Power transmission and transformation project cost intelligent evaluation method based on cost fluctuation characteristics

Abstract

The invention discloses a power transmission and transformation project cost intelligent review method based on cost fluctuation characteristics, and relates to the technical field of intelligent review, and the method comprises the following steps: collecting historical cost data, preprocessing the data based on NACEMD, and recognizing power transmission and transformation project cost influence factors; judging main influence factors influencing the cost by adopting correlation analysis, and carrying out clustering analysis on different project types to obtain different types of power transmission and transformation project sample sets; on the basis of the historical cost data and the transmission and transformation project sample set, cost intervals of different types of costs of the transmission and transformation project are calculated through a Cheff-Snow ratio inequality for different types of costs, and fluctuation intervals of different types of costs are judged; corresponding intelligent methods are set for different types of fees, and based on prediction analysis of the three types of fees, evaluation standard determination of the power transmission and transformation project cost is realized through input of related project parameters. And the scientificity and accuracy of the review process are ensured through model selection for different fluctuation types.

Description

Technical Field

[0001] The present invention relates to the technical field of project data statistical analysis, and more specifically, to an intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics. Background Art

[0002] In the construction and management of modern power transmission and transformation projects, cost control is one of the key links. Since this field involves the interaction of multiple factors, traditional cost assessment methods are often unable to effectively cope with complex market changes and uncertainties. This restricts the accuracy of project cost forecasting, thus affecting the overall economic benefits of the project.

[0003] With the development of information technology, data-driven methods have gradually been applied to the review process of engineering cost. However, existing models mostly focus on the prediction of a single type of cost, lacking a comprehensive analysis and dynamic evaluation of cost fluctuation characteristics. In addition, traditional methods are usually unable to effectively handle noise and outliers in the data, resulting in inaccurate cost review results.

[0004] In order to solve the above problems, by combining modern data analysis technology and intelligent algorithms, through in-depth mining and analysis of historical data, the key factors affecting the cost of power transmission and transformation projects can be better identified and quantified. The introduction of machine learning algorithms, such as support vector machines (SVM) and recurrent neural networks (RNN), can improve the accuracy and adaptability of cost prediction. In addition, the use of methods such as cluster analysis and correlation analysis can more comprehensively identify the characteristics of cost fluctuations, thereby providing support for subsequent intelligent reviews.

[0005] In this context, this paper proposes an intelligent cost review technology for power transmission and transformation projects based on cost fluctuation characteristics, aiming to provide scientific and reasonable cost review standards by systematically combing and identifying influencing factors, combining historical data analysis with intelligent prediction models. This method not only improves the accuracy of cost review, but also provides effective support for the dynamic management and optimization of project costs.

[0006] The above disclosed technical solutions have at least the following technical problems: the existing models mostly focus on the prediction of a single type of cost, lacking a comprehensive analysis and dynamic evaluation of the cost fluctuation characteristics. In addition, traditional methods are usually unable to effectively handle noise and outliers in the data, resulting in inaccurate results of cost review. In view of the above problems, the present invention proposes a solution. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an intelligent evaluation method for the cost of power transmission and transformation projects based on cost fluctuation characteristics. Based on the predictive analysis of three types of costs and through the input of relevant engineering parameters, the evaluation criteria for the cost of power transmission and transformation projects are determined to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions: An intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics, comprising the following steps: collecting historical cost data, preprocessing the data based on NACEMD, and identifying factors affecting the cost of power transmission and transformation projects, wherein the historical cost data includes project type, construction plan, construction environment, and the cost of various costs; Correlation analysis is used to determine the main factors affecting the cost, and cluster analysis is performed on different types of projects to obtain a sample set of power transmission and transformation projects of different categories; Based on the historical cost data and the sample set of power transmission and transformation projects, the cost ranges of different types of power transmission and transformation projects are calculated by using the Chevshcherby inequality, and the fluctuation ranges of different types of costs are determined, which are divided into three parts: stable trend, orderly fluctuation and disordered oscillation. Set corresponding intelligent methods for different types of expenses. For the disordered oscillation part, build a GRU prediction model based on improved particle swarm optimization. For the stable trend part, build a Markov Fourier modified grey prediction model. For the orderly fluctuation part, build a support vector machine model based on genetic algorithm optimization. Based on the forecast analysis of three types of costs and through the input of relevant project parameters, the evaluation criteria for the transmission and transformation project cost are determined.

[0009] In a preferred embodiment, the historical cost data is collected and preprocessed based on NACEMD, specifically: execute M initialization CEMD, and add white noise with an amplitude of k for a certain number of times; add the white noise to the original data sequence of the historical cost data to obtain a complex signal; project the complex signal onto φk, where φk is the projection direction; when the sine value of φk is not zero, solve the maximum point, and perform cubic spline interpolation on it to obtain the upper and lower envelopes in each projection direction; calculate the average value of the boundary envelope line, and obtain each component signal according to the average value of the boundary envelope line; calculate the mean value of the IMF component and the residual; and calculate the mean value of each IMF component obtained after M CEMD processes as the output.

[0010] In a preferred embodiment, the correlation analysis is adopted to determine the main influencing factors affecting the cost, and cluster analysis is performed on different project types to obtain a sample set of power transmission and transformation projects of different categories, specifically: the correlation coefficient between each influencing factor and the project cost is calculated based on the Pearson correlation coefficient according to the identified influencing factors of the power transmission and transformation project cost; the correlation between each factor and the project cost is explained according to the calculated correlation coefficient; the main factors that have a significant impact on the project cost are screened out according to the interpretation results of the correlation coefficient; the screened out main influencing factors are used as input variables for cluster analysis; the input variables are used as samples and the maximum sum is selected Minimum value, based on the preset formula, standardize the data to obtain the standardized sample data; take the set of samples with similar characteristics and attributes in the sample data as a cluster; take the average distance between the sample and other points in the cluster as the cohesion within the cluster; take the average distance between the sample and all points in the nearest cluster as the separation from the nearest cluster; divide the difference between the cluster separation and the cluster cohesion by the larger number of the two to obtain the silhouette coefficient; evaluate the clustering effect according to the range of the silhouette coefficient; the closer the silhouette coefficient is to 1, the better the clustering effect; the closer it is to -1, the worse the clustering effect; close to 0 means that the sample is at the boundary of two clustering results. If the silhouette coefficient is low, adjust the number of clusters, re-perform cluster analysis, and calculate the new silhouette coefficient; by comparing the silhouette coefficients under different cluster numbers, find the clustering scheme with the largest silhouette coefficient; According to the best clustering scheme, the clustering results are sorted out to obtain a sample set of power transmission and transformation projects of different categories.

[0011] In a preferred embodiment, based on historical cost data and a sample set of power transmission and transformation projects, different types of costs are calculated through the Chebyshev inequality to obtain cost ranges of different types of costs for power transmission and transformation projects, specifically: calculating the median, upper quartile, lower quartile, maximum value, minimum value, sample mean and sample variance of the sample data; using a box plot to eliminate outliers to perform an outlier analysis on the sample data, wherein the outlier is less than the lower quartile minus 1.5 times the interquartile distance and greater than the upper quartile plus 1.5 times the interquartile distance; presetting a confidence level, and calculating the cost range of each cost type at the confidence level according to the Chebyshev inequality.

[0012] In a preferred embodiment, for the disordered oscillation part, an improved particle swarm optimized GRU prediction model is constructed, specifically: the GRU network structure is divided into two parts, training and prediction, and the prediction part includes three parts: input layer, hidden layer and output layer; for a given training set, a number of parameter vectors with a dimension of E are initialized; according to the sample loss, the gradient descent algorithm is used to train the GRU network so that the sample loss function value is minimized, and the weights of each network parameter are updated; the initial weight matrix and parameter vector of the GRU constitute a particle swarm individual, and for each individual in the group, each gate control unit is calculated, and the candidate Select the state and the output of the network; calculate the GRU prediction result corresponding to each particle, take the root mean square error as the fitness function of the IPSO algorithm, and optimize the particle corresponding to the GRU with the smallest root mean square error; update the group optimal and individual optimal and calculate the aggregation degree; when the aggregation degree reaches the threshold, update the particle speed and position based on the new dynamic topology and particle search behavior; after reaching the condition of the number of iterations of the particle swarm termination, complete the initialization process of the GRU model; after completing the initialization process of the GRU, build and train the GRU model for each module respectively, and finally integrate the prediction results of each component through the new GRU network.

[0013] In a preferred embodiment, for the ordered fluctuation part, a support vector machine model based on genetic algorithm optimization is constructed, specifically: a fitness function is defined, and relative mean square error is used as the objective function: a prediction model is established according to the training sample data, and the test sample data is predicted, and the objective function is used to measure the quality of the learning algorithm; the parameters of the model are binary-encoded, and an initial population is randomly generated; the chromosomes in the population are decoded, and the objective function corresponding to each individual is calculated; it is determined whether the performance of the population meets the optimization criteria, and if the optimization criteria are met, the optimal parameters are output, otherwise, according to the genetic strategy, selection, crossover and mutation operators are applied to the population to generate the next generation population and start a new generation of inheritance; the objective function obtained by learning is checked, and if the selected parameter value meets the MAPE error requirement, binary encoding is continued; the obtained optimal parameter value is assigned to the prediction learning algorithm, trained with the training sample set, and then the test sample is predicted to obtain the corresponding prediction result.

[0014] The technical effects and advantages of the intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics of the present invention are as follows: 1. The present invention introduces a variety of advanced models, including the GRU prediction model with improved particle swarm optimization, the Markov Fourier modified grey prediction model and the support vector machine (SVM) model based on genetic algorithm optimization. These models not only take into account the historical data of costs, but also combine real-time environmental variables, which can effectively improve the prediction accuracy of the transmission and transformation project cost. By selecting models for different types of fluctuations, the scientificity and accuracy of the review process are ensured.

[0015] 2. In the disordered oscillation part, the GRU model with improved particle swarm optimization has a strong time series prediction capability and can identify potential cost fluctuation patterns by analyzing the time dependency of historical data. The gating mechanism of this model enables it to make flexible adjustments in long-term dependency and short-term fluctuation, enhancing the ability to predict future cost changes, thereby achieving dynamic monitoring and management.

[0016] 3. Through the Markov Fourier modified grey prediction model, the state transfer characteristics of the Markov chain are introduced in the cost prediction of the stable trend part, combined with the frequency domain analysis of Fourier transform, which improves the ability to capture long-term trends. The design of this model enables managers to adjust project budgets based on real-time data and optimize capital allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The present invention is a schematic diagram of the structure of an intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics.

[0018] Figure 2 is the criterion for determining relevance. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Embodiment 1, Figure 1 The present invention provides an intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics, comprising the following steps: Identify the factors affecting the cost of power transmission and transformation projects, collect historical cost data and pre-process the data based on CEMD. The historical cost data includes the project type, construction plan, construction environment, and the cost of various expenses; Correlation analysis is used to determine the main factors affecting the cost, and cluster analysis is performed on different types of projects to obtain a sample set of power transmission and transformation projects of different categories; Based on historical data, the cost range of different types of costs is calculated by using the Chevshcherby inequality, and the fluctuation range of different types of costs is determined, which is divided into three parts: stable trend, orderly fluctuation and disordered oscillation. Set corresponding intelligent methods for different types of expenses. For the disordered oscillation part, build an improved particle swarm optimized GRU prediction model. Based on the stable trend part, build a Markov Fourier modified grey prediction model. Based on the disordered oscillation part, build a support vector machine model based on genetic algorithm optimization. Based on the forecast analysis of three types of costs and through the input of relevant project parameters, the evaluation criteria for the transmission and transformation project cost are determined.

[0021] S1, identifying the factors affecting the cost of power transmission and transformation projects, collecting historical cost data and pre-processing the data based on NACEMD, wherein the historical cost data includes the project type, construction plan, construction environment, and the cost of various expenses; The identification of factors affecting the cost of power transmission and transformation projects is specifically as follows: Based on the analysis of the influencing factors of cost structure in the dimensions of engineering technology, policy standards, management specifications, natural conditions, and market elements, other relevant factors of power grid projects are introduced, and the influencing factors are summarized as follows: project division domain, cost structure domain, engineering (equipment) quantity domain, factor price domain, subdivided cost domain, natural environment domain, engineering technology domain, market element domain, policy standard domain, and management specification domain.

[0022] Among them, the indicator data in the project division domain, cost structure domain, engineering (equipment) quantity domain, factor price domain, and subdivided cost domain are relatively mature, and the internal data information quantification ability and data integration and collection ability are strong.

[0023] The project division domain is based on the "Regulations on the Preparation and Calculation of Power Grid Project Construction Budgets", which divides power transmission and transformation projects into three categories: power transformation projects, overhead power transmission line projects, and cable power transmission line projects; the cost composition domain is based on the "Regulations on the Preparation and Calculation of Power Grid Project Construction Budgets", which brings together basic cost structures such as construction project costs, installation project costs, and equipment purchase costs; the engineering (equipment) quantity domain includes the quantity of each sub-item or sub-item project and component expressed in physical measurement units or natural measurement units in the equipment engineering quantity, construction engineering quantity, and installation engineering quantity; the factor price domain refers to the price level of the factors involved in the construction process, which is determined according to the standard price or market price for the calculation of the cost, and the factors involved include labor, materials, machinery, and equipment; the subdivided cost domain is based on the cost calculation regulations, except for the direct engineering costs and equipment purchase costs in the construction engineering costs and installation engineering costs, which need to be determined by quantity and price, including the measures costs, indirect costs, profits, taxes, etc. of the construction (installation) project.

[0024] When collecting data, there may be certain data errors. The CEMD method can effectively separate the noise components from the original data, thereby improving the signal-to-noise ratio of the data. The denoised data can better reflect the true characteristics and trends of the data, thereby providing more reliable and accurate information when conducting data mining, pattern recognition or predictive analysis.

[0025] The captured data is denoised based on the NACEMD method, specifically: Perform M initialization CEMD (complex data empirical mode decomposition) and add white noise with amplitude k for each time. Add the white noise xn(z) to the original data sequence x0(z) of the historical cost data to form a complex signal xc(z)=x0(z)+ixn(z), where z is a time series variable; Project the complex signal onto φk, where φk is the projection direction, φk=2πk / N, 1≤k≤N: , Substituting Euler's formula into the above formula: , when When , the direction of the original extreme value selection changes, and the solution is The maximum point of , and perform cubic spline interpolation on it to obtain the upper and lower envelopes in each projection direction; Calculate the average value m(z) of the boundary envelope, and then obtain each component signal h(z)=x(z)-m(z) according to the average value m(z) of the boundary envelope, where x(z) is the cost data sequence; Get the mean of IMF components and residuals; The mean of each IMF component obtained after M CEMD processes is calculated as the output.

[0026] S2, conduct correlation analysis to determine the main factors affecting the cost, and perform cluster analysis on different project types to obtain a sample set of different types of power transmission and transformation projects; The correlation analysis is adopted to determine the main factors affecting the construction cost, and cluster analysis is performed on different types of projects to obtain different types of power transmission and transformation project sample sets, specifically: According to the identified factors affecting the cost of power transmission and transformation projects, the correlation coefficient between each factor and the project cost is calculated based on the Pearson correlation coefficient; Based on the calculated correlation coefficient, explain the correlation between each factor and the project cost; According to the interpretation results of the correlation coefficient, the main factors that have a significant impact on the project cost are screened out; The screened main influencing factors were used as input variables for cluster analysis; Take the input variable as a sample and select the maximum and minimum values, perform standardization based on the preset formula, and obtain the standardized sample data; The collection of samples with similar characteristics and attributes in the sample data is regarded as a cluster; The average distance between the sample and other points in the cluster is taken as the cohesion within the cluster; The average distance between a sample and all points in the nearest cluster is considered as the separation from the nearest cluster; The silhouette coefficient is obtained by dividing the difference between the separation degree of the cluster and the cohesion degree of the cluster by the larger number of the two; The clustering effect is evaluated according to the value range of the silhouette coefficient. The closer the silhouette coefficient is to 1, the better the clustering effect is; the closer it is to -1, the worse the clustering effect is; and if it is close to 0, it means that the sample is at the boundary of two clustering results.

[0027] If the silhouette coefficient is low, adjust the number of clusters, re-perform cluster analysis, and calculate a new silhouette coefficient; By comparing the silhouette coefficients under different cluster numbers, find the clustering scheme with the largest silhouette coefficient; According to the best clustering scheme, the clustering results are sorted out to obtain a sample set of power transmission and transformation projects of different categories.

[0028] The correlation analysis method adopts the Pearson correlation coefficient analysis method to quantitatively analyze the key factors affecting the cost of substation engineering. The Pearson correlation coefficient is a statistical analysis method used to quantify the linear relationship between two variables. Its core principle is to determine the degree of association between them by calculating the ratio of the covariance between variables to their respective standard deviations. Specifically, the value of the Pearson correlation coefficient ranges from -1 to +1, indicating a complete negative correlation and a complete positive correlation, and a value close to 0 indicates that there is almost no linear relationship between the two. This method is widely used in data analysis, especially in the fields of economics, finance, and engineering, and can help identify the key factors affecting the cost, thereby providing data support for decision-making. By analyzing the correlation between influencing factors and the cost of substation engineering, decision makers can more effectively control costs and allocate resources.

[0029] The main calculation principle of the Pearson correlation coefficient analysis method is as follows: Calculate the standard deviation of each variable itself:

[0030]

[0031] Compute the variance between variables:

[0032] Calculate the correlation coefficient between variables:

[0033] In the formula, For variables and The correlation coefficient of For variables and The variance of are the standard deviations of the two variables, is the sample size, They are The values ​​of the two variables at the sample points, are the means of the two variables.

[0034] The correlation coefficient is used to determine the correlation of variables. The value range of the correlation coefficient is between 1 and -1. -1 indicates a completely negative correlation, 1 indicates a completely positive correlation, and the closer the absolute value is to 1, the stronger the correlation is. 0 indicates that the two variables are unrelated. Figure 2 : Through correlation analysis, we can determine the main factors affecting each cost, and on the other hand, we can combine the fluctuation of factors to determine the degree of influence of different factors on the cost, which is used for the comprehensive calculation of the subsequent warning threshold. Combined with the results of correlation analysis, we select the main factors with a correlation greater than 0.8.

[0035] In this embodiment, after correlation analysis of the influencing factors of construction projects, installation projects, equipment purchase projects and other projects, the main factors with correlation greater than 0.8 are: The construction project items include the number of transformers, voltage level, support and foundation construction costs, pollution level, ground bearing capacity, construction concrete usage, and earthwork volume; The installation project items include the number of main transformers, number of lightning arresters, number of voltage transformers, unit price of post insulators, and quantity of post insulators used; Equipment purchase items include the number of main transformers, capacity for this period, number of lightning arresters, unit price of main transformers, and scale of outgoing lines for this period; Other project aspects include the full set of startup trial operation fees, special commissioning fees, project construction technical service fees, and project construction management fees.

[0036] The cluster analysis of different engineering types is specifically as follows: The K-means clustering model is mainly used to support the research. Cluster analysis is mainly based on the distance between classification objects, which is easily affected by the measurement unit of the clustering variable. The larger the magnitude of the variable, the greater the impact on the distance calculation result, and it will dominate the clustering process, thereby covering up other variables with smaller magnitudes and causing deviations in the clustering results.

[0037] Therefore, it is necessary to introduce data standardization methods to process sample data, remove the influence of dimensions, convert dimensional values ​​into dimensionless pure values, and solve the problem of large numerical differences between features to avoid affecting clustering results. Commonly used data standardization methods include range standardization and Z-score standardization.

[0038] The range standardization method selects the maximum and minimum values ​​of the sample data, and standardizes the original data using the following formula, ultimately mapping all data to the interval [0,1].

[0039]

[0040] In the formula, , Respectively represent the sample data after and before standardization; , Respectively represent the maximum and minimum values ​​in the sample data.

[0041] When the maximum and minimum values ​​of a certain indicator are unknown, or there are outliers that exceed the range, the range standardization method is no longer applicable. Another most commonly used method for data standardization can be used, namely Z-score standardization, also called standard deviation standardization, which is based on the mean of the original data. and standard deviation Standardize the data and convert the original value of A Use the Z-score method to standardize The specific calculation formula is shown below.

[0042]

[0043] Reasonable selection of the number of clusters is crucial to the clustering effect and analysis depth. If the number of clusters is small, the samples contained in each cluster will be greatly different from their corresponding cluster centers, and the cluster samples will be relatively scattered, which cannot effectively reflect the characteristics of each category. If the number of clusters is large, the classification will be too refined, resulting in smaller differences between clusters, and the clustering results will not be able to reflect the characteristics of each cluster.

[0044] Obviously, as the number of clusters k increases, the sample division becomes more refined, the degree of aggregation of each cluster gradually increases, and the sum of squared errors (SSE) naturally decreases. When k is less than the optimal number of clusters, the increase in k will greatly increase the degree of aggregation of each cluster, so the decrease in SSE will be large; and when k reaches the actual number of clusters, the return of the degree of aggregation obtained by increasing k will rapidly decrease, so the decrease in SSE will decrease sharply, and then tend to be flat as the k value continues to increase. In other words, the relationship between SSE and k is in the shape of an elbow, and the k value corresponding to this elbow is the optimal number of clusters for the data. The calculation method of SSE is shown in the following formula.

[0045]

[0046] In the formula, The clustering result is cluster; The clustering result is The cluster centers.

[0047] Silhouette analysis is also a calculation method for measuring the concentration of samples in a cluster. It is also applicable to other methods besides K-means clustering. This topic uses silhouette analysis as a method to test clustering results.

[0048] The core of profile analysis is to calculate the profile coefficient of a single sample. The calculation steps are divided into the following three steps: (a) The average distance between the sample and other points in the cluster is taken as the cohesion within the cluster; (b) The average distance between a sample and all points in the nearest cluster is considered as the separation from the nearest cluster; (c) The silhouette coefficient is calculated by dividing the difference between the cluster separation and the cluster cohesion by the larger number of the two. The calculation formula is shown below.

[0049]

[0050] The value of the silhouette coefficient is between -1 and 1. When the cluster cohesion and separation are equal, the silhouette coefficient is 0. When , the silhouette coefficient is approximately 1, and the model performance is optimal. In the interval [-1,1], it is generally believed that a silhouette coefficient greater than 0.7 indicates an excellent clustering result.

[0051] S3, based on historical data analysis, uses the Chevshcherby inequality to calculate the cost ranges of different types of costs for power transmission and transformation projects, and determines the fluctuation ranges of different types of costs, which are divided into three parts: stable trend, orderly fluctuation, and disordered oscillation; Calculate the median, upper quartile, lower quartile, maximum value, minimum value, sample mean and sample variance of the sample data; A box plot is used to remove outliers to perform an outlier analysis on the sample data, wherein the outlier is less than the lower quartile QL minus 1.5 times the interquartile range QIQR, and greater than the upper quartile QU plus 1.5 times the interquartile range QIQR; According to Chebyshev’s inequality principle, at least 80% of the statistics fall in the interval Φ; Φ={x|μ-kσ≤x≤μ+kσ} Among them, μ is the mean value of the sample data, σ is the standard deviation of the sample data, and k is the interval coefficient; Trend analysis: Use moving average or linear regression methods to analyze the data. If the data shows a stable upward or downward trend, it is considered to be stable; if the trend fluctuates greatly and it is impossible to determine a clear upward or downward direction, it is considered unstable.

[0052] Volatility analysis: Calculate the standard deviation and coefficient of variation of the data. If the standard deviation is small and the coefficient of variation is low, it means that the data has low volatility and is a stable trend; if the standard deviation is moderate and the fluctuation range is relatively fixed, it is an orderly fluctuation; and if the standard deviation is large and the fluctuation is irregular, it means disordered oscillation.

[0053] Time series decomposition: Use time series decomposition techniques to decompose data into trend, seasonality, and residual components. By analyzing the pattern of residuals, the fluctuation characteristics of the data can be determined.

[0054] S4, respectively set corresponding intelligent methods for different types of expenses. For the disordered oscillation part, build an improved particle swarm optimized GRU prediction model. Based on the stable trend part, build a Markov Fourier modified grey prediction model. Based on the disordered oscillation part, build a support vector machine model based on genetic algorithm optimization. 1) No need for vibration part GRU is a network model for processing time series. Compared with the traditional BP algorithm, GRU has stronger learning ability and can selectively forget or retain the input historical information through "gating", so as to achieve the prediction of future data. The network structure is divided into two parts: training and prediction. The prediction part includes three parts: input layer, hidden layer and output layer. According to the sample loss, the gradient descent algorithm is used to train the GRU network to minimize the sample loss function value and update the weights of each network parameter.

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] The phase point of the phase space reconstruction matrix is ​​used as the input part x of the network, and the sequence As the output part , is the hidden layer, is the sigmoid activation function, They are the update gate and reset gate of the recurrent network respectively. is the weight between the input layer and the update gate at time t. is the weight between the hidden layer at time t-1 and the update gate at time t, is the weight between the input layer and the reset gate at time t. is the weight between the hidden layer at time t-1 and the reset gate at time t, is the weight between the hidden layer and the output layer at time t. is the Hadamard product of matrices.

[0061] GRU has a fast learning speed and a small training error, but the initialization weights between its hidden layers and between the hidden layer and the output layer are randomly selected, which lacks pertinence for the subsequent training process. The improved particle swarm optimization algorithm has a strong global optimization capability. Introducing it into the network training process of GRU can further improve its model prediction ability. The GRU weight initialization process based on the improved particle swarm algorithm is as follows: For a given training set ,initialization A parameter vector with dimension E (E=Kn+1, where K is the number of hidden layers of GRU and n is the number of input neurons) (q=1, 2, ..., ), the value range of any dimension is [-1, 1], and g is the number of iterations.

[0062] The initial weight matrix and bias vector of GRU constitute the individual particle swarm. For each individual in the swarm , calculate each gating unit, candidate state and the output results of the network.

[0063] Calculate the GRU prediction result corresponding to each particle, take the root mean square error as the fitness function of the IPSO algorithm, and optimize the particle corresponding to the GRU with the smallest root mean square error.

[0064] Update group optimality and individual optimal .

[0065] Calculate the degree of aggregation.

[0066] When the aggregation degree reaches a threshold, the proposed new dynamic topology and particle search behavior updates the velocity and position of the particles.

[0067] After reaching the particle swarm termination iteration condition, the initialization process of the GRU model is completed. After completing the initialization process of the GRU, continue to use the algorithm to learn the network parameters. Build and train the GRU model for each module separately, and finally integrate the prediction results of each component through the new GRU network.

[0068] 2) Trend stability part The specific modeling of Markov Fourier modified grey prediction model is as follows: (1) A sequence of n nodes, Perform an accumulation to obtain the cumulative sequence:

[0069] in,

[0070] Because the sequence With gray finger distribution, construct its first-order differential equation:

[0071] Then The sequence is averaged and the sequence is generated ,

[0072] in,

[0073] Therefore, the differential equation transforms into

[0074] in

[0075] (3) Then refer to the grey prediction method and use the least squares method to estimate the parameter vector

[0076] (4-14) in,

[0077] Thus, the estimated parameters a and b are obtained and the grey prediction equation is constructed:

[0078] (4) Cumulative subtraction to obtain the predicted value of the stable trend cost type

[0079] Therefore, according to the above formula, the stable trend cost type forecast sequence is:

[0080] (5) Use the residual formula to calculate the residual and obtain the residual sequence

[0081]

[0082] (6) Use Fourier series to correct residuals.

[0083] 1) First, convert the residual sequence into a Fourier series sequence

[0084] 2) Then use the formula Calculate each parameter and obtain the parameter estimate of the Fourier series sequence

[0085]

[0086]

[0087] 3) After correction by Fourier series, the predicted value of the stable trend fee type is modified to:

[0088] (7) Perform state division of the Markov chain. Based on the historical data of the stable trend fee type, divide the data into n states.

[0089]

[0090] in , the number of state partitions n and It can be determined based on the original data.

[0091] (8) According to the state transition probability formula, calculate the transition probability between the states divided in step (7) , and then construct the state transition probability matrix P.

[0092]

[0093]

[0094] in, Indicates the relevant parameters from After k steps, the state is transferred to The probability value of the state. The relevant parameters representing the data are represented by After k steps, the state is transferred to The original number of samples of the state, Indicates that The raw number of samples for the state.

[0095] (9) Based on step (8) and the state transition probability matrix obtained, the cost level of the future stable trend cost is finally predicted based on the state transition matrix and the formula.

[0096]

[0097] 3) Orderly fluctuation part Support vector machines were originally used to solve classification problems, and prediction technology was developed on their basis. Therefore, we start from classification theory and expand to the research and discussion of prediction technology. Suppose that we consider the training sample set {(xi,yi),i=1,2,…,l}, where xi∈Rn is the i-th example of the input pattern, and yi∈{+1,-1} is the target output corresponding to the two-category classification problem. The goal of the support vector classifier is to construct a decision function (classification hyperplane) to classify the test data as correctly as possible, which is discussed below.

[0098] If there is a classification hyperplane:

[0099] Satisfy the constraints:

[0100] The training set is said to be linearly separable. represents the inner product of vectors, is an adjustable weight vector and b∈R is a bias.

[0101] According to statistical learning theory, if the training sample set is not separated by the hyperplane error, and the distance between the sample data closest to the hyperplane and the classification hyperplane is the largest, then the hyperplane is the optimal hyperplane (OptimalHyperplane), and the decision function obtained is:

[0102] Its promotion ability is the best, among which is a symbolic function.

[0103] Using the Lagrange optimization method and Wofle duality theory, the above problem can be transformed into its dual problem, that is, maximizing the functional:

[0104] Among them, αi is the Lagrange multiplier corresponding to sample i.

[0105] The above formula is a quadratic programming problem with inequality constraints. There is a unique solution, and only a small number of αi in the solution are not zero. The corresponding samples are support vectors. Solving the above problem to obtain the optimal classification function is:

[0106] Among them, nsv is the number of support vectors, and b is the classification threshold, which can be obtained by taking the median of any pair of support vectors in the two categories.

[0107] By introducing the relaxation term ξi, a generalized classification surface is realized to solve the problem that the training samples are linearly inseparable, that is, a compromise is made between the minimum misclassified samples and the maximum classification interval. For nonlinear problems, they can be transformed into linear problems in a certain high-dimensional space through nonlinear transformation, and then the optimal classification surface is sought in this high-dimensional space. Only the inner product operation (xi•xj) between samples is involved, so only the inner product operation needs to be performed in the high-dimensional space, and the inner product operation can be realized by the function in the original space. According to the Hilbert-schmidt principle, as long as the kernel function K(xi•xj) satisfies the Mercer condition, it corresponds to the inner product in a certain exchange space.

[0108] Therefore, by replacing the inner product in the formula with the kernel function K(xi•xj) that satisfies the Mercer condition, linear classification after a certain nonlinear transformation can be achieved, and the optimal classification function becomes: , In short, SVM maps the input vector to a high-dimensional feature space through some pre-selected nonlinear mapping, and constructs the optimal classification hyperplane in this feature space. In form, the SVM classification function is similar to a neural network, and the output is a linear combination of intermediate nodes. Each intermediate node corresponds to a support vector. The number of intermediate nodes in a neural network must be selected based on experience or comparative experiments. Different algorithms will have a great impact on network performance; while the number of intermediate nodes of SVM is automatically determined by calculation.

[0109] Vapnik generalized the results obtained in SVM classification theory by introducing an insensitive loss function, making it applicable to function fitting. With this generalization, the SVM method becomes a general method for expressing functions in high-dimensional space. It can be used for various function estimation problems, including density estimation problems and solving linear operator equations. and , construct the following optimization problem:

[0110] Among them, the constant C is the penalty coefficient.

[0111] The optimization problem in the dual space is:

[0112] Among them, αi, αi* are Lagrange multipliers.

[0113] Solving this problem, we can obtain the optimal Lagrange multipliers αi, αi*, and thus obtain the fitting function:

[0114] Taking into account the high dimension of actual cost information, the numerous and interrelated interference factors, and the nonlinear properties between input variables that will affect the prediction accuracy and generalization ability of the support vector machine learning model, we first use nonlinear kernel principal component analysis to compress the input variables and extract information, eliminate the correlation between the input variables, and extract the principal components to lay a data foundation for support vector machine training.

[0115] Prediction accuracy of support vector machine and parameter penalty factor C, loss function parameters , the selected kernel function and kernel function width coefficient Therefore, the genetic algorithm is used to find the optimal value of each parameter of the support vector machine within a certain range of values, and then obtain the best prediction performance.

[0116] Determine the parameter C. and radial basis kernel function The value range of The value range of is [0, 0.08], the value range of penalty factor C is [0.0001, 500], and the width coefficient of radial basis kernel function is =[0.01,5].

[0117] Define the fitness function and apply the relative mean square error as the objective function:

[0118] Where αi is the actual value, fi is the test value, and N is the number of samples in the training set. Determine the appropriate error requirement between the actual value and the predicted value.

[0119] When evaluating the objective function, we first use the training sample data to establish a prediction model, predict the test sample data, and use the objective function to measure the quality of the learning algorithm. The smaller the error, the better the individual performance.

[0120] Binary encode the above parameters and randomly generate the initial population; Decode the chromosomes in the population and calculate the objective function corresponding to each individual; Determine whether the group performance meets the optimization criteria. If so, output the optimal parameters. Otherwise, according to the genetic strategy, apply selection, crossover and mutation operators to the group to generate the next generation of groups and start a new generation of inheritance. Check the objective function obtained through learning. If the selected parameter values ​​meet the MAPE error requirements, continue with binary encoding.

[0121] Finally, the optimal parameter value is assigned to the predictive learning algorithm, trained with the training sample set, and then the test sample is predicted to obtain the corresponding prediction result.

[0122] Given a set of samples x1, x2, ..., xn, x1, x2, ..., xn is a set of influencing factors. The nonlinear mapping Φ is used to map the original data xi to the space F. The image of the original data xi in the mapping space F is Φ(xk), and its mean is assumed to be 0, that is, , then the mapping data covariance matrix can be expressed as:

[0123] Perform eigenvector analysis on the feature space covariance matrix C: λv=Cv Where the eigenvalue λ≥0, the inner product of the projection vector and the formula is obtained:

[0124] Solving the equation yields the eigenvector v corresponding to the non-zero eigenvalue. Its solution must be in the space Φ(x1),…,Φ(xn), so v can be expressed as:

[0125] get

[0126] Where: k=1,…,n.

[0127] Assume an n×n matrix in, .

[0128] It can be simplified to nλα=Kα where a=[a1, a2,…,an]T Solving the equation, we can get the kth eigenvector ak of matrix K. The eigenvector ak can determine the principal element direction vk of the feature space. The eigenvector a of matrix K can be used to obtain the eigenvector v of C, that is, the principal element direction of the mapping space:

[0129] Where k is the number of principal components. The projection of a point x in the feature space to the kth core principal component vk is:

[0130] In space x, it is represented as The selection of kernel function form has an important influence on the analysis of nonlinear systems. Kernel functions have different forms. This application uses radial basis kernel function: .

[0131] From the perspective of minimizing the redundancy of the principal component while ensuring the integrity of the extracted principal component features, the following fitness function is defined:

[0132] in is the cumulative variance contribution rate, which indicates the proportion of the characteristic information extracted by the m principal components to the original variable information. λi is the eigenvalue corresponding to the i-th eigenvector. From the fitness function, we can see that when the cumulative variance contribution rate meets the given requirements, the smaller m is, the smaller CPV is. Therefore, by finding an optimal kernel parameter c, the fitness function CPV is minimized, thereby ensuring the best effect of kernel principal component analysis.

[0133] Set the key elements of the same type of scheme, identify the historical engineering information of the same type as the scheme through similar engineering search methods, and obtain the cost reference standard of the project through data analysis.

[0134] Specifically, we first need to construct an initial population, where each individual represents a set of SVM parameters. Then, we use the fitness function to evaluate the performance of each individual, and usually use the cross-validation method to measure the classification accuracy of the model. Next, we generate a new population through selection, crossover, and mutation operations, where high-quality individuals are retained and low-quality individuals are eliminated.

[0135] After multiple iterations, the genetic algorithm can gradually approach the optimal parameter configuration, thereby improving the classification accuracy and generalization ability of the SVM model. Ultimately, the optimized SVM model can effectively process data with nonlinear characteristics, enabling accurate decision support in application scenarios such as power transmission and transformation project cost prediction.

[0136] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0137] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0138] Those of ordinary skill 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 to be beyond the scope of this application.

[0139] 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.

[0140] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0141] 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 protection scope of the present invention.

Claims

1. An intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics, characterized in that: The steps include: Collect historical cost data, pre-process the data based on NACEMD and identify the factors affecting the cost of power transmission and transformation projects. The historical cost data includes the type of project, construction plan, construction environment, and the cost of various expenses; Correlation analysis is used to determine the main factors affecting the cost, and cluster analysis is performed on different types of projects to obtain a sample set of power transmission and transformation projects of different categories; Based on the historical cost data and the sample set of power transmission and transformation projects, the cost ranges of different types of power transmission and transformation projects are calculated by using the Chevshcherby inequality, and the fluctuation ranges of different types of costs are determined, which are divided into three parts: stable trend, orderly fluctuation and disordered oscillation. Set corresponding intelligent methods for different types of expenses. For the disordered oscillation part, build a GRU prediction model based on improved particle swarm optimization. For the stable trend part, build a Markov Fourier modified grey prediction model. For the orderly fluctuation part, build a support vector machine model based on genetic algorithm optimization. Based on the forecast analysis of three types of costs and through the input of relevant project parameters, the evaluation criteria for the transmission and transformation project cost are determined.

2. According to claim 1, a method for intelligent evaluation of power transmission and transformation project cost based on cost fluctuation characteristics is characterized in that: The collection of historical cost data and preprocessing of the data based on NACEMD are specifically as follows: Execute M initialization CEMDs, adding white noise with amplitude k for each time; Adding the white noise to the original data sequence of the historical construction cost data to obtain a complex signal; Projecting the complex signal onto φk, where φk is the projection direction; When the sine value of φk is not zero, find the maximum point and perform cubic spline interpolation on it to obtain the upper and lower envelopes in each projection direction; Calculating the average value of the boundary envelope, and obtaining each component signal according to the average value of the boundary envelope; The mean of the IMF components and residuals are calculated; The mean of each IMF component obtained after M CEMD processes is taken as the output.

3. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 2 is characterized in that: The correlation analysis is adopted to determine the main factors affecting the construction cost, and cluster analysis is performed on different types of projects to obtain different types of power transmission and transformation project sample sets, specifically: According to the identified factors affecting the cost of power transmission and transformation projects, the correlation coefficient between each factor and the project cost is calculated based on the Pearson correlation coefficient; Based on the calculated correlation coefficient, explain the correlation between each factor and the project cost; According to the interpretation results of the correlation coefficient, the main factors that have a significant impact on the project cost are screened out; The screened main influencing factors were used as input variables for cluster analysis; Take the input variable as a sample and select the maximum and minimum values, perform standardization based on the preset formula, and obtain the standardized sample data; The collection of samples with similar characteristics and attributes in the sample data is regarded as a cluster; The average distance between the sample and other points in the cluster is taken as the cohesion within the cluster; The average distance between a sample and all points in the nearest cluster is considered as the separation from the nearest cluster; The silhouette coefficient is obtained by dividing the difference between the separation degree of the cluster and the cohesion degree of the cluster by the larger number of the two; The clustering effect is evaluated according to the value range of the silhouette coefficient; the closer the silhouette coefficient is to 1, the better the clustering effect is; the closer it is to -1, the worse the clustering effect is; A value close to 0 indicates that the sample is at the boundary of two clustering results. If the silhouette coefficient is low, adjust the number of clusters, re-perform cluster analysis, and calculate a new silhouette coefficient; By comparing the silhouette coefficients under different cluster numbers, find the clustering scheme with the largest silhouette coefficient; According to the best clustering scheme, the clustering results are sorted out to obtain a sample set of power transmission and transformation projects of different categories.

4. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 3 is characterized in that: Based on the historical cost data and the sample set of power transmission and transformation projects, the cost ranges of different types of power transmission and transformation projects are calculated by using the Chevshcherby inequality for different types of costs, which are specifically: Calculate the median, upper quartile, lower quartile, maximum value, minimum value, sample mean and sample variance of the sample data; A box plot is used to remove outliers to perform an outlier analysis on the sample data. The outlier is smaller than the lower quartile minus 1.5 times the interquartile range and larger than the upper quartile plus 1.5 times the interquartile range. A confidence level is preset, and based on the Chebyshev inequality, the cost range of each cost type at the confidence level is calculated.

5. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 4 is characterized in that: For the disordered oscillation part, an improved particle swarm optimized GRU prediction model is constructed, specifically: The GRU network structure is divided into two parts: training and prediction. The prediction part includes three parts: input layer, hidden layer and output layer. For a given training set, initialize several parameter vectors of dimension E; According to the sample loss, the gradient descent algorithm is used to train the GRU network to minimize the sample loss function value and update the weights of each network parameter; The initial weight matrix and parameter vector of GRU are used to form individual particle swarms. For each individual in the swarm, each gating unit, candidate state, and network output are calculated. Calculate the GRU prediction result corresponding to each particle, take the root mean square error as the fitness function of the IPSO algorithm, and optimize the particle corresponding to the GRU with the smallest root mean square error; Update the group optimum and individual optimum and calculate the aggregation degree; When the aggregation reaches a threshold, the velocity and position of the particles are updated based on the new dynamic topology and particle search behavior; After the particle swarm termination iteration condition is reached, the initialization process of the GRU model is completed; After completing the GRU initialization process, the GRU model is constructed and trained for each module separately, and finally the prediction results of each component are integrated through the new GRU network.

6. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 5 is characterized in that: For the ordered fluctuation part, a support vector machine model based on genetic algorithm optimization is constructed, specifically: Define the fitness function and apply the relative mean square error as the objective function: Establish a prediction model based on the training sample data, predict the test sample data, and use the objective function to measure the quality of the learning algorithm; Binary encode the model parameters and randomly generate the initial population; Decode the chromosomes in the population and calculate the objective function corresponding to each individual; Determine whether the group performance meets the optimization criteria. If it does, output the optimal parameters. Otherwise, according to the genetic strategy, apply selection, crossover and mutation operators to the group to generate the next generation of groups and start a new generation of genetics. Check the learned objective function. If the selected parameter value meets the MAPE error requirement, continue with binary encoding. The obtained optimal parameter value is assigned to the predictive learning algorithm, trained with the training sample set, and then the test sample is predicted to obtain the corresponding prediction result.

7. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 6 is characterized in that: The optimal parameter support vector machine model is specifically: Among them, nsv is the number of support vectors, and b is the classification threshold, which can be obtained by taking the median of any pair of support vectors in the two categories.

8. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 7 is characterized in that: For the stable trend part, a Markov Fourier modified grey prediction model is constructed, specifically: A sequence of n nodes, Perform an accumulation to obtain the cumulative sequence: in, Because the sequence With gray finger distribution, construct its first-order differential equation: right The sequence is averaged and the sequence is generated , in, Transform the differential equation into in Based on the grey prediction method, the parameter vector is estimated using the least squares method. in, Thus, the estimated parameters a and b are obtained and the grey prediction equation is constructed: The predicted value of the stable trend cost type is obtained by cumulative subtraction; According to the above formula, the stable trend cost type forecast sequence is: Use the residual formula to calculate the residual and get the residual sequence The residual correction is performed using Fourier series to obtain the Markov Fourier corrected grey prediction model.

9. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 8 is characterized in that: The residual correction using Fourier series is specifically as follows: Convert the residual sequence into a Fourier series sequence; Utilization Calculate each parameter and obtain the parameter estimation value of the Fourier series sequence; After correction by Fourier series, the predicted value of the stable trend fee type is modified to: Perform state division of the Markov chain and divide the data into n states according to the historical data of the stable trend fee type; in , the number of state partitions n and It can be determined based on the original data. According to the state transition probability formula, calculate the transition probability between the divided states , and then construct the state transition probability matrix P; in, Indicates the relevant parameters from After k steps, the state is transferred to The probability value of the state. The relevant parameters representing the data are represented by After k steps, the state is transferred to The original number of samples of the state, Indicates that The raw number of samples for the state. According to the obtained state transition probability matrix, the cost level of future stable trend cost is predicted according to the state transition matrix and formula; 。 10. The intelligent evaluation method for power transmission and transformation project cost based on cost fluctuation characteristics according to claim 9 is characterized in that: The weights of the network parameters are specifically: The phase point of the phase space reconstruction matrix is ​​used as the input part x of the network, and the sequence As the output part , is the hidden layer, is the sigmoid activation function, They are the update gate and reset gate of the recurrent network, is the weight between the input layer and the update gate at time t, is the weight between the hidden layer at time t-1 and the update gate at time t, is the weight between the input layer and the reset gate at time t, is the weight between the hidden layer at time t-1 and the reset gate at time t, is the weight between the hidden layer and the output layer at time t, is the Hadamard product of matrices.

Images