added to the original quantifiable logical rule base to complete the incremental update of the rule base, so as to ensure that the rule base covers the latest project constraints.

[0122] Next, the fusion model handles conflicts between new and old data. This involves resolving conflicts between new and existing rules, and aligning new entities with existing entities. For example, if the energy consumption constraint for a new project is no more than 10kW, and the energy consumption constraint for an existing project is no more than 15kW, the fusion model, based on the priorities in the document, retains the new rules and corrects the corresponding entity relationships in the existing rules. For newly extracted semantic entities, the fusion model aligns them with similar entities in the existing graph to avoid entity redundancy.

[0123] Finally, the information output by the fusion model is used to update the existing knowledge graph in accordance with step S204 above. If a new entity needs to be added, a new node is added to the knowledge graph, and the relationship between this node and other related nodes is established, such as establishing a link between a new policy clause and project compliance requirements. If the relationship between existing entities is updated, the relationship attributes of the existing entities are modified. For example, if the energy consumption constraint threshold changes, the relationship between the project and the energy consumption constraint is updated from no more than 15kW to no more than 10kW.

[0124] Figure 5This is a schematic diagram of a project design and optimization system connection provided in an embodiment of this application. For example... Figure 5 As shown, a project design and optimization system includes: a data acquisition module, a preprocessing module, a knowledge graph module, a modeling module, and a solution generation module.

[0125] The system comprises the following modules: A data acquisition module obtains initial project information. A preprocessing module cleans, normalizes, and performs correlation processing on the initial project information to obtain processed project information, including processed historical project information and information from the current project. A knowledge graph module uses natural language processing algorithms and a fusion model to process the processed historical project information and construct a knowledge graph. A modeling module uses machine learning methods to build a project performance prediction model based on the processed historical project information and the knowledge graph. A solution generation module obtains project requirements, determines constraints based on these requirements, generates candidate projects based on the constraints, the project performance prediction model, and the processed current project information, and finally determines the final project based on these candidate projects.

[0126] The other functions performed by the above-mentioned data acquisition module, preprocessing module, knowledge graph module, modeling module, and solution generation module, as well as the technical details of each function, are the same as or similar to the corresponding features in the project design and optimization method described above, so they will not be repeated here.

[0127] This application also provides a computer storage medium storing a computer program that, when run on a computer, enables the computer to execute the steps in the project design and optimization method described above.

[0128] It should be understood that although the steps in the flowcharts in the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps, and they can be performed in other orders.

[0129] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A project design and optimization method, characterized by, The method includes: Initial project information is obtained, and the initial project information is cleaned, normalized, and correlated to obtain processed project information, wherein the processed project information includes processed historical project information and processed current project information; A knowledge graph is constructed by processing the processed historical project information using natural language processing algorithms and fusion models. Based on the processed historical project information and knowledge graph, a project performance prediction model is constructed using machine learning methods. Obtain project requirements, determine constraints based on the project requirements, and obtain candidate projects based on the constraints, the project performance prediction model, and the processed project information. The final project is determined based on the candidate projects; The step of constructing a knowledge graph by processing the processed historical project information using natural language processing algorithms and fusion models includes: Semantic item information is obtained by semantic parsing the processed historical item information using natural semantic algorithms, wherein the semantic item information includes structured semantic item information and unstructured semantic item information; Historical constraints are extracted based on the structured and unstructured semantic item information to obtain a quantifiable logical rule base. The integrated project information is obtained by processing the logical rule base using a fusion model. The integrated project information is converted into a graph structure to obtain a knowledge graph; Among them, the fusion model is used to solve problems of entity alignment, conflict resolution and heterogeneous data integration; The performance prediction model includes a base layer and an algorithm layer. The construction of the project performance prediction model based on the processed historical project information and knowledge graph using machine learning methods includes: A foundational layer is constructed based on the processed historical project information and the knowledge graph. The base layer is trained using the random forest algorithm to obtain the daily prediction pattern of the algorithm layer in the project performance prediction model; The base layer is trained using the LSTM algorithm to obtain the adjustment period prediction pattern of the algorithm layer in the project performance prediction model; During periods when there are no policy adjustments, information representing daily life is selected from the basic layer, and an initial learning model is trained using the random forest algorithm to obtain the algorithm layer of the project performance prediction model in the daily prediction mode. During policy adjustment periods, information about the policy adjustment period is selected from the basic layer, and an initial learning model is trained using the LSTM algorithm to obtain the algorithm layer of the project performance prediction model for the prediction mode during the adjustment period.

2. The method according to claim 1, characterized in that, The process of cleaning, normalizing, and correlating the initial project information to obtain processed project information includes: The initial project information is cleaned by performing missing value processing, outlier identification and correction, and redundant data merging operations to obtain cleaned project information. The cleaned project information is subjected to format standardization, encoding mapping, and semantic consistency operations to complete the normalization process and obtain normalized project information; The normalized project information is subjected to multi-source data association and business rule association operations to complete the association processing and obtain the processed project information.

3. The method according to claim 2, characterized in that, The process of performing multi-source data association and business rule association operations on the normalized project information to obtain processed project information includes: Obtain the data source and rules, and use the data source and rules to construct the row headers and column headers of the table to obtain a blank table; The normalized project information is filled into the blank table according to the row and column headings to obtain the processed project information.

4. The method according to claim 1, characterized in that, The constraints determined based on the project requirements include: Determine whether there is a conflict between the sub-requirements of the project requirements. If not, generate corresponding constraint sub-conditions based on each sub-requirement of the project, wherein the constraint conditions include several constraint sub-conditions. If they exist, obtain conflicting project sub-requirements and non-conflicting project sub-requirements, and determine whether the correct project sub-requirement in the conflicting project sub-requirements can be determined based on the non-conflicting project sub-requirements. If so, generate corresponding constraint sub-conditions based on the non-conflicting project sub-requirements and the correct project sub-requirements. If not, obtain the correction requirement, and generate corresponding constraint sub-conditions based on the non-conflicting project sub-requirement and the correction requirement.

5. The method according to claim 1, characterized in that, The candidate projects obtained based on the constraints, the project performance prediction model, and the processed project information include: Obtain the penalty function, and adjust the crossover and mutation in the preset multi-objective genetic algorithm based on the constraints, penalty function and project performance prediction model to obtain the adjusted multi-objective genetic algorithm; The adjusted multi-objective genetic algorithm is used to process the processed project information to obtain candidate projects.

6. The method according to claim 1, characterized in that, The process of determining the final project based on the candidate projects includes: The fuzzy comprehensive evaluation method was used to determine the scheme scores of the candidate projects; The scheme with the highest score among the scheme scores is determined as the final project.

7. A project design and optimization system, characterized in that, The system includes: a data acquisition module, a preprocessing module, a knowledge graph module, a modeling module, and a solution generation module; wherein, The acquisition module is used to obtain initial project information; The preprocessing module is used to clean, normalize, and perform correlation processing on the initial project information to obtain processed project information, wherein the processed project information includes processed historical project information and processed current project information. The knowledge graph module is used to construct a knowledge graph by processing the processed historical project information using natural language processing algorithms and fusion models; The modeling module is used to construct a project performance prediction model based on the processed historical project information and knowledge graph using machine learning methods. The solution generation module is used to obtain project requirements, determine constraints based on the project requirements, obtain candidate projects based on the constraints, the project performance prediction model and the processed project information, and determine the final project based on the candidate projects. The step of constructing a knowledge graph by processing the processed historical project information using natural language processing algorithms and fusion models includes: Semantic item information is obtained by semantic parsing the processed historical item information using natural semantic algorithms, wherein the semantic item information includes structured semantic item information and unstructured semantic item information; Historical constraints are extracted based on the structured and unstructured semantic item information to obtain a quantifiable logical rule base. The integrated project information is obtained by processing the logical rule base using a fusion model. The integrated project information is converted into a graph structure to obtain a knowledge graph; Among them, the fusion model is used to solve problems of entity alignment, conflict resolution and heterogeneous data integration; The performance prediction model includes a base layer and an algorithm layer. The construction of the project performance prediction model based on the processed historical project information and knowledge graph using machine learning methods includes: A foundational layer is constructed based on the processed historical project information and the knowledge graph. The base layer is trained using the random forest algorithm to obtain the daily prediction pattern of the algorithm layer in the project performance prediction model; The base layer is trained using the LSTM algorithm to obtain the adjustment period prediction pattern of the algorithm layer in the project performance prediction model; During periods when there are no policy adjustments, information representing daily life is selected from the basic layer, and an initial learning model is trained using the random forest algorithm to obtain the algorithm layer of the project performance prediction model in the daily prediction mode. During policy adjustment periods, information about the policy adjustment period is selected from the basic layer, and an initial learning model is trained using the LSTM algorithm to obtain the algorithm layer of the project performance prediction model for the prediction mode during the adjustment period.

8. A computer-readable storage medium having a computer program stored thereon that can run on a processor, characterized in that, When the computer program is executed by the processor, it implements a project design and optimization method as described in any one of claims 1 to 6.

Images