Distribution network engineering mechanized construction quota optimization method and system based on artificial intelligence

Abstract

The invention discloses a distribution network engineering mechanization construction quota optimization method and system based on artificial intelligence, and relates to the technical field of quota optimization, and the method comprises the following steps: constructing a multi-target construction collaborative optimization model, solving and generating a quota detection optimization scheme, the solution is to obtain an initial solution of a simulated annealing algorithm based on a genetic algorithm and optimize and screen; obtaining quota detection data based on a quota detection optimization scheme to construct a quota prediction model, and performing quota anomaly judgment based on a dynamic threshold value which is obtained based on a variational auto-encoder; the root cause features of the abnormal result are recognized, the contribution degree of the root cause features is analyzed in combination with an SHAP method, and the root cause features are recognized through a causal reasoning model; according to the invention, through dynamic detection and intelligent optimization of the mechanical construction quota, the problems of low fixed threshold determination sensitivity and incapability of adapting to construction dynamic change in quota detection, and weak abnormal traceability and lack of interpretable support in the quota optimization process in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanized construction quota optimization, and more specifically, to an artificial intelligence-based method and system for optimizing mechanized construction quotas for distribution network projects. Background Art

[0002] Currently, distribution network construction is gradually transitioning toward mechanization and intelligentization. This is particularly true when developing and optimizing power construction quotas, which face multiple challenges, including the complexity of construction tasks, diverse resource types, and increasing carbon emission requirements. Traditional construction quotas, primarily based on historical experience and manual qualitative analysis, lack the ability to deeply mine multi-source data and coordinate multi-objective optimization, making them inadequate for the current demands of efficient, green, and intelligent distribution network construction.

[0003] For example, the invention patent publication number CN119272977A discloses a method for public building energy consumption quotas based on multi-level classification and dynamic data correction, including: obtaining energy consumption parameters, constructing a multi-feature data set and preprocessing it; performing primary classification based on functionality and energy consumption attributes; performing secondary classification based on the K-means model; performing energy consumption quotas based on the quota level and ranking coupling method and the simulation quota method, and performing weighted average to obtain a comprehensive energy consumption quota; performing energy consumption feature importance analysis based on the SHAP value model; performing linear regression based on annual data of important features, and calculating the predicted values ​​and change rates of important features to correct and derive the comprehensive energy consumption quotas for various types of buildings in the future; judging the rationality of energy use and the initiative to save energy based on the change rate of important features of a single building, and correcting the comprehensive energy consumption quota. The present invention corrects two energy consumption quotas for multi-level classified buildings based on dynamic data, improving the previous energy consumption quota method that ignores the impact of different behaviors, properties, and characteristics between buildings on building energy consumption.

[0004] For example, the invention patent announcement with announcement number: CN118504833A discloses a configuration quota method and system based on configuration relationship splitting, including: iteratively disassembling the target disassembly set according to the first logical condition of the current iteration to obtain a number of disassembly subsets, until all disassembly subsets do not meet the second logical condition of the current iteration, merging the disassembly subsets of the current iteration into the disassembly output result, and outputting the merged disassembly output result; adjusting the quota of the target configuration block according to the output disassembly output result. The present invention solves the problem of inaccurate configuration quota results caused by repeated quotas due to complex configurations in the prior art during configuration solution. The present invention greatly improves the accuracy of configuration quotas.

[0005] The above disclosed technical solutions have at least the following technical problems: With the widespread application of mechanized construction in distribution network projects, the existing construction quota system has exposed many problems. Traditional construction quotas are mainly based on historical experience and manual qualitative formulation. They lack in-depth mining of multi-source data and multi-objective collaborative optimization, making it difficult to adapt to the current demand for efficient and intelligent construction of distribution network construction. On the one hand, existing quota optimization methods are mostly centered on the single objective of minimizing cost or construction period, ignoring the coordination relationship between multi-dimensional objectives, resulting in the lack of globalization and practicality of the optimization results. On the other hand, there is a large amount of dynamically changing construction data during the construction process. The traditional static quota formulation method cannot capture the actual situation such as resource intensity fluctuations and construction sequence adjustments during task execution in real time, which in turn limits the scientific nature and responsiveness of the quota.

[0006] Furthermore, existing methods for detecting anomalies in quotas are often based on fixed thresholds and rule matching, making them difficult to adapt to the nonlinear changes in tasks and resource inputs during different construction phases. This makes them prone to missed or misjudgment, impacting quota reliability. After anomaly determinations are made, systematic root cause analysis methods are often lacking, making it impossible to identify the source of the anomaly at the causal level. This also makes it difficult to provide transparent and explainable decision support for quota optimization, further limiting the intelligent level of quota management. To address these issues, the present invention proposes a solution. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a method and system for optimizing the mechanized construction quota of distribution network engineering based on artificial intelligence. By introducing multi-agent optimization, deep generative modeling, causal reasoning and feature contribution analysis technology, dynamic detection and intelligent optimization of the mechanized construction quota of distribution network engineering are realized, solving the problems of low sensitivity of fixed threshold judgment in quota detection, inability to adapt to dynamic changes in construction, weak abnormality tracing capability in prior art, and lack of explainable support for the quota optimization process.

[0008] To achieve the above object, the present invention provides the following technical solutions: An artificial intelligence-based method for optimizing the construction quota of mechanized distribution network projects includes the following steps: constructing a multi-objective construction collaborative optimization model, solving and generating a quota detection optimization plan, wherein the solution is based on a genetic algorithm to obtain an initial solution of a simulated annealing algorithm and optimize the screening; obtaining quota detection data based on the quota detection optimization plan to construct a quota prediction model, and making quota anomaly judgments based on a dynamic threshold, wherein the dynamic threshold is obtained based on a variational autoencoder; identifying the root cause features of the abnormal results, and analyzing the contribution of the root cause features in combination with the SHAP method, wherein the root cause features are identified through a causal inference model.

[0009] In a preferred embodiment, a multi-objective construction collaborative optimization model is constructed, specifically: the first data in each task is obtained to construct a construction task set; the second data in each type of resource is obtained to construct a resource set; and an objective function and allocation constraints are established based on the construction task set and the resource set, wherein the objective function includes carbon emission intensity, maximizing resource utilization, and task priority objective functions.

[0010] In a preferred embodiment, the initial solution of the simulated annealing algorithm is obtained based on the genetic algorithm, specifically: an initial scheduling scheme code is generated based on the construction task set and the resource set, and the initial scheduling scheme code includes a task sequence and a resource allocation matrix; the objective function is respectively based on the entropy weight method to obtain the information entropy of each objective function; a multi-objective fitness evaluation model is constructed based on the linear weighted summation method according to the information entropy, and the fitness is output; the initial scheduling scheme code is based on a strategy combining elite retention and tournament selection method, and a selection operation is performed through fitness; crossover and mutation operations are performed on the tournament selection results to generate new individuals, and the complete next generation is formed in combination with the elite retention results, the crossover operation includes performing order-preserving recombination crossover on the task sequence and performing partial matching crossover on the resource allocation matrix, and the mutation operation is performed by perturbing the task execution time and resource matching relationship; when the fitness is continuously iterated and reaches the preset algebraic termination, the current optimal solution is output as the initial solution of the simulated annealing algorithm.

[0011] In a preferred embodiment, crossover and mutation operations are performed on the tournament selection results to generate new individuals, and the elite retention results are combined to form a complete next generation, specifically: based on the output of the multi-objective fitness evaluation model, each individual in the initial scheduling plan code is sorted in descending order according to fitness, and a preset number of optimal individuals are selected to form an elite subset; the elite subset is copied into the next generation without participating in crossover and mutation operations; the probability of the individual with the best fitness is obtained based on descending sorting, and individuals for the tournament selection method are selected from the individuals remaining in the elite retention; the individuals selected in the tournament are sorted according to fitness, and the optimal individual is selected as the parent generation based on the probability; crossover and mutation operations are performed on the selected parent generation to generate new individuals.

[0012] In a preferred embodiment, the initial solution of the simulated annealing algorithm is obtained and optimized and screened, specifically: the initial solution, initial temperature, temperature update formula and temperature threshold are set, and the output value of the multi-objective fitness evaluation model is used as the objective function value for iteration; the current temperature in each round of iteration is obtained, and based on the current temperature, a neighborhood perturbation operation is performed on the current solution to generate a new candidate solution, wherein the neighborhood perturbation operation includes task sequence perturbation and resource allocation perturbation; judgment is made by calculating the objective function value of the candidate solution; after each round of iteration is completed, the temperature is lowered based on the temperature update formula, and when the temperature threshold is reached, the iteration is stopped, and the globally optimal new solution is output as the quota detection optimization solution.

[0013] In a preferred embodiment, quota anomaly judgment is performed based on a dynamic threshold, specifically: based on a variational autoencoder, the outputs of several quota prediction models are input into the encoder as original samples, and combined with the decoder output to reconstruct samples; a reconstruction error is obtained based on the original samples and the reconstructed samples; the KL divergence of the potential distribution is obtained, and an initial judgment model is constructed in combination with the reconstruction error; based on the Bayesian optimization method, the optimal hyperparameter combination is searched to optimize the initial judgment model to obtain a quota anomaly judgment model; a quota anomaly judgment value is obtained based on the quota anomaly judgment model; based on the sliding window method, several quota anomaly judgment values ​​are used as calculation windows; based on a statistical method, the mean and standard deviation of the quota anomaly judgment values ​​in the window are calculated, and the upper and lower thresholds are set as the dynamic threshold range; quota anomaly judgment is performed within the dynamic threshold range.

[0014] In a preferred embodiment, the root cause characteristics of the abnormal results are identified, and the contribution of the root cause characteristics is analyzed in combination with the SHAP method. The root cause characteristics are identified through a causal reasoning model. Specifically, based on the abnormal judgment result, a causal reasoning model is constructed, and the causal relationship between the quota detection data and the historical construction data is analyzed through the causal reasoning model to identify the key factor characteristics that affect the quota abnormality. The causal reasoning model is constructed based on a graph structure modeling method; the SHAP method is introduced to quantify the key factor characteristics in the causal reasoning model, and the contribution of each feature to the quota abnormality judgment result is calculated; the feature importance is sorted according to the contribution size, and interpretability optimization is performed based on the sorting result.

[0015] In a preferred embodiment, the quota detection data includes resource intensity deviation data, and the specific acquisition method is as follows: the construction period is divided into several stages according to the changing trend of resource input through a change point detection algorithm, and the change point detection algorithm is based on a time series segmentation model of the minimum description length criterion; the resource intensity data under each stage is obtained, and the standardized resource intensity is calculated based on the cost weight normalization method; a standardized resource intensity curve is constructed based on the standardized resource intensity, and the standardized resource intensity curve is fitted using the kernel regression method; the actual resource intensity is obtained, and the resource intensity deviation data is obtained by calculating the average absolute deviation between the actual resource intensity and the fitted standardized resource intensity curve In a preferred embodiment, the quota detection data includes sub-item quota anomaly data, and the specific acquisition steps are as follows: obtain the quota data of several sub-items and construct a multi-dimensional feature vector set; establish a domain relationship diagram between sub-items for the multi-dimensional feature vector set based on a density clustering algorithm, in which nodes in the domain relationship diagram represent sub-items and edges represent adjacent relationships between two nodes, and the adjacent relationships are obtained by obtaining the similarity between nodes using the Mahalanobis distance method; based on the LOF method, calculate the local anomaly factor of each sub-item according to the domain relationship diagram; obtain all local anomaly factors that are greater than a preset local anomaly judgment threshold, and quantify the sub-item quota anomaly data according to the proportion of the local anomaly factors. One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By integrating genetic algorithm and simulated annealing algorithm, a multi-objective construction collaborative optimization model is constructed for minimizing carbon emission intensity, maximizing resource utilization and optimizing task priority. It can efficiently generate high-quality, environmentally friendly and logically rigorous quota inspection optimization plans; global search and population diversity maintenance are achieved through genetic algorithm, and local search capability and the ability to escape local optimality are enhanced through simulated annealing algorithm, so as to achieve balanced exploration of solution space and stable improvement of solution quality. Ultimately, it provides scientific, systematic, dynamic and intelligent optimization support for the acquisition of quota inspection data in complex construction scenarios, significantly improving plan reliability and comprehensive construction benefits.

[0016] 2. By integrating three types of quota detection data, namely resource intensity deviation, sub-item quota anomalies, and corresponding residuals, a multi-source feature fusion model is constructed to comprehensively characterize the abnormal characteristics of the construction process; a variational autoencoder is introduced to improve the recognition ability of nonlinear and sparse anomalies, and combined with a sliding window and dynamic threshold strategy, the model's adaptability to construction fluctuations is enhanced, effectively reducing false alarms and missed alarms; at the same time, causal reasoning and the SHAP method are used to conduct interpretable analysis and root cause location of the detection results, significantly improving the system's practicality and decision-making support capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of the method for optimizing the mechanized construction quota of distribution network projects based on artificial intelligence provided in an embodiment of the present application.

[0018] Figure 2 Schematic diagram of the structure of the artificial intelligence-based distribution network engineering mechanized construction quota optimization system provided in the embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Example 1, Figure 1 The flowchart of the method for optimizing the mechanized construction quota of a distribution network project based on artificial intelligence provided in the embodiment of the present application includes the following steps: S1, build a multi-objective construction collaborative optimization model, solve and generate a quota inspection optimization plan, and the solution is based on the genetic algorithm to obtain the initial solution of the simulated annealing algorithm and optimize the screening.

[0021] In this embodiment, the multi-objective collaborative optimization strategy that integrates the genetic algorithm and the simulated annealing algorithm makes the quota detection optimization scheme highly feasible, highly accurate and highly adaptable, providing a more scientific, systematic, dynamic and intelligent optimization support system for the acquisition of quota detection data.

[0022] Through a multi-objective collaborative optimization model, carbon emission intensity, resource utilization, and task priority are integrated into a unified evaluation system, ensuring that optimization results are more aligned with actual construction efficiency goals and effectively avoiding the bias caused by single-objective optimization in traditional quota-based testing solutions. The benefits include: more environmentally friendly quota data, a more realistic reflection of resource utilization efficiency, and optimized sequencing to prioritize critical paths and improve plan reliability.

[0023] A high-quality initial feasible solution is generated through a genetic algorithm, and excellent individuals are retained using the elite retention + tournament selection mechanism to avoid solution space degradation. The local search capability is then enhanced through order-preserving recombination crossover + resource perturbation mutation, so that: the initial quota detection scheme has a high optimization level and representativeness, avoids premature convergence, accelerates the global convergence speed, and improves the balance and diversity of multi-objective solution space exploration.

[0024] The simulated annealing algorithm performs perturbation-based local search on the basis of the genetic algorithm, introduces the "acceptance of differential solutions" mechanism in the iteration, and effectively escapes the local optimum through the annealing process: further optimizing the initial solution, improving the stability and global optimality of the quota inspection scheme, and supporting fine-tuning and refined analysis of the scheme in complex construction scenarios.

[0025] The multi-objective construction collaborative optimization model is constructed as follows: Acquire first data in each task to construct a construction task set, wherein the first data includes resource demand information, task duration, task priority, and carbon emissions; Acquire second data in each type of resource to construct a resource set, wherein the second data includes resource availability time, maximum available amount of the resource, and usage amount of the resource; Establishing an objective function between resource use and tasks and allocation constraints based on the construction task set and resource set, wherein the objective function includes carbon emission intensity, maximizing resource utilization and task priority, and the allocation constraints include time window constraints, resource capacity constraints and task priority constraints; Among them, the objective function of carbon emission intensity is specifically:

[0026] Where, To minimize the carbon emission intensity objective function, For the The carbon emissions of each task, For the task The start time, For the task The end time, is the total number of tasks; The objective function of maximizing resource utilization is:

[0027] Where, To maximize the resource utilization objective function, is the resource availability time, For construction resources, For the task The amount of resources allocated, is the total construction resources; The objective function of task priority is:

[0028] Where, To minimize the objective function of task priority, For the task priority; The allocation constraints include time window constraints, resource capacity constraints, and task priority constraints, as follows: The time window constraints are specifically:

[0029] Where, is the selected time window value, is the minimum value of the time window, is the maximum value of the time window; The resource capacity constraints are specifically:

[0030] Where, is the usage of construction resources in the selected time window, is the maximum available amount of construction resources, is the construction resource within an arbitrarily selected time window; The task priority constraints are specifically:

[0031] In the formula, For the Whether the construction task is carried out within the selected time window, Any time for construction tasks, For the The subsequent construction tasks of the construction tasks, For the Whether the subsequent construction tasks are carried out within the selected time window.

[0032] It should be noted that the resource capacity constraint limits the total amount of resources used in the construction process within the selected time window to not exceed the maximum available resources. The task priority constraint ensures that the order of construction tasks must meet the construction logic, that is, the predecessor task must be completed before the successor task. The task execution constraint requires that each construction task must and can only be executed once within the selected time window to avoid repeated execution or omission of tasks. ,like =1, indicating that the construction task is carried out within the selected time window; if =0, indicating that the construction task was not carried out during the preset time period.

[0033] The initial solution of the simulated annealing algorithm based on the genetic algorithm is specifically: Generate an initial scheduling solution code based on the construction task set and the resource set, the code including a task sequence and a resource allocation matrix; The objective functions are respectively subjected to the entropy weight method to obtain the information entropy of each objective function; Constructing a multi-objective fitness evaluation model based on the information entropy and a linear weighted summation method, and outputting the fitness; The initial scheduling scheme is encoded based on a strategy combining elite retention and tournament selection. A selection operation is performed based on fitness. Crossover and mutation operations are performed on the tournament selection results to generate new individuals. These individuals are then combined with the elite retention results to form the next generation. The crossover operation includes performing order-preserving reorganization crossover on the task sequence and partial matching crossover on the resource allocation matrix. The mutation operation is performed by perturbing the relationship between task execution time and resource matching, enhancing local search capabilities. When the fitness continuously iterates and reaches the preset algebraic termination, the current optimal solution is output as the initial solution of the simulated annealing algorithm.

[0034] The multi-objective fitness evaluation model is specifically:

[0035] Where, For fitness, To minimize the carbon emission intensity objective function, To maximize the resource utilization objective function, To minimize the objective function of task priority, , , are weight coefficients respectively.

[0036] It should be noted that a task sequence is a one-dimensional permutation list formed by arranging all tasks in a construction task set in a certain order. It represents the order in which tasks are executed. Each element represents a specific construction task. The order of the sequence affects the scheduling start and end times of tasks, as well as the resource utilization rhythm and overall project duration. Tasks may have precedence constraints (e.g., some tasks must begin after others have completed). These dependencies must be considered during scheduling. During crossover and mutation, the task sequence must ensure that each task appears only once. The resource allocation matrix is ​​a two-dimensional matrix structure that specifies specific resource allocation decisions for each construction task based on a given task sequence. It describes the matching and utilization of resources during the execution of each task. During crossover (partial matching crossover), resource allocation can be locally adjusted to explore more optimal resource utilization plans. During mutation, perturbations (such as small increases / decreases in allocation quantities or changes in resource types) can be used to enhance the diversity of the search space and avoid being trapped in local optima.

[0037] The encoding of the initial scheduling scheme is based on a strategy combining elite retention and tournament selection, performing selection operations through fitness, and performing crossover and mutation operations on the tournament selection results to generate new individuals, specifically: Based on the output of the multi-objective fitness evaluation model, each individual in the initial scheduling scheme code is sorted in descending order according to fitness, and a preset number of optimal individuals are selected to form an elite subset; The elite subset is directly copied into the next generation without participating in crossover and mutation operations, preventing excellent solutions from being destroyed by genetic operations and ensuring that the optimal solution of the population is monotonically non-inferior. Obtain the probability of the individual with the best fitness based on descending sorting, and select the individual for the tournament selection method from the remaining individuals of the elite retention selection; Sort the individuals selected by the tournament by fitness and select the best individual as the parent according to the probability; Perform crossover and mutation operations on the selected parents to generate new individuals.

[0038] The initial solution of the simulated annealing algorithm is obtained and optimized and screened, specifically: Set the initial solution, initial temperature, temperature update formula and temperature threshold, and use the output value of the multi-objective fitness evaluation model as the objective function value for iteration; Obtain the current temperature in each iteration. Based on the current temperature, perform a neighborhood perturbation operation on the current solution. The neighborhood perturbation operation includes task order perturbation and resource allocation perturbation. The task order perturbation is to randomly swap the positions of two tasks in the scheduling order while satisfying the task priority constraint. The resource allocation perturbation is to select a task and adjust its corresponding resource configuration while satisfying the resource capacity constraint and time window availability. Generate new candidate solutions based on neighborhood perturbation operations and calculate the objective function value of the candidate solutions for judgment; If the objective function value of the candidate solution is higher than the current solution, the new solution is accepted; If the objective function value of the candidate solution is lower than the current solution, it is accepted in the form of probability; The probability form is accepted, and the specific calculation formula is as follows:

[0039] Where, is the probability of acceptance, is the objective function value difference, is the current temperature; After each round of iteration, the temperature is lowered based on the temperature update formula until the temperature threshold is reached, and the iteration is stopped. Finally, the global optimal new solution is output as the quota detection optimization solution.

[0040] The temperature update formula is specifically calculated as follows:

[0041] Where, is the current temperature, is the temperature coefficient.

[0042] It's important to note that the quota inspection optimization plan reflects the optimal combination of tasks and resource allocation, avoiding resource conflicts and idleness, and provides the optimal quota consumption and usage intensity for different resource and task categories. In short, the quota inspection optimization plan reflects the resource input, task arrangement, and energy consumption patterns under ideal construction conditions, and is the best mapping of theoretical quota standards to actual construction organization.

[0043] S2, based on the quota detection optimization plan, obtains quota detection data to build a quota prediction model, and makes quota anomaly judgment based on a dynamic threshold. The dynamic threshold is obtained based on a variational autoencoder. The quota detection data includes resource intensity deviation data, sub-item quota anomaly data and corresponding residual data.

[0044] In this embodiment, traditional quota detection relies on manual experience or static standards, which can easily become disconnected from specific construction scenarios. The quota detection optimization scheme is generated by considering factors such as construction task coordination constraints, resource conflicts, and time windows. The quota detection optimization scheme provides a unified and structured data foundation for obtaining quota detection data and has the characteristics of strong scenario adaptability. The quota detection optimization scheme includes multiple information such as task execution order, resource intensity, and carbon emissions. Combined with quota detection data, it can more comprehensively construct deep feature relationships and enhance the generalization ability of the model.

[0045] The quota detection data includes resource intensity deviation data, sub-item quota abnormal data and corresponding residual data.

[0046] Furthermore, resource intensity deviation data is used to measure the degree of fluctuation in the input of construction resources (manpower, materials, and machinery) at different stages of construction per unit of work. Predicting quota anomalies by analyzing resource intensity deviation data has the following advantages: Improve the sensitivity and accuracy of anomaly detection: Resource intensity deviation data reflects the real-time fluctuations in resource input per unit of construction during the construction process. It can keenly capture intensity mutations caused by abnormal inputs (such as excessive use of manpower or equipment, material waste, etc.), thereby effectively assisting in identifying anomalies in quota usage (such as overestimation, underestimation or missetting of quota values).

[0047] Enhanced timeliness and early warning: Resource intensity deviation data has distinct time-series characteristics and can be continuously generated and dynamically updated during the project execution phase. By setting thresholds or using anomaly pattern recognition algorithms, deviation trends can be detected early or mid-construction, providing decision support for quota adjustments or on-site corrections, reducing subsequent rectification costs.

[0048] Improve the level of refinement in quota management: By analyzing the deviation in resource intensity at different stages (such as foundation construction, structural construction, and decorative construction), it is possible to identify which types of work, processes, or sections have unreasonable resource allocation or low execution efficiency, thereby providing quantitative support for local optimization of the quota database and promoting the transformation of quota management from experience-driven to data-driven.

[0049] The resource intensity deviation data is obtained in the following specific manner: The construction period is divided into several stages according to the changing trend of resource input using a change point detection algorithm based on a time series segmentation model with a minimum description length criterion. The changing trend of resource input refers to the fluctuation characteristics of the input of various resources (labor, materials, machinery) over time during the construction process. Obtain resource intensity data for each stage and calculate standardized resource intensity based on cost weight normalization. The resource intensity data includes labor input, material input, and machinery input. Based on the standardized resource intensity, a standardized resource intensity curve is constructed and fitted using the kernel regression method. The actual resource intensity is obtained, and the resource intensity deviation data is obtained by calculating the average absolute deviation between the actual resource intensity and the fitted standardized resource intensity curve.

[0050] The specific calculation formula for the standardized resource intensity is as follows:

[0051] The specific calculation formula for the fitting is as follows:

[0052] The specific calculation formula for the resource intensity deviation data is as follows:

[0053] In the formula, To standardize resource intensity, For the The manpower input per unit of work in each stage is For the The mechanical input per unit of engineering volume in each stage is: For the Material input per unit of work in each stage, is the market average cost per unit of labor, is the average market cost per unit of machinery, is the average market cost per unit of material, 、 、 is the resource importance weighting coefficient, For the moment The kernel fitted estimate of resource intensity, For the Each stage represents a point in time. is the number of divided construction stages, The bandwidth is The kernel function, is the bandwidth of the kernel function, Deviation data for resource intensity.

[0054] Sub-item quota anomaly data is used to measure whether the quota values ​​of several sub-items in an engineering project deviate intensively in a certain local concentrated area, and to assist in identifying situations where local anomalies are not identified by overall detection. By analyzing sub-item quota anomaly data to predict quota anomalies, it is possible to reveal "locally concentrated" quota anomalies and supplement the overall detection blind spots: overall quota deviation analysis tends to smooth local errors and easily ignores highly concentrated anomalies in a small area. Sub-item quota anomaly data, by evaluating the density of deviations from the quota values ​​of multiple adjacent or similar sub-items, can reveal, for example: a certain type of work is generally overestimated in a local area; the labor / material quotas of a certain type of structural component are systematically set too low; if such "local systematic deviations" are not identified, the stability and adaptability of the quota standards will be reduced.

[0055] The specific steps for obtaining the abnormal data of the itemized quota are as follows: Obtain quota data for several sub-projects and construct a multi-dimensional feature vector set, wherein the quota data includes unit quota value, labor ratio, material ratio, and construction period; Establishing a domain relationship graph between sub-items for the multidimensional feature vector set based on a density clustering algorithm, wherein nodes in the domain relationship graph represent sub-items, and edges represent adjacent relationships between two nodes. The adjacent relationship is the similarity between nodes obtained based on the Mahalanobis distance method. If the similarity is less than a preset similarity threshold, the nodes are adjacent; Based on the LOF method, the local anomaly factor of each item is calculated according to the domain relationship graph; All local anomaly factors greater than a preset local anomaly determination threshold are obtained, and the sub-item quota anomaly data is quantified according to the proportion of the local anomaly factors.

[0056] The specific calculation formula of the local anomaly factor is as follows:

[0057] The specific calculation formula for the abnormal data of the itemized quota is as follows:

[0058] Where, For the The local anomaly factor of each item, For the The nearest neighbor of each item, is the number of neighbors, For neighbors The local reachable density of For the The local reachability density of each item, For abnormal data of sub-item quota, is the total number of sub-items, is the indicator function, is the preset local anomaly determination threshold.

[0059] It should be noted that in the indicator function, if > , then it is 1, otherwise it is 0.

[0060] Correspondence residual data is used to measure the degree of mismatch between current construction quotas and similar historical projects. It is particularly suitable for detecting unreasonable new quotas set by humans. Analyzing correspondence residual data to predict quota anomalies has the following advantages: Effectively identifies artificially set unreasonable new quotas: Correspondence residuals essentially represent the difference between the current quota value and similar historical projects using the same process, resource mix, and quantity characteristics. When the residual value is significantly greater than the historical fluctuation range, it often indicates: the quota setting process lacks a basis, deviates from actual consumption patterns, or ignores existing empirical data without sufficient reference to historical quota levels. Such anomalies are typically difficult to detect through deviation testing during the construction phase, but correspondence residuals can provide early warning at the initial stage of quota entry.

[0061] Revealing consistency and integrity deficiencies within the quota system: Within a large-scale quota data system, different types of work, components, or construction phases should exhibit a certain degree of mathematical and logical consistency. If certain quota residuals are systematically high or low, this may indicate: an imbalance between the sub-item quota and the parent item quota; a disconnect between some quotas and the actual resource pricing system; conflicts in standard updates; database merging errors; and other technical issues. Residual analysis can serve as part of a systematic verification mechanism to improve the consistency and integrity of the quota data system.

[0062] The specific method for obtaining the corresponding residual data is as follows: Obtain engineering data of the current construction project and construct an engineering feature vector, wherein the engineering data includes engineering quantity, structure type, region, construction period, project, and construction method; The comprehensive distance between the current construction project and all projects in the historical database is obtained based on the Mahalanobis distance method through the project feature vector, and the historical project with the smallest distance is selected as the reference set; Obtain resource quota data for the current construction project and corresponding historical resource quota data in a reference set, wherein the resource quota data includes labor quota, material quota, machinery quota, and management quota; Calculate resource deviations between resource quota data and historical resource quota data; Calculate the corresponding residual mean and residual standard deviation based on resource deviation; The standardized residual coefficient is calculated by the residual mean and residual standard deviation based on the standardized Z-score method; Based on the principal component residual fusion model, the corresponding residual data are calculated according to the standardized residual coefficients.

[0063] The specific calculation formula for the resource deviation is as follows:

[0064] The specific calculation formulas for the residual mean and residual standard deviation are as follows:

[0065]

[0066] The specific calculation formula of the standardized residual coefficient is as follows:

[0067] The specific calculation formula for the corresponding residual data is as follows:

[0068] Where, For the Resource deviation, Resource quota data for the current project. The historical resource quota data corresponding to the current project, is the residual mean, is the number of historical projects in the reference set, is the residual standard deviation, is the standardized residual coefficient, is the corresponding residual data, For the The weight of a resource, is the number of features in the resource quota data.

[0069] It should be noted that the standardized residual coefficient represents the standard deviation multiple of the average difference between the current project and the historical project, and is used to measure the degree of deviation. The resource weight represents the contribution of a certain feature quantity in the resource quota data to the overall deviation.

[0070] Furthermore, a quota prediction model is constructed based on BP neural network through resource intensity deviation data, sub-item quota abnormal data and corresponding residual data.

[0071] The specific calculation formula of the quota forecast model is as follows:

[0072] In the formula, is the quota forecast value, For resource intensity deviation data, For abnormal data of sub-item quota, is the corresponding residual data, 、 、 are weights respectively.

[0073] It should be noted that building a quota forecasting model using resource intensity deviation data, abnormal sub-item quota data, and corresponding residual data has the following advantages: Multi-dimensional fusion of data structures improves the model's perception of quota anomalies: three types of data form a three-dimensional input system of time series, space and reference from the construction execution layer, local structure layer and historical comparison layer: by integrating input features of different dimensions, a multi-dimensional characterization of the input volatility, local anomalies and historical consistency of quota information is achieved. The BP neural network can automatically learn the representation characteristics of various types of abnormal signals during training, effectively breaking through the limitations of traditional quota verification that only relies on static indicators or manual judgment.

[0074] Enhanced local anomaly detection capabilities address overall model deficiencies: Traditional quota forecasting models often use holistic statistical or machine learning methods, which are sensitive to overall trends in the data but insensitive to local outliers or concentrated shifts. By introducing abnormal data for sub-item quotas, the model possesses local recognition capabilities, effectively identifying fluctuations in sub-item quotas caused by process changes, geological differences, or local design adjustments, thereby improving the local robustness of the forecast.

[0075] Realize historical similarity mapping and build an analogy reasoning mechanism: The corresponding residual data establishes the quota deviation distribution between the current quota and historical high-similarity engineering cases, which can provide analogy support when data is sparse or new types of engineering samples are insufficient, thereby enhancing the model's generalization and migration capabilities.

[0076] In addition, the traditional quota model focuses on static prediction based on direct variables such as project quantity and category. This model innovatively introduces the characteristics of resource allocation volatility, spatial density characteristics of quota deviation, and cross-project matching deviation characteristics, shifting from static parameter prediction to dynamic process drive and behavioral characteristic analysis.

[0077] Furthermore, variational autoencoders are a type of generative model that has been widely used in many tasks, especially in image generation, anomaly detection, representation learning, etc.

[0078] The quota anomaly determination is performed based on a dynamic threshold, which is obtained based on a variational autoencoder. Specifically, Based on the variational autoencoder, the outputs of several quota prediction models are input into the encoder as original samples, and the samples are reconstructed by combining with the decoder output; Obtain reconstruction error based on original samples and reconstructed samples; Obtain the KL divergence of the potential distribution and build an initial decision model based on the reconstruction error; Based on the Bayesian optimization method, the optimal hyperparameter combination is searched for to optimize the initial judgment model and obtain the quota anomaly judgment model. The optimal hyperparameter combination is the parameters for controlling the dimension of the latent space and the weights for controlling the KL divergence that can maximize the quality of potential representation and anomaly detection performance. Obtaining a quota abnormality determination value based on a quota abnormality determination model; Based on the sliding window method, several quota abnormality judgment values ​​are used as calculation windows; Calculate the mean and standard deviation of the quota abnormality judgment value within the window based on statistical methods, and set the upper and lower thresholds as the dynamic threshold range; The quota abnormality judgment is performed through the dynamic threshold range. If the quota abnormality judgment value is greater than the upper threshold or less than the lower threshold, it is judged as abnormal.

[0079] The specific calculation formula of the quota abnormality determination model is as follows:

[0080] The upper threshold is specifically calculated as follows:

[0081] The specific calculation formula for the lower threshold is as follows:

[0082] Where, is the quota abnormality judgment value, For the original sample, To reconstruct the sample, To control the weight of KL divergence, is the KL divergence of the potential distribution, is the latent space vector, is the latent variable distribution inferred by the encoder based on the input sample, is the target prior distribution, is the mean value output by the quota forecast model, is the standard deviation of the quota forecast model output, is the upper threshold, is the lower threshold, is a hyperparameter.

[0083] It should be noted that using a variational autoencoder combined with a quota prediction model to determine quota anomalies has the following advantages: Variational autoencoders (VAEs) have the ability to learn unsupervised. By self-learning the latent representation of data, they can detect anomalies even in the absence of labeled data. This is particularly important for determining quota anomalies, as anomaly data is often difficult to collect and label, and traditional methods may rely on large amounts of labeled data.

[0084] VAEs effectively capture the complex structure and nonlinear characteristics of data. By modeling the data's latent space, they can identify subtle differences between normal and abnormal patterns. In quota anomaly detection, quota data often involves the interaction of multiple factors, and VAEs can discover potential abnormal patterns in high-dimensional space.

[0085] During the reconstruction process of input data, VAEs strive to learn a compressed latent representation. By comparing the error between the original data and the reconstructed data, anomalous data can be identified. For quota forecasting models, anomalous data often exhibits significant differences in reconstruction error from normal data. Monitoring these differences allows for effective identification of anomalies.

[0086] The outputs of multiple quota prediction models are used as raw sample inputs into a variational autoencoder, avoiding potential biases or blind spots associated with relying on a single model's predictions and improving the diversity and representativeness of the input features. The performance of a variational autoencoder is heavily dependent on the dimensions of the latent space and the weights of the KL term. Manual adjustment is inefficient and prone to falling into local optima. By introducing Bayesian optimization for adaptive hyperparameter adjustment, we can quickly identify hyperparameter combinations that improve both the quality of the latent representation and anomaly detection performance, enhancing the model's generalization and adaptability to new samples.

[0087] It's important to note that the sliding window method can dynamically adjust the threshold based on changes in time series data, allowing anomaly detection to automatically adjust to changes in the data at different stages. VAEs can generate the distribution of the latent space and calculate the reconstruction error, but the sliding window method can adjust the threshold based on the actual distribution of the current data, making the threshold more flexible and adaptable to different time periods or data patterns. This avoids the situation where traditional static threshold methods are inapplicable when data patterns change. Furthermore, the sliding window mechanism can focus on local changes and fluctuations in the data, thereby capturing anomalies that may occur in the short term. In determining quota anomalies, data fluctuations may be short-term emergencies, and using a fixed threshold is often prone to miss or misjudgment. However, the sliding window method can dynamically adjust the anomaly criteria within each window segment, more accurately judging abnormal short-term fluctuations.

[0088] Sliding windows allow for localized analysis of abnormal patterns over time, rather than relying solely on global static thresholds. This results in more accurate anomaly detection. For example, if quota data shows a gradual upward trend over a period of time, a sliding window can dynamically adjust the threshold so that it changes with the data, thus avoiding false positives.

[0089] S3, based on the judgment results, identifies the root cause characteristics of the quota anomaly based on the causal reasoning model, combines the SHAP method to analyze the contribution of the root cause characteristics, and achieves interpretability optimization.

[0090] In this example, causal reasoning and SHAP analysis methods are used to optimize the root cause characteristics of quota anomalies, which has the following advantages: Traditional data-driven methods can only identify surface features with strong correlation, but it is difficult to determine which are the fundamental driving factors. The introduction of causal reasoning models can remove false correlations from complex causal chains and lock in the starting variables of quota anomalies. In addition, prioritizing the optimization of root cause features can significantly improve regulatory efficiency and reduce resource waste caused by blind adjustments.

[0091] The SHAP method, through a precise marginal contribution allocation mechanism, quantifies the impact of each feature on quota anomalies. The causal relationship possesses a certain structural stability, preserving the core reasoning framework even when construction plans and personnel structures change. Compared to supervised learning methods that rely solely on training set models, this method is more suitable for heterogeneous scenario migration in complex engineering environments.

[0092] According to the judgment results, the root cause characteristics of the quota anomaly are identified based on the causal reasoning model, and the contribution of the root cause characteristics is analyzed in combination with the SHAP method. Specifically: Based on the abnormality determination results, a causal inference model is constructed. The causal relationship between quota detection data and historical construction data is analyzed through the causal inference model to identify the key factor characteristics that affect quota abnormalities. The causal inference model is constructed based on a graph structure modeling method, which uses the graph model to represent the causal relationship between each link in the construction process, resource utilization and quota data performance. The key factor characteristics include the deviation value of actual resource intensity, the deviation of mechanical equipment utilization rate, the difference between actual material consumption and quota ratio, the actual deviation rate of construction progress, and the construction period. The SHAP method is introduced to quantify the key factor features in the causal reasoning model and calculate the contribution of each feature to the quota abnormality judgment result; The importance of features is ranked according to their contribution, and interpretability optimization is performed based on the ranking results. The interpretability optimization is to adjust the construction plan and optimize resource allocation based on the contribution of each feature to the quota anomaly, and finally form a more targeted and effective quota optimization strategy to ensure that the quota execution during the construction process is more accurate and reasonable.

[0093] The specific calculation formula for the contribution is as follows:

[0094] Where, For contribution, For a feature among all features that do not contain key factors The feature set of is a feature subset, Features When inserted into all possible permutations, the feature The probability of the following occurrence, To add features The subsequent abnormality judgment value, To use only a subset of features The quota abnormal judgment value at this time.

[0095] It should be noted that the causal inference model described above is constructed using a graph-based modeling approach, utilizing a graph model to represent the causal relationships between various construction processes, resource utilization, and quota data performance. This can be understood as a graph-based approach where nodes in the graph represent variables (e.g., labor input, material consumption, construction plan deviation), while edges represent causal relationships between variables (i.e., the direct impact of one factor on another). This graph explicitly models the causal chain between construction processes, resource utilization behaviors, and quota data performance (e.g., quota overruns and quota anomalies). Leveraging the causal inference capabilities of the graph model, combined with observed quota anomalies, causal chain backtracking analysis is performed to determine which intermediate nodes play a role in the causal path. Causal inference techniques (e.g., Do-Calculus, backdoor adjustments, and structural equation models) can be used to locate key nodes. Once identified, further screening is conducted to identify key factor characteristics at the source of the causal chain that are decisive for the anomalous outcome. Root cause characteristics are characteristic variables that are at the beginning of the causal chain that triggers quota anomalies and have a substantial causal influence on the outcome variable. In this solution, the root cause characteristics refer to the key factor characteristics that lead to abnormal construction quotas.

[0096] It's important to note that historical construction data includes records related to quota execution, including historical quota standard datasets (quota standards for different projects, phases, and types of work), actual resource consumption records (consumption values ​​per unit of work for labor, materials, machinery, and other items), quota deviation data, construction log data, construction resource allocation and scheduling data, and management and control measures data. Historical construction data provides a training foundation for causal inference models based on time series causal paths; supports modeling causal relationships between quota anomalies and external factors or construction strategies; and, in conjunction with current inspection data, uses structured learning (such as DAG modeling and Granger causal analysis) to restore the chain causal mechanism that leads to quota anomalies.

[0097] It's important to note that the core of the causal inference model is causal analysis. It goes beyond simply considering data correlations and instead analyzes the causal relationships between different factors to identify the true causes of quota violations. In this scenario, the model combines quota inspection data with historical construction data, comparing and analyzing the causal relationships between these data to identify potential causes of quota anomalies.

[0098] Example 2, Figure 2 This is a schematic diagram of the structure of the distribution network engineering mechanized construction quota optimization system based on artificial intelligence provided in an embodiment of the present application, including a quota detection optimization solution generation module, a quota prediction module, and a quota optimization module, with connections between the modules: The quota detection optimization scheme is based on the genetic algorithm to obtain the initial solution of the simulated annealing algorithm and optimize the screening; A quota prediction module is used to obtain quota detection data based on the quota detection optimization solution to build a quota prediction model, and to determine quota anomalies based on dynamic thresholds obtained based on variational autoencoders; The quota optimization module is used to identify the root cause characteristics of abnormal results and analyze the contribution of the root cause characteristics in combination with the SHAP method. The root cause characteristics are identified through a causal reasoning model.

[0099] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0100] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0101] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0102] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0103] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0104] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The method for optimizing the mechanized construction quota of distribution network engineering based on artificial intelligence is characterized by: The steps include: Construct a multi-objective construction collaborative optimization model to generate a quota inspection optimization plan. The solution is based on the genetic algorithm to obtain the initial solution of the simulated annealing algorithm and optimize the screening; Based on the quota detection optimization solution, quota detection data is obtained to build a quota prediction model, and quota anomaly judgment is made based on a dynamic threshold obtained based on a variational autoencoder; Identify the root cause features of abnormal results and analyze the contribution of the root cause features using the SHAP method. The root cause features are identified through a causal reasoning model.

2. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 1 is characterized in that: The multi-objective construction collaborative optimization model is constructed as follows: The multiple objectives include multiple tasks and multiple resources; Get the first data construction task set in each task; Obtain the second data in each type of resource to construct a resource collection; An objective function and allocation constraints are established based on the construction task set and the resource set. The objective function includes the objective functions of carbon emission intensity, maximizing resource utilization, and task priority.

3. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 2 is characterized in that: The initial solution of the simulated annealing algorithm based on the genetic algorithm is specifically: Generate an initial scheduling solution code based on the construction task set and the resource set, wherein the initial scheduling solution code includes a task sequence and a resource allocation matrix; The objective functions are respectively based on the entropy weight method to obtain the information entropy of each objective function; According to the information entropy, a multi-objective fitness evaluation model is constructed based on the linear weighted summation method to output the fitness; The initial scheduling scheme is encoded based on a strategy combining elite retention and tournament selection, and the selection operation is performed through fitness; Performing crossover and mutation operations on the tournament selection results to generate new individuals, and combining them with the elite retention results to form the next generation. The crossover operation includes performing order-preserving recombinant crossover on the task sequence and partial matching crossover on the resource allocation matrix. The mutation operation is performed by perturbing the relationship between task execution time and resource matching. When the fitness continuously iterates and reaches the preset algebraic termination, the current optimal solution is output as the initial solution of the simulated annealing algorithm.

4. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 3 is characterized in that: The crossover and mutation operations are performed on the tournament selection results to generate new individuals, and combined with the elite retention results to form a complete next generation, specifically: Based on the output of the multi-objective fitness evaluation model, each individual in the initial scheduling scheme code is sorted in descending order according to fitness, and a preset number of optimal individuals are selected to form an elite subset; The elite subset is copied into the next generation without participating in crossover and mutation operations; Obtain the probability of the individual with the best fitness based on descending sorting, and select the individuals used for the tournament selection method from the individuals remaining in the elite reserve; Sort the individuals selected by the tournament by fitness and select the best individual as the parent according to the probability; Perform crossover and mutation operations on the selected parents to generate new individuals.

5. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 1 is characterized in that: The initial solution of the simulated annealing algorithm is obtained and optimized and screened, specifically: Set the initial solution, initial temperature, temperature update formula and temperature threshold, and use the output value of the multi-objective fitness evaluation model as the objective function value for iteration; Obtain the current temperature in each iteration, and based on the current temperature, perform a neighborhood perturbation operation on the current solution to generate a new candidate solution. The neighborhood perturbation operation includes task sequence perturbation and resource allocation perturbation. A new solution is obtained by calculating the objective function value of the candidate solution and making a judgment; After each round of iteration, the temperature is lowered based on the temperature update formula. When the temperature threshold is reached, the iteration is stopped and the global optimal new solution is output as the quota detection optimization solution.

6. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 1 is characterized in that: The above-mentioned determination of quota abnormality based on the dynamic threshold is specifically as follows: Based on the variational autoencoder, the outputs of several quota prediction models are input into the encoder as original samples, and the samples are reconstructed by combining with the decoder output; Obtain reconstruction error based on original samples and reconstructed samples; Obtain the KL divergence of the potential distribution and build an initial decision model based on the reconstruction error; Based on the Bayesian optimization method, the optimal hyperparameter combination is searched to optimize the initial judgment model and obtain the quota abnormality judgment model; Obtaining a quota abnormality determination value based on a quota abnormality determination model; Based on the sliding window method, several quota abnormality judgment values ​​are used as calculation windows; Calculate the mean and standard deviation of the quota abnormality judgment value within the window based on statistical methods, and set the upper and lower thresholds as the dynamic threshold range; The quota anomaly determination is performed through the dynamic threshold range.

7. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 1 is characterized in that: The root cause features of the identified abnormal results are analyzed in combination with the SHAP method to analyze the contribution of the root cause features. The root cause features are identified through a causal reasoning model, specifically: Based on the abnormality determination results, a causal reasoning model is constructed. The causal relationship between the quota inspection data and the historical construction data is analyzed through the causal reasoning model to identify the key factor characteristics that affect the quota abnormality. The causal reasoning model is constructed based on a graph structure modeling method. The SHAP method is introduced to quantify the key factor features in the causal reasoning model and calculate the contribution of each feature to the quota abnormality judgment result; The importance of features is sorted according to their contribution, and interpretability is optimized based on the sorting results.

8. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 1 is characterized in that: The quota detection data includes resource intensity deviation data, and the specific acquisition method is as follows: The construction period is divided into several stages according to the changing trend of resource input through a change point detection algorithm based on a time series segmentation model with a minimum description length criterion; Obtain resource intensity data for each stage and calculate standardized resource intensity based on cost weight normalization; Based on the standardized resource intensity, a standardized resource intensity curve is constructed and fitted using the kernel regression method. The actual resource intensity is obtained, and the resource intensity deviation data is obtained by calculating the average absolute deviation between the actual resource intensity and the fitted standardized resource intensity curve.

9. The method for optimizing the construction quota of mechanized distribution network engineering based on artificial intelligence according to claim 1, characterized in that: The quota detection data includes abnormal data of sub-item quotas, and the specific acquisition steps are as follows: Obtain quota data for several sub-projects and construct a multi-dimensional feature vector set; Based on the density clustering algorithm, a domain relationship graph between sub-items is established for the multi-dimensional feature vector set, wherein nodes in the domain relationship graph represent sub-items, and edges represent the adjacent relationship between two nodes. The adjacent relationship is obtained by obtaining the similarity between nodes using the Mahalanobis distance method; Based on the LOF method, the local anomaly factor of each item is calculated according to the domain relationship graph; All local anomaly factors that are greater than a preset local anomaly determination threshold are obtained, and the sub-item quota anomaly data is quantified according to the proportion of the local anomaly factors.

10. A system using the artificial intelligence-based distribution network engineering mechanized construction quota optimization method according to any one of claims 1 to 9, characterized in that: It includes the quota detection optimization plan generation module, the quota prediction module and the quota optimization module. There are connections between the modules: The quota inspection optimization scheme generation module is used to build a multi-objective construction collaborative optimization model and solve and generate the quota inspection optimization scheme. The solution is based on the genetic algorithm to obtain the initial solution of the simulated annealing algorithm and optimize the screening; A quota prediction module is used to obtain quota detection data based on the quota detection optimization plan to build a quota prediction model, and to determine quota anomalies based on dynamic thresholds obtained based on variational autoencoders; The quota optimization module is used to identify the root cause characteristics of abnormal results and analyze the contribution of the root cause characteristics in combination with the SHAP method. The root cause characteristics are identified through a causal reasoning model.

Images