Compressor data asset value evaluation method based on optimization algorithm

Abstract

The invention provides a compressor data asset value evaluation method based on an optimization algorithm. The compressor data asset value evaluation method comprises the following steps: S1, determining a compressor data standard, quality, an application scene and an influence mechanism on an operation mode and cost reduction and benefit improvement; s2, on the basis of key parameters such as the pressure, the flow and the temperature collected in the S1, abnormal value elimination and standardization preprocessing are carried out; s3, constructing a nonlinear correlation model of unit consumption, efficiency and power consumption by adopting a multi-kernel weighted regularization partial least squares-radial basis kernel coupling algorithm, and determining dynamic influence weights of the parameters; s4, extracting surge characteristic components through a multi-scale wavelet packet decomposition-Hilbert-Huang transform coupling surge analysis algorithm, and analyzing the influence rule of the surge characteristic components on energy consumption, yield and load; and S5, in combination with the model in S3, the rule in S4, the energy efficiency attenuation coefficient and other data, calculating the energy consumption difference before and after data application by adopting an improved particle swarm gradient descent hybrid optimization algorithm, and determining the data asset value. Accurate evaluation of the compressor data asset value is realized, and a basis is provided for cost reduction and benefit increase.

Description

Technical Field

[0001] This invention relates to the field of compressor data evaluation, and in particular to a method for evaluating the value of compressor data assets based on optimization algorithms. Background Technology

[0002] Compressors, as key equipment in the industrial sector, are widely used in various industrial processes. Their operation relies on external power to meet the different pressure gas requirements of various industries, and they are also classified as high-energy-consuming equipment. During long-term operation, compressors continuously generate multi-dimensional operational data covering pressure, flow rate, temperature, vibration, and other dimensions. With the development of industrial data analysis technology, by mining and applying this operational data, it is possible to discover optimization opportunities in the compressor's operation, thereby achieving the goals of improving compressor operating efficiency, increasing output, and reducing operating costs.

[0003] Furthermore, compressor operating data not only supports the optimization of individual units, but the cost reduction and efficiency improvement solutions derived from this data can also provide direct guidance for compressor manufacturers to optimize product design and for compressor users to improve their operation and maintenance strategies, thus bringing clear economic value to relevant stakeholders. However, the industry currently lacks a systematic and scientific method to link the cost reduction and efficiency improvement results brought by compressor operating data with the value of the data itself. This makes it impossible to effectively measure the actual value of compressor data as an asset. This situation results in a lack of clear basis for the management, evaluation, and subsequent application of compressor data assets, hindering the further release of the data's value in the compressor field. Summary of the Invention

[0004] The main objective of this invention is to provide a method for evaluating the value of compressor data assets based on optimization algorithms. This method solves the technical problem that the industry currently lacks a systematic and scientific approach, making it impossible to correlate the cost reduction and efficiency improvement results brought about by compressor operation data with the value of the data itself, and thus impossible to effectively measure the actual value of compressor data assets.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for evaluating the value of compressor data assets based on optimization algorithms, the method comprising: S1. Determine the compressor data standards, data quality, data application scenarios, and the impact mechanism of data on compressor operation mode and cost reduction and efficiency improvement; S2. Based on the compressor data determined in S1, and combined with the data points and instrument configuration, collect key parameters of compressor pressure, flow rate, and temperature, and perform outlier removal and standardization preprocessing on the collected data; S3. A multi-kernel weighted regularized partial least squares-radial basis kernel coupling algorithm is adopted to perform feature mapping and correlation modeling on the parameters preprocessed in S2, construct a nonlinear correlation model of compressor unit consumption, efficiency and power consumption, and determine the dynamic influence weight of each parameter on operating performance. S4. A multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm is adopted to decompose the vibration signals of each stage of the compressor in the time and frequency domain, extract surge characteristic components, and analyze the coupling effect of surge amplitude on energy consumption and the dynamic effect of surge margin on output and load. S5. Based on the nonlinear correlation model of S3 and the surge law of S4, combined with the compressor energy efficiency attenuation coefficient, capacity fluctuation coefficient and loss quantification data, an improved particle swarm optimization-gradient descent hybrid optimization algorithm is used to calculate the energy consumption difference before and after the application of compressor data, and to determine the value of data assets.

[0006] In the preferred scheme, when constructing the nonlinear correlation model using the multi-kernel weighted regularized partial least squares-radial basis kernel coupling algorithm in S3, the parameters are first mapped to high-dimensional features using the radial basis kernel function. The mapping formula is as follows: ; in, Let be the parameter vectors at the i-th and j-th sampling times, respectively. The radial base core width parameter, The kernel matrix, calculated from the radial basis function kernel, is used to describe the two parameter vectors. Similarity between them; then introduce a multi-core weighting factor ( Construct a multi-core matrix: ; in, For polynomial kernel functions, The kernel matrix, calculated using a polynomial kernel function, is also used to describe the similarity between parameter vectors. For scale parameters, For offset parameters, Let the degree be a polynomial. A multi-kernel matrix is ​​used to integrate radial basis function kernels and polynomial kernels to improve the comprehensiveness of feature mapping; finally, an L2 regularization term is added to construct the objective function and solve for the regression coefficients. ; in, For the regression coefficient vector, As slack variables, For regularization parameters, For penalty parameters, Let be the actual value of the compressor efficiency at the i-th sampling time; after solving, the nonlinear correlation model is obtained: ; in, This is a multi-kernel matrix for the test and training sets, used to establish the correlation between the test set parameters and the training set model. This is the optimal regression coefficient vector. This is the predicted value for compressor efficiency. For the actual compressor efficiency, the model prediction error must meet the following requirements: ; The nonlinear correlation model accurately quantifies the mapping relationship between parameters and compressor efficiency, providing core model support for subsequent calculations of efficiency differences before and after data application, and thus assessing the value of data assets.

[0007] In the preferred scheme, when determining the dynamic influence weights of parameters in S3, the contribution of each parameter is calculated based on the eigenvalue decomposition of the kernel matrix. The contribution formula is as follows: ; in, This is the k-th eigenvalue of the multi-kernel matrix, used to reflect the importance of the corresponding dimension of the kernel matrix. Let j be the j-th component of the k-th feature vector, used to describe the distribution of the j-th parameter along the k-th feature dimension. The number of non-zero eigenvalues. Let j be the contribution of the j-th parameter; the dynamic weight update formula is: ; in, Let j be the weight of the j-th parameter at time t. The weight at time t-1, This is used to update the weighting coefficients, ensuring that the weights are dynamically adjusted according to changes in operating conditions. This contribution calculation clarifies the priority of each parameter's impact on compressor efficiency.

[0008] In the preferred scheme, when processing vibration signals using the multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm in S4, the vibration acceleration signal is first analyzed. Wavelet packet decomposition is performed with 4 decomposition levels. The db8 wavelet basis function is used, and the wavelet packet coefficient reconstruction formula for each node is as follows: ; in, This represents the reconstruction signal of the nth node in the mth layer. These are the corresponding wavelet packet coefficients. For wavelet packet basis functions, The sampling period is specified; empirical mode decomposition is performed on the reconstructed signals of each layer to obtain the intrinsic mode functions. And then Perform Hilbert transform: ; in, for The Hilbert transform result, Represents the principal value of the integral. For the integration variable; construct the analytic signal ,in The imaginary unit is used to extract the instantaneous frequency. With instantaneous amplitude ,in To analyze the signal The phase angle; The intrinsic mode functions with instantaneous frequencies in the range of [10, 500 Hz] and instantaneous amplitude variation coefficients greater than 0.3 are identified as surge characteristic components. The signal processing process accurately extracts surge-related features, providing a reliable data foundation for subsequent quantification of the impact of surge on energy consumption and output, realizing the key link of "from surge data to value loss assessment".

[0009] In the preferred scheme, when analyzing the coupled effect of surge amplitude on energy consumption in S4, a surge amplitude index is constructed based on the instantaneous amplitude of the surge characteristic components. Establish a coupling equation between energy consumption and surge amplitude, operating pressure, and flow rate: ; in, Energy consumption per unit time For average operating pressure, For average flow rate, For surge amplitude index, The coupling coefficient is used to quantify the impact of each variable on energy consumption; a genetic algorithm is used to fit and solve the historical data, and the fitting objective function is: ; in, The number of historical data samples. Let i be the actual energy consumption of the i-th sample. Let be the surge amplitude index for the i-th sample. Let be the average operating pressure of the i-th sample. To predict energy consumption using the model, the sum of squared residuals after fitting must satisfy... The coupled equation quantifies the relationship between surge amplitude and energy consumption. By calculating the energy consumption reduction after "eliminating / suppressing surge", it is directly converted into the cost reduction benefit of data assets, which is an important quantitative basis for the valuation of data assets.

[0010] In the preferred scheme, when analyzing the dynamic effect of surge margin on output and load in S4, surge margin is defined. ,in The surge critical pressure, To determine the actual operating pressure, a dynamic correlation equation is constructed: ; ; in, Let be the output at time t. For the rated output, Let be the surge margin at time t. This is the output impact coefficient. The attenuation coefficient is... The angular frequency of the wave. This is the initial phase. For integration variables; Let t be the load at time t. For rated load, This is the load influence factor. The rate of change of surge margin is used; the parameters of the dynamic equation are calibrated by least squares support vector machine, and the prediction error is less than 2% after calibration; the dynamic correlation equation can calculate the output increase brought about by "optimizing surge margin", which is transformed into the efficiency benefit of data assets, and together with the cost reduction benefit, constitutes the comprehensive value of data assets.

[0011] In the preferred scheme, when calculating the energy consumption difference using the improved particle swarm optimization-gradient descent hybrid optimization algorithm in S5, the efficiency output by the S3 correlation model is used first. Energy consumption of the output of the S4 coupled equation For the target variable, construct the optimization objective function: ; in, Due to differences in energy consumption, Energy consumption before and after data application, For compressor efficiency after data application, For optimal efficiency. For output after data application, To achieve optimal output, These are weighting coefficients used to balance the importance of different objectives; particle positions are initialized during the particle swarm optimization phase. With speed The updated formula is: ; ; in, The velocities of the i-th particle in the j-th dimension at time t+1 and t are respectively. For inertial weights, As a learning factor, A random number in the range [0,1]. Let be the optimal position of the i-th particle in the j-th dimension. Let be the globally optimal position in the j-th dimension. Let be the positions of the i-th particle in the j-th dimension at time t+1 and t, respectively; when the number of particle swarm optimization iterations reaches 60% of the total number of iterations, switch to the gradient descent algorithm, and the gradient update formula is: ; in, For learning rate, For the objective function in The gradient at point is iterated until the objective function value is less than 1. Stop at the appropriate time and output the optimal value. .

[0012] In the preferred scheme, outlier removal in S2 employs a method based on the local outlier factor, calculating the local outlier factor for each data point. The formula is: ; in, For the number of nearest neighbors, For the data points to be evaluated, for Nearest neighbor data points, for The set of k nearest neighbors, for arrive The reachable distance, for arrive The reachable distance, for and Euclidean distance, for The distance to its k-th nearest neighbor; when Values ​​identified as outliers are removed; standardization preprocessing uses Z-score standardization, with the formula: ; in, This represents the value of the j-th parameter at the i-th sampling time after standardization. This represents the original value of the j-th parameter at the i-th sampling time before standardization. Let j be the mean of the j-th parameter. Let be the standard deviation of the j-th parameter; the data preprocessing process can remove invalid interference data and unify the parameter dimensions, ensuring the data quality of subsequent S3 modeling and S4 analysis, and avoiding deviations in value assessment results caused by abnormal data.

[0013] In the preferred scheme, the compressor energy efficiency attenuation coefficient in S5 Calculate using the following formula: ; in, For the evaluation period, For the efficiency at time t, For initial efficiency, The integral value of efficiency over the evaluation period; capacity fluctuation coefficient. The calculation is as follows: ; in, For the rated output, Let be the output at time t. To assess average yield over the evaluation period; loss quantification data Including heat loss Mechanical loss Leakage loss The calculation formula is: ; ; ; in, For heat dissipation coefficient, For heat dissipation area, The average temperature. The ambient temperature; For mechanical loss coefficient, For average operating pressure, Average flow rate; Leakage coefficient, The leakage area; Total loss; By calculating the energy efficiency decay coefficient, the capacity fluctuation coefficient, and loss quantification data, the calculation results of energy consumption differences can be corrected, and the interference of non-data factors on value assessment can be eliminated.

[0014] In the preferred scheme, when determining the value of data assets in S5, it is based on differences in energy consumption. The unit price of energy type Annual operating time of equipment Equipment depreciation factor Construct a value calculation formula:

[0015] in, For the value of data assets, Annual cost of the data acquisition system The annual maintenance cost of the algorithm model; when the energy type is steam, It needs to be converted into equivalent electrical energy. The conversion formula is:

[0016] in, For the difference in equivalent electrical energy consumption, Due to differences in steam consumption, This refers to the enthalpy of vapor. For vapor density, For boiler efficiency, For turbine efficiency; the value calculation result is rounded to two decimal places; The value calculation formula transforms "energy consumption difference" into "annual net income" and deducts data-related costs to obtain the actual value of data assets.

[0017] This invention provides a method for evaluating the value of compressor data assets based on optimization algorithms. This method effectively addresses the lack of a systematic approach to compressor data asset valuation by identifying core elements of compressor data, collecting key parameters, and constructing correlation models and analyzing surge effects using specific optimization algorithms. It can accurately quantify the correlation between compressor data and operating efficiency, energy consumption, and output, clearly defining the actual role of data in cost reduction and efficiency improvement, and providing a scientific basis for measuring the value of compressor data assets. It not only provides direction for screening high-value data and optimizing data collection priorities, but also transforms the data optimization effect into quantifiable economic benefits by calculating the difference in energy consumption before and after data application, thus clarifying the actual value of data assets. Simultaneously, this method eliminates interference from non-data factors in the valuation, ensuring that the valuation results only reflect the benefits brought by data application. This provides reliable data value references for compressor manufacturers to optimize product design and for users to improve operation and maintenance strategies, helping to promote the effective application and value release of data in the compressor field, and further improving the scientific and economic aspects of compressor operation management. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the data asset valuation method of the present invention; Figure 2 The evaluation results of this invention are shown in the data system page diagram; Figure 3 This is the state diagram of the compressor data asset valuation algorithm of the present invention. Detailed Implementation

[0019] Example 1 like Figure 1-3 As shown, a method for evaluating the value of compressor data assets based on an optimization algorithm is proposed. This method includes: S1. Determine the compressor data standards, data quality, data application scenarios, and the impact mechanism of data on compressor operation mode and cost reduction and efficiency improvement; S2. Based on the compressor data determined in S1, and combined with the data points and instrument configuration, collect key parameters of compressor pressure, flow rate, and temperature, and perform outlier removal and standardization preprocessing on the collected data; S3. A multi-kernel weighted regularized partial least squares-radial basis kernel coupling algorithm is adopted to perform feature mapping and correlation modeling on the parameters preprocessed in S2, construct a nonlinear correlation model of compressor unit consumption, efficiency and power consumption, and determine the dynamic influence weight of each parameter on operating performance. S4. A multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm is adopted to decompose the vibration signals of each stage of the compressor in the time and frequency domain, extract surge characteristic components, and analyze the coupling effect of surge amplitude on energy consumption and the dynamic effect of surge margin on output and load. S5. Based on the nonlinear correlation model of S3 and the surge law of S4, combined with the compressor energy efficiency attenuation coefficient, capacity fluctuation coefficient and loss quantification data, an improved particle swarm optimization-gradient descent hybrid optimization algorithm is used to calculate the energy consumption difference before and after the application of compressor data, and to determine the value of data assets.

[0020] It clarifies the entire process from defining the basic elements of compressor data to determining the final value of data assets, connecting key links such as data processing, algorithm modeling, impact analysis, and value calculation. It provides overall guidance for the subsequent detailed development of each dependent claim, ensuring the completeness and logical coherence of the assessment methodology system.

[0021] Breaking away from the traditional limitation of using compressor data solely for equipment monitoring, this system directly links data with cost reduction and efficiency improvement benefits through multi-algorithm collaboration, establishing a systematic path for assessing the value of compressor data assets. This addresses the difficulty in quantifying data value within the industry and provides a standardized method for the management, application, and value mining of data assets.

[0022] Clearly defining the core elements of compressor data standards and quality is a prerequisite for subsequent data collection and analysis, ensuring data usability and consistency, and avoiding deviations in evaluation results due to inconsistent data foundations.

[0023] Key parameters such as pressure, flow rate, and temperature are collected and preprocessed to provide high-quality data input for algorithm modeling and reduce the interference of abnormal data on the accuracy of the model.

[0024] Three specific optimization algorithms are used to achieve parameter correlation modeling, surge impact analysis and energy consumption difference calculation, respectively, to complete the transformation from data to value in a progressive manner. Each algorithm addresses specific technical pain points in the evaluation process, and together they ensure the accuracy and scientific nature of the value assessment.

[0025] In the preferred scheme, when constructing the nonlinear correlation model using the multi-kernel weighted regularized partial least squares-radial basis kernel coupling algorithm in S3, the parameters are first mapped to high-dimensional features using the radial basis kernel function. The mapping formula is as follows: ; in, Let be the parameter vectors at the i-th and j-th sampling times, respectively. The radial base core width parameter, The kernel matrix, calculated from the radial basis function kernel, is used to describe the two parameter vectors. Similarity between them; then introduce a multi-core weighting factor ( Construct a multi-core matrix: ; in, For polynomial kernel functions, The kernel matrix, calculated using a polynomial kernel function, is also used to describe the similarity between parameter vectors. For scale parameters, For offset parameters, Let the degree be a polynomial. A multi-kernel matrix is ​​used to integrate radial basis function kernels and polynomial kernels to improve the comprehensiveness of feature mapping; finally, an L2 regularization term is added to construct the objective function and solve for the regression coefficients. ; in, For the regression coefficient vector, As slack variables, For regularization parameters, For penalty parameters, Let be the actual value of the compressor efficiency at the i-th sampling time; after solving, the nonlinear correlation model is obtained: ; in, This is a multi-kernel matrix for the test and training sets, used to establish the correlation between the test set parameters and the training set model. This is the optimal regression coefficient vector. This is the predicted value for compressor efficiency. For the actual compressor efficiency, the model prediction error must meet the following requirements: ; The nonlinear correlation model accurately quantifies the mapping relationship between parameters and compressor efficiency, providing core model support for subsequent calculations of efficiency differences before and after data application, and thus assessing the value of data assets.

[0026] The "Multi-core Weighted Regularized Partial Least Squares-Radial Basis Kernel Coupling Algorithm" of S3 clarifies the specific process of constructing a nonlinear correlation model, fills the gap in the core algorithm operation level, ensures that the model construction is reproducible and feasible, and provides reliable model support for subsequent efficiency difference calculations.

[0027] By leveraging the advantages of multi-kernel matrix fusion of radial basis kernel and polynomial kernel, the comprehensiveness of parameter feature mapping is enhanced. Furthermore, by combining L2 regularization to avoid model overfitting, the constructed nonlinear correlation model can accurately quantify the mapping relationship between parameters and compressor efficiency. The model prediction error is controlled within an extremely low range, ensuring the accuracy of subsequent value assessment results based on efficiency differences.

[0028] Radial basis kernel function mapping formula: ; Letter explanation: This is the parameter vector (containing parameters such as pressure, flow rate, and temperature) at the i and j sampling times. The radial base core width parameter, This is the radial basis function kernel matrix. It maps low-dimensional compressor operating parameters to a high-dimensional space, capturing the complex nonlinear relationships between parameters. This lays the foundation for building a precise correlation model and solves the problem of unclear linear relationships between parameters and efficiency in low-dimensional space.

[0029] Multi-core matrix construction formula: (in ); This is a multi-core weighting factor (the sum of the two is 1). It is a polynomial kernel matrix. For scale parameters, Here, d is the offset parameter, and d is the degree of the polynomial. This is the fused multi-core matrix.

[0030] Combining the advantages of two kernel functions, Skilled at capturing local features Skilled at capturing global features It can comprehensively reflect the similarity between parameter vectors, improve the integrity of feature mapping, and avoid feature omission problems caused by a single kernel function.

[0031] Objective function formula: ; For the regression coefficient vector, As slack variables, For regularization parameters, For penalty parameters, Let be the actual value of the compressor efficiency at the i-th sampling time, and n be the total number of samples.

[0032] pass The complexity of regression coefficients is controlled to avoid model overfitting; Item and The model is balanced between fitting accuracy and generalization ability to ensure good performance on both the training and test sets.

[0033] Formula for nonlinear correlation model: ; This is the predicted value for compressor efficiency. The multi-kernel matrix for the test set and training set. This is the optimal regression coefficient vector. This is the actual value of the compressor efficiency.

[0034] Directly outputting compressor efficiency prediction results is the core basis for calculating the efficiency difference before and after data application. and Error control ( This ensures accurate calculation of efficiency differences, thereby guaranteeing the reliability of data asset valuation results.

[0035] In the preferred scheme, when determining the dynamic influence weights of parameters in S3, the contribution of each parameter is calculated based on the eigenvalue decomposition of the kernel matrix. The contribution formula is as follows: ; in, This is the k-th eigenvalue of the multi-kernel matrix, used to reflect the importance of the corresponding dimension of the kernel matrix. Let j be the j-th component of the k-th feature vector, used to describe the distribution of the j-th parameter along the k-th feature dimension. The number of non-zero eigenvalues. Let j be the contribution of the j-th parameter; the dynamic weight update formula is: ; in, Let j be the weight of the j-th parameter at time t. The weight at time t-1, This is used to update the weighting coefficients, ensuring that the weights are dynamically adjusted according to changes in operating conditions. This contribution calculation clarifies the priority of each parameter's impact on compressor efficiency.

[0036] The process of "determining the dynamic influence weight of parameters" in S3 is refined, and the contribution calculation method based on kernel matrix eigenvalue decomposition and the dynamic weight update logic are clarified, so that the parameter importance assessment can be quantified and dynamically adjusted, providing a specific basis for data collection optimization and high-value data screening.

[0037] By calculating the contribution value, the priority of each parameter's impact on compressor efficiency is clearly defined, avoiding the blindness of data collection and enabling targeted improvement in the collection accuracy and frequency of high-contribution parameters. Dynamic weight updates ensure that the parameter importance assessment still reflects the actual operating conditions when the compressor's operating conditions change, guaranteeing the timeliness and accuracy of "data value and parameter importance correlation" in subsequent value assessments.

[0038] Contribution calculation formula: ; The k-th eigenvalue of the multi-kernel matrix. Let be the j-th component of the k-th eigenvector, and p be the number of non-zero eigenvalues. The contribution of the j-th parameter, where j is the parameter number (e.g., pressure, flow rate).

[0039] Reflecting the importance of the corresponding dimension of the kernel matrix, This reflects the distribution of the j-th parameter along that dimension, and the calculation is performed by combining both. It can quantify the impact of each parameter on efficiency, providing a direct basis for screening high-value data. For example, the "pressure" parameter, which has a high contribution, needs to be collected in a key manner to improve the cost-effectiveness of data collection.

[0040] Dynamic weight update formula: ; Let j be the weight of the j-th parameter at time t. The weight at time t-1, Update the coefficients for the weights.

[0041] Through historical weighting Compared with current contribution The weighted fusion achieves smooth updates of weights, avoids sudden changes in parameter weights due to abrupt changes in operating conditions, ensures stable and reliable parameter importance assessment, and provides a continuous and accurate basis for subsequent data value association.

[0042] In the preferred scheme, when processing vibration signals using the multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm in S4, the vibration acceleration signal is first analyzed. Wavelet packet decomposition is performed with 4 decomposition levels. The db8 wavelet basis function is used, and the wavelet packet coefficient reconstruction formula for each node is as follows: ; in, This represents the reconstruction signal of the nth node in the mth layer. These are the corresponding wavelet packet coefficients. For wavelet packet basis functions, The sampling period is specified; empirical mode decomposition is performed on the reconstructed signals of each layer to obtain the intrinsic mode functions. And then Perform Hilbert transform: ; in, for The Hilbert transform result, Represents the principal value of the integral. For the integration variable; construct the analytic signal ,in The imaginary unit is used to extract the instantaneous frequency. With instantaneous amplitude ,in To analyze the signal The phase angle; The intrinsic mode functions with instantaneous frequencies in the range of [10, 500 Hz] and instantaneous amplitude variation coefficients greater than 0.3 are identified as surge characteristic components. The signal processing process accurately extracts surge-related features, providing a reliable data foundation for subsequent quantification of the impact of surge on energy consumption and output, realizing the key link of "from surge data to value loss assessment".

[0043] The "multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm" of S4 is refined, clarifying the specific steps of vibration signal processing and the criteria for judging surge feature components. This solves the problem of difficulty in accurately extracting compressor surge features and provides high-quality feature data for subsequent quantification of the impact of surge on energy consumption and output.

[0044] By using multi-scale wavelet packet decomposition to achieve fine time-frequency domain decomposition of vibration signals, and combining Hilbert-Huang transform to extract instantaneous frequency and amplitude, the characteristic components related to surge can be accurately identified, other vibration interferences (such as normal mechanical vibration) can be eliminated, the purity of surge characteristic data can be ensured, reliable input can be provided for subsequent coupling effect analysis, and value assessment bias caused by inaccurate surge feature extraction can be avoided.

[0045] Wavelet packet coefficient reconstruction formula: ; This is the reconstruction signal of the nth node in the mth layer. These are wavelet packet coefficients. For wavelet packet basis functions, is the sampling period, and k is the discrete time index.

[0046] The decomposed wavelet packet coefficients are reconstructed into signals corresponding to the frequency bands, realizing multi-scale decomposition of vibration signals. Vibration components in different frequency ranges can be separated, laying the foundation for subsequent extraction of surge characteristics at specific frequencies and solving the problem of mixed frequencies in the original vibration signals and difficulty in distinguishing surge signals.

[0047] Hilbert transform formula: ; Letter explanation: For the k-th intrinsic modulus function The Hilbert transform result, Represents the principal value of the integral. Let t be the integral variable and t be the time variable.

[0048] The intrinsic mode functions obtained from empirical mode decomposition are transformed to provide imaginary components for constructing analytic signals, thereby extracting instantaneous frequency and amplitude, realizing dynamic capture of the time-frequency characteristics of surge signals, and solving the problem that traditional Fourier transform cannot handle non-stationary surge signals.

[0049] Analytical signal and instantaneous parameter extraction: , , ; Letter explanation: For analytic signals, j is the imaginary unit. Instantaneous frequency, To analyze the phase angle of the signal, This represents the instantaneous amplitude.

[0050] By analyzing the signal to simultaneously obtain the instantaneous frequency and amplitude, the time-varying characteristics of the surge signal can be dynamically described. Combined with the judgment criteria of "instantaneous frequency in the range of [10, 500 Hz] and instantaneous amplitude variation coefficient greater than 0.3", surge characteristic components can be accurately screened out, providing pure characteristic data for subsequent quantification of the impact of surge on energy consumption and output.

[0051] In the preferred scheme, when analyzing the coupled effect of surge amplitude on energy consumption in S4, a surge amplitude index is constructed based on the instantaneous amplitude of the surge characteristic components. Establish a coupling equation between energy consumption and surge amplitude, operating pressure, and flow rate: ; in, Energy consumption per unit time For average operating pressure, For average flow rate, For surge amplitude index, The coupling coefficient is used to quantify the impact of each variable on energy consumption; a genetic algorithm is used to fit and solve the historical data, and the fitting objective function is: ; in, The number of historical data samples. Let i be the actual energy consumption of the i-th sample. Let be the surge amplitude index for the i-th sample. Let be the average operating pressure of the i-th sample. To predict energy consumption using the model, the sum of squared residuals after fitting must satisfy... The coupled equation quantifies the relationship between surge amplitude and energy consumption. By calculating the energy consumption reduction after "eliminating / suppressing surge", it is directly converted into the cost reduction benefit of data assets, which is an important quantitative basis for the valuation of data assets.

[0052] This section explains the process of "analyzing the coupled impact of surge amplitude on energy consumption" in S4, clarifies the construction method of surge amplitude index and the establishment and solution logic of energy consumption coupling equation, transforms the abstract operating state of surge into a quantifiable energy consumption influencing factor, and provides a specific basis for calculating the cost reduction benefits of data assets.

[0053] By constructing a coupled equation that includes surge amplitude, operating pressure, and flow rate, the synergistic impact of these three factors on energy consumption is quantified, avoiding energy consumption assessment bias caused by single-factor analysis. By combining genetic algorithms to fit and solve historical data, the accuracy of equation parameters is ensured. The value of surge data can be directly converted into cost reduction benefits by calculating the energy consumption reduction after "eliminating / suppressing surge", making data asset value assessment more practical.

[0054] Surge amplitude index and coupling equation: , ; For surge amplitude index, Let E be the instantaneous amplitude of the surge characteristic component, and E be the energy consumption per unit time. For average operating pressure, For average flow rate, is the coupling coefficient.

[0055] By taking the maximum instantaneous amplitude, the severity of surge is intuitively reflected; the coupled equations consider the synergistic effect of surge amplitude with pressure and flow rate, comprehensively describing the impact of each factor on energy consumption, avoiding the limitations of single-variable analysis, and providing a model basis for accurately calculating energy loss caused by surge.

[0056] Formula for fitting the objective function: ; Letter explanation: N represents the number of historical data samples. Let i be the actual energy consumption of the i-th sample. Let be the surge amplitude index for the i-th sample. Let be the average operating pressure of the i-th sample. To predict energy consumption for the model.

[0057] With the objective of minimizing the sum of squared residuals between actual and predicted energy consumption, a genetic algorithm is used to solve for the optimal coupling coefficient. This ensures that the coupled equations can accurately fit historical energy consumption data; combined with the "residual sum of squares" The constraint of "" ensures the accuracy of the equation prediction, provides a reliable model for subsequent calculation of surge-related energy consumption reduction, and thus accurately assesses the cost reduction value of surge data.

[0058] In the preferred scheme, when analyzing the dynamic effect of surge margin on output and load in S4, surge margin is defined. ,in The surge critical pressure, To determine the actual operating pressure, a dynamic correlation equation is constructed: ; ; in, Let be the output at time t. For the rated output, Let be the surge margin at time t. This is the output impact coefficient. The attenuation coefficient is... The angular frequency of the wave. This is the initial phase. For integration variables; Let t be the load at time t. For rated load, This is the load influence factor. The rate of change of surge margin is used; the parameters of the dynamic equation are calibrated by least squares support vector machine, and the prediction error is less than 2% after calibration; the dynamic correlation equation can calculate the output increase brought about by "optimizing surge margin", which is transformed into the efficiency benefit of data assets, and together with the cost reduction benefit, constitutes the comprehensive value of data assets.

[0059] The explanation of "analyzing the dynamic effect of surge margin on output and load" in S4 clarifies the definition of surge margin and the construction and calibration logic of the dynamic correlation equation between output and load, directly linking surge margin with production indicators, and providing a specific path for calculating the benefits of data asset efficiency.

[0060] The time-varying impact of surge margin on output and load is quantified by dynamic correlation equations. The output equation introduces integral and sine terms to reflect the cumulative and fluctuating effects of surge margin, while the load equation introduces a rate of change term to reflect the impact of sudden changes in surge margin on load. Combined with parameter calibration of least squares support vector machine, the prediction error of the equation is ensured to be less than 2%, which can accurately calculate the output increase after "optimizing surge margin", transforming the value of surge data into efficiency gains and improving the comprehensive value assessment system of data assets.

[0061] Surge margin definition: ; M represents the surge margin. The surge critical pressure, This represents the actual operating pressure.

[0062] By using the ratio of the difference between the critical pressure and the actual pressure, the safety level of the compressor away from surge can be intuitively reflected, providing a quantifiable state indicator for subsequent analysis of its impact on production indicators, and solving the problem of the difficulty in concretizing the safety level of surge.

[0063] Production dynamic correlation equation: ; Let be the output at time t. For the rated output, Let be the surge margin at time t. This is the output impact coefficient. The attenuation coefficient is... The angular frequency of the wave. This is the initial phase. It is the integral variable.

[0064] This item reflects the immediate impact of surge margin. The term reflects the cumulative attenuation effect of surge margin. The three factors reflect the cyclical fluctuations in output, and their combination comprehensively describes the dynamic effect of surge margin on output, providing model support for accurately calculating output increases.

[0065] Load dynamic correlation equation: ; Let t be the load at time t. For rated load, This is the load influence factor. This represents the rate of change of surge margin.

[0066] remove In addition to the immediate impact of the item, This item reflects the impact of sudden changes in surge margin on the load. This item reflects the nonlinear effect of surge margin, which can accurately describe the load change pattern under different surge margin states, provide a basis for optimizing load distribution and improving production efficiency, and thus quantify the efficiency-enhancing value of surge data.

[0067] In the preferred scheme, when calculating the energy consumption difference using the improved particle swarm optimization-gradient descent hybrid optimization algorithm in S5, the efficiency output by the S3 correlation model is used first. Energy consumption of the output of the S4 coupled equation For the target variable, construct the optimization objective function: ; in, Due to differences in energy consumption, Energy consumption before and after data application, For compressor efficiency after data application, For optimal efficiency. For output after data application, To achieve optimal output, These are weighting coefficients used to balance the importance of different objectives; particle positions are initialized during the particle swarm optimization phase. With speed The updated formula is: ; ; in, The velocities of the i-th particle in the j-th dimension at time t+1 and t are respectively. For inertial weights, As a learning factor, A random number in the range [0,1]. Let be the optimal position of the i-th particle in the j-th dimension. Let be the globally optimal position in the j-th dimension. Let be the positions of the i-th particle in the j-th dimension at time t+1 and t, respectively; when the number of particle swarm optimization iterations reaches 60% of the total number of iterations, switch to the gradient descent algorithm, and the gradient update formula is: ; in, For learning rate, For the objective function in The gradient at point is iterated until the objective function value is less than 1. Stop at the appropriate time and output the optimal value. .

[0068] This paper explains the process of calculating energy consumption differences using the "improved particle swarm optimization-gradient descent hybrid optimization algorithm" in S5, clarifies the construction of the optimization objective function and the switching logic between the particle swarm optimization and gradient descent algorithms, addresses the limitations of a single algorithm in terms of optimization accuracy and speed, ensures accurate calculation of energy consumption differences, and provides core quantitative indicators for data asset value assessment.

[0069] By constructing a multi-objective optimization function that integrates energy consumption differences, efficiency, and output, the one-sidedness of value evaluation caused by single-objective optimization is avoided. The global optimization capability of the particle swarm optimization algorithm can quickly locate the optimal solution region, while the local optimization capability of the gradient descent algorithm can accurately converge to the optimal solution. Combining the two ensures optimization speed while controlling the objective function value within a certain range. Therefore, ensuring the accuracy of energy consumption difference calculations directly determines the accuracy of data asset valuation results.

[0070] Optimize the objective function formula:

[0071] Letter explanation: Due to differences in energy consumption, Energy consumption before and after data application, For compressor efficiency after data application, For optimal efficiency. For output after data application, To achieve optimal output, These are the weighting coefficients.

[0072] With the core objective of "approaching the actual difference in energy consumption", we also take into account efficiency and output to approach the optimal value, avoiding the single pursuit of energy consumption reduction while ignoring efficiency or output. Balancing the importance of different objectives and ensuring that optimization results align with the actual production needs of "equal emphasis on cost reduction and efficiency improvement" provides a comprehensive optimization basis for data asset value assessment.

[0073] Particle swarm velocity and position update formulas: , ; Let be the velocities of the i-th particle in the j-th dimension at times t+1 and t, respectively. For inertial weights, As a learning factor, A random number in the range [0,1]. Let be the optimal position of the i-th particle in the j-th dimension. The globally optimal position in the j-th dimension. These are the positions of the i-th particle in the j-th dimension at times t+1 and t, respectively.

[0074] Through inertia weight The ability to balance global exploration and local development of particles, learning factor Guide particles toward their individual and global optimal positions using random numbers. Increasing the randomness of the optimization process can quickly locate the optimal solution region for energy consumption differences, laying the foundation for accurate optimization in subsequent gradient descent algorithms.

[0075] Gradient descent update formula: ; For learning rate, For the objective function in The gradient at that point.

[0076] Within the optimal solution region located by the particle swarm optimization algorithm, the gradient direction guides the particle position update, achieving localized and precise optimization and compensating for the insufficient convergence accuracy of the particle swarm optimization algorithm near the optimal solution; iterating until the objective function value is less than... Stop at the appropriate time to ensure output. Precision provides reliable core quantitative indicators for calculating the value of data assets.

[0077] In the preferred scheme, outlier removal in S2 employs a method based on the local outlier factor, calculating the local outlier factor for each data point. The formula is: ; in, For the number of nearest neighbors, For the data points to be evaluated, for Nearest neighbor data points, for The set of k nearest neighbors, for arrive The reachable distance, for arrive The reachable distance, for and Euclidean distance, for The distance to its k-th nearest neighbor; when Values ​​identified as outliers are removed; standardization preprocessing uses Z-score standardization, with the formula: ; in, This represents the value of the j-th parameter at the i-th sampling time after standardization. This represents the original value of the j-th parameter at the i-th sampling time before standardization. Let j be the mean of the j-th parameter. Let be the standard deviation of the j-th parameter; the data preprocessing process can remove invalid interference data and unify the parameter dimensions, ensuring the data quality of subsequent S3 modeling and S4 analysis, and avoiding deviations in value assessment results caused by abnormal data.

[0078] Explaining the process of "outlier removal and standardization preprocessing" in S2, clarifying the outlier identification method based on local outlier factors and the Z-score standardization logic, ensures the quality of input algorithm parameters from the data source and avoids interference from poor data on subsequent modeling and analysis results, which is the foundation for the reliable operation of the entire evaluation method.

[0079] The local outlier method can identify outliers that deviate from the normal data distribution (such as abnormal pressure values ​​caused by sensor failure), thus avoiding their interference with model fitting. Z-score standardization can unify the numerical range of parameters with different dimensions (such as pressure unit MPa and flow unit m³ / h), eliminate parameter weight bias caused by differences in dimensions, and ensure that subsequent algorithm modeling and parameter analysis are based on high-quality, same-scale data, thereby guaranteeing the reliability of the evaluation results.

[0080] Local outlier formula: ; in ; Let be the local outlier factor of data point x, k be the number of nearest neighbors, and y be the nearest neighbor data points of x. Let x be the set of k nearest neighbors. Let x be the reachable distance from y. Let y be the distance from its k-th nearest neighbor. Let x be the Euclidean distance between x and y.

[0081] The degree of anomaly of a data point is quantified by calculating the ratio of its reachability distance to its nearest neighbors; when By identifying and removing outliers in real time, it is possible to effectively remove poor-quality data caused by sensor malfunctions, transient interference, etc., ensuring the quality of data input to subsequent algorithms and avoiding model fitting deviations caused by outliers.

[0082] Z-score normalization formula: ; This represents the value of the j-th parameter at the i-th sampling time after standardization. The original values ​​before standardization. Let j be the mean of the j-th parameter. Let be the standard deviation of the j-th parameter.

[0083] Parameters with different dimensions and numerical ranges (such as pressure values ​​typically ranging from 0 to 10 MPa and flow rates typically ranging from 0 to 1000 m³ / h) are transformed into standardized data with a mean of 0 and a standard deviation of 1. This eliminates the impact of dimensional differences on algorithm modeling. For example, it avoids over-weighting due to the large range of flow rate values, ensuring that each parameter has an equal basis of influence in the modeling process, and guaranteeing the fairness and accuracy of the model.

[0084] In the preferred scheme, the compressor energy efficiency attenuation coefficient in S5 Calculate using the following formula: ; in, For the evaluation period, For the efficiency at time t, For initial efficiency, The integral value of efficiency over the evaluation period; capacity fluctuation coefficient. The calculation is as follows: ; in, For the rated output, Let be the output at time t. To assess average yield over the evaluation period; loss quantification data Including heat loss Mechanical loss Leakage loss The calculation formula is: ; ; ; in, For heat dissipation coefficient, For heat dissipation area, The average temperature. The ambient temperature; For mechanical loss coefficient, For average operating pressure, Average flow rate; Leakage coefficient, The leakage area; Total loss; By calculating the energy efficiency decay coefficient, the capacity fluctuation coefficient, and loss quantification data, the calculation results of energy consumption differences can be corrected, and the interference of non-data factors on value assessment can be eliminated.

[0085] Explain the process in S5 of “combining compressor energy efficiency attenuation coefficient, capacity fluctuation coefficient, and loss quantification data”, clarify the calculation logic of the three types of coefficients / data, and correct the calculation results of energy consumption differences by quantifying the impact of non-data factors on energy consumption and output, so as to ensure that the data asset value assessment results only reflect the actual benefits of “data application” and avoid interference from non-data factors.

[0086] The energy efficiency degradation coefficient can quantify the efficiency decline caused by equipment aging, the capacity fluctuation coefficient can quantify the output fluctuation caused by changes in market demand, and the loss quantification data can quantify inherent losses such as heat dissipation and mechanical wear. The three types of data together constitute a quantitative system for the influence of non-data factors. By correcting for differences in energy consumption and eliminating factors unrelated to data such as equipment aging and market fluctuations, the final assessment of "data asset value" is more accurate and aligns with the core objective of "measuring data value by the benefits of data application".

[0087] Formula for energy efficiency degradation coefficient: ; The energy efficiency degradation coefficient is T, where T is the evaluation period. For the efficiency at time t, For initial efficiency, To evaluate the efficiency integral value within the evaluation period.

[0088] Quantify the degree of energy efficiency degradation of equipment due to non-data factors such as aging and wear during the evaluation period. For example, if This indicates that the 5% efficiency decline is caused by equipment aging and should be deducted from the total efficiency improvement to ensure that the remaining efficiency improvement is solely attributable to data applications. This corrects the calculation results of energy consumption differences and avoids including non-data benefits in the value of data assets.

[0089] Formula for capacity fluctuation coefficient: ; This is the capacity fluctuation coefficient. For the rated output, Let be the output at time t. To evaluate the average yield within the evaluation period, T represents the number of samples taken within the evaluation period.

[0090] Quantify the degree of production capacity fluctuations caused by non-data factors such as changes in market demand and fluctuations in raw material supply. For example, if This indicates that the 3% production fluctuation was caused by non-data factors, and this part needs to be deducted from the total production increase to ensure that the remaining production increase is solely attributable to data application, correct the efficiency gain calculation results, and ensure the accuracy of data asset value assessment.

[0091] Loss quantification formula: , , , ; Letter explanation: These are heat dissipation, mechanical, and leakage losses, respectively. For heat dissipation coefficient, For heat dissipation area, The average temperature. The ambient temperature; For mechanical loss coefficient, For average operating pressure, Average flow rate; Leakage coefficient, The leakage area; This represents the total loss.

[0092] Purpose: To quantify the unavoidable inherent losses during compressor operation (such as heat loss, which is caused by the physical characteristics of the equipment and is unrelated to the data), this portion of the inherent loss reduction (if any) needs to be deducted from the total energy consumption reduction to ensure that the remaining energy consumption reduction is solely attributable to data application, avoid misrepresenting changes in inherent losses as data asset value, and ensure the accuracy and fairness of the assessment results.

[0093] In the preferred scheme, when determining the value of data assets in S5, it is based on differences in energy consumption. The unit price of energy type Annual operating time of equipment Equipment depreciation factor Construct a value calculation formula:

[0094] in, For the value of data assets, Annual cost of the data acquisition system The annual maintenance cost of the algorithm model; when the energy type is steam, It needs to be converted into equivalent electrical energy. The conversion formula is:

[0095] in, For the difference in equivalent electrical energy consumption, Due to differences in steam consumption, This refers to the enthalpy of vapor. For vapor density, For boiler efficiency, For turbine efficiency; the value calculation result is rounded to two decimal places; The value calculation formula transforms "energy consumption difference" into "annual net income" and deducts data-related costs to obtain the actual value of data assets.

[0096] This section explains the process of "determining the value of data assets" in S5, clarifies the construction logic of the value calculation formula, including energy type adaptation (electricity / steam), cost deduction and value result processing, transforms the abstract "energy consumption difference" into a specific "annual net income", fully realizes the core objective of "assessing the value of data assets with cost reduction and efficiency improvement benefits", and makes the assessment results have practical economic significance and application value.

[0097] By distinguishing between electricity and steam as two energy types, and adapting to the energy consumption scenarios of different compressors, the value calculation is made in line with reality. By deducting the costs of data collection and model maintenance, the calculation avoids only calculating the revenue while ignoring the investment in data application, resulting in a more realistic "net revenue". Finally, the output is a specific value with two decimal places, so that the value of data assets has a clear economic quantification result, which can be directly used for data asset management, trading or investment decisions.

[0098] Formula for calculating the value of data assets:

[0099] V represents the value of data assets. Due to differences in energy consumption, For the annual operating time of the equipment, For energy unit price, This is the equipment depreciation factor. Annual cost of the data acquisition system The annual maintenance cost of the algorithm model.

[0100] Transforming "energy consumption differences" into annual economic benefits, among which... Calculate annual energy cost savings. Adjust the impact of equipment depreciation on revenue (e.g., appropriately reduce the revenue share of older equipment); after deducting data-related costs, obtain the annual net revenue of data assets, giving the value of data assets a clear economic quantification result, for example... A sum of tens of thousands of yuan can be directly used to measure the actual economic value of data assets and support management decisions for data assets.

[0101] Steam equivalent electrical energy conversion formula:

[0102] For the difference in equivalent electrical energy consumption, Due to differences in steam consumption, This refers to the enthalpy of vapor. For vapor density, For boiler efficiency, For turbine efficiency.

[0103] For compressors that use steam as energy, the difference in steam consumption is converted into a unit (kW·h) that is consistent with that of electrical energy, which solves the problem that different energy types cannot be directly compared and calculated. This ensures that regardless of whether the compressor uses electricity or steam, the difference in energy consumption and value can be calculated through a unified "equivalent electrical energy" caliber, thus guaranteeing the consistency and comparability of data asset value assessment for different types of compressors.

[0104] Example 2 Further explanation in conjunction with Example 1, such as Figure 1-3 The structure shown above, when actually deploying and using the compressor data asset valuation method based on the optimization algorithm, requires following these steps, along with corresponding software and modules: First, based on the compressor data standards, quality, and application scenarios specified in S1 of the solution, configure the data acquisition module in the industrial control software Siemens WinCC and Rockwell FactoryTalk. Combine the compressor data points and instruments (pressure, flow, and temperature sensors), and achieve real-time acquisition of the key parameters in S2 through the OPC UA / DA protocol. At the same time, in the data preprocessing module of the software, call the Python script based on the pandas library to perform outlier removal preprocessing using the local outlier factor algorithm and Z-score standardization.

[0105] Next, the preprocessed data is synchronized to the time-series databases InfluxDB and Prometheus via the data transmission module Kafka to meet the data storage and time-series requirements of the solution. Subsequently, in the algorithm deployment platforms TensorFlowServing and PyTorchServe, a multi-kernel weighted regularized partial least squares-radial basis kernel coupling algorithm module developed based on Python combined with scikit-learn and PyWavelets libraries is loaded to execute the parameter feature mapping and nonlinear correlation model construction in S3. At the same time, the parameter contribution calculation and dynamic weight update are implemented through the NumPy library.

[0106] Subsequently, a multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm module was loaded onto the same algorithm platform. Based on the PyWavelets and Hilbert-Huang-Transform libraries, the module performed time-frequency domain decomposition and surge feature component extraction in S4 on the vibration signals of each stage of the compressor in the time series database. Then, the coupled equations of surge amplitude and energy consumption, and the dynamic correlation equations of surge margin and output / load were constructed using Python combined with the scipy library. Finally, the genetic algorithm was used to complete the equation fitting.

[0107] Finally, an improved particle swarm optimization-gradient descent hybrid optimization algorithm module based on the pygmo and scipy.optimize libraries was loaded onto the algorithm platform. Combining the correlation model of S3, the surge law of S4, and the energy efficiency decay coefficient, capacity fluctuation coefficient, and loss quantification data in S5, the energy consumption difference was calculated using Python. Then, the data asset value was calculated using Excel or the Python pandas library according to the value calculation formula in the plan, combined with parameters such as energy unit price and annual equipment operating time. Finally, the evaluation results were displayed through the data visualization software Grafana to realize the asset value table and subsequent monitoring.

[0108] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for evaluating the value of compressor data assets based on optimization algorithms, characterized by: The method includes: S1. Determine the compressor data standards, data quality, data application scenarios, and the impact mechanism of data on compressor operation mode and cost reduction and efficiency improvement; S2. Based on the compressor data determined in S1, and combined with the data points and instrument configuration, collect key parameters of compressor pressure, flow rate, and temperature, and perform outlier removal and standardization preprocessing on the collected data; S3. A multi-kernel weighted regularized partial least squares-radial basis kernel coupling algorithm is adopted to perform feature mapping and correlation modeling on the parameters preprocessed in S2, construct a nonlinear correlation model of compressor unit consumption, efficiency and power consumption, and determine the dynamic influence weight of each parameter on operating performance. S4. A multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm is adopted to decompose the vibration signals of each stage of the compressor in the time and frequency domain, extract surge characteristic components, and analyze the coupling effect of surge amplitude on energy consumption and the dynamic effect of surge margin on output and load. S5. Based on the nonlinear correlation model of S3 and the surge law of S4, combined with the compressor energy efficiency attenuation coefficient, capacity fluctuation coefficient and loss quantification data, an improved particle swarm optimization-gradient descent hybrid optimization algorithm is used to calculate the energy consumption difference before and after the application of compressor data, and to determine the value of data assets.

2. The compressor data asset valuation method based on optimization algorithm according to claim 1, characterized in that: S3 When constructing a nonlinear correlation model using the multi-kernel weighted regularized partial least squares-radial basis kernel coupling algorithm, the parameters are first mapped to high-dimensional features using the radial basis kernel function. The mapping formula is as follows: ； in, Let be the parameter vectors at the i-th and j-th sampling times, respectively. The radial base core width parameter, The kernel matrix, calculated from the radial basis function kernel, is used to describe the two parameter vectors. Similarity between them; then introduce a multi-core weighting factor ( Construct a multi-core matrix: ； in, For polynomial kernel functions, The kernel matrix, calculated using a polynomial kernel function, is also used to describe the similarity between parameter vectors. For scale parameters, For offset parameters, Let the degree be a polynomial. A multi-kernel matrix is ​​used to integrate radial basis function kernels and polynomial kernels to improve the comprehensiveness of feature mapping; finally, an L2 regularization term is added to construct the objective function and solve for the regression coefficients. ； in, For the regression coefficient vector, As slack variables, For regularization parameters, For penalty parameters, Let be the actual value of the compressor efficiency at the i-th sampling time; after solving, the nonlinear correlation model is obtained: ； in, This is a multi-kernel matrix for the test and training sets, used to establish the correlation between the test set parameters and the training set model. This is the optimal regression coefficient vector. This is the predicted value for compressor efficiency. For the actual compressor efficiency, the model prediction error must meet the following requirements: ; The nonlinear correlation model accurately quantifies the mapping relationship between parameters and compressor efficiency, providing core model support for subsequent calculations of efficiency differences before and after data application, and thus assessing the value of data assets.

3. The compressor data asset valuation method based on optimization algorithm according to claim 1, characterized in that: S3 When determining the dynamic influence weights of parameters, the contribution of each parameter is calculated based on the eigenvalue decomposition of the kernel matrix. The contribution formula is as follows: ； in, This is the k-th eigenvalue of the multi-kernel matrix, used to reflect the importance of the corresponding dimension of the kernel matrix. Let j be the j-th component of the k-th feature vector, used to describe the distribution of the j-th parameter along the k-th feature dimension. The number of non-zero eigenvalues. Let j be the contribution of the j-th parameter; the dynamic weight update formula is: ； in, Let j be the weight of the j-th parameter at time t. The weight at time t-1, This is used to update the weighting coefficients, ensuring that the weights are dynamically adjusted according to changes in operating conditions. This contribution calculation clarifies the priority of each parameter's impact on compressor efficiency.

4. The method for evaluating the value of compressor data assets based on optimization algorithms according to claim 1, characterized in that: When processing vibration signals using the multi-scale wavelet packet decomposition-Hilbert-Huang transform coupled surge analysis algorithm in S4, the vibration acceleration signal is first analyzed. Wavelet packet decomposition is performed with 4 decomposition levels. The db8 wavelet basis function is used, and the wavelet packet coefficient reconstruction formula for each node is as follows: ； in, This represents the reconstruction signal of the nth node in the mth layer. These are the corresponding wavelet packet coefficients. For wavelet packet basis functions, The sampling period is specified; empirical mode decomposition is performed on the reconstructed signals of each layer to obtain the intrinsic mode functions. And then Perform Hilbert transform: ； in, for The Hilbert transform result, Represents the principal value of the integral. For the integration variable; construct the analytic signal ,in The imaginary unit is used to extract the instantaneous frequency. With instantaneous amplitude ,in To analyze the signal The phase angle; The intrinsic mode functions with instantaneous frequencies in the range of [10, 500 Hz] and instantaneous amplitude variation coefficients greater than 0.3 are identified as surge characteristic components. The signal processing process accurately extracts surge-related features, providing a reliable data foundation for subsequent quantification of the impact of surge on energy consumption and output, realizing the key link of "from surge data to value loss assessment".

5. The compressor data asset valuation method based on optimization algorithm according to claim 1, characterized in that: S4 When analyzing the coupled effect of surge amplitude on energy consumption, a surge amplitude index is constructed based on the instantaneous amplitude of the surge characteristic components. Establish a coupling equation between energy consumption and surge amplitude, operating pressure, and flow rate: ； in, Energy consumption per unit time To average operating pressure, For average flow rate, The surge amplitude index The coupling coefficient is used to quantify the impact of each variable on energy consumption; a genetic algorithm is used to fit and solve the historical data, and the fitting objective function is: ； in, The number of historical data samples. Let i be the actual energy consumption of the i-th sample. Let be the surge amplitude index for the i-th sample. Let be the average operating pressure of the i-th sample. To predict energy consumption using the model, the sum of squared residuals after fitting must satisfy... The coupled equation quantifies the relationship between surge amplitude and energy consumption. By calculating the energy consumption reduction after "eliminating / suppressing surge", it is directly converted into the cost reduction benefit of data assets, which is an important quantitative basis for the valuation of data assets.

6. The compressor data asset valuation method based on optimization algorithm according to claim 1, characterized in that: S4 When analyzing the dynamic effects of surge margin on output and load, surge margin is defined. ,in The surge critical pressure, To determine the actual operating pressure, a dynamic correlation equation is constructed: ； ； in, Let be the output at time t. For the rated output, Let be the surge margin at time t. This is the output impact coefficient. The attenuation coefficient is... The angular frequency of the wave. This is the initial phase. For integration variables; Let t be the load at time t. For rated load, This is the load influence factor. The rate of change of surge margin is used; the parameters of the dynamic equation are calibrated by least squares support vector machine, and the prediction error is less than 2% after calibration; the dynamic correlation equation can calculate the output increase brought about by "optimizing surge margin", which is transformed into the efficiency benefit of data assets, and together with the cost reduction benefit, constitutes the comprehensive value of data assets.

7. The compressor data asset valuation method based on optimization algorithm according to claim 1, characterized in that: When calculating energy consumption differences using the improved particle swarm optimization-gradient descent hybrid optimization algorithm in S5, the efficiency output by the S3 correlation model is used first. Energy consumption of the output of the S4 coupled equation For the target variable, construct the optimization objective function: ； in, Due to differences in energy consumption, Energy consumption before and after data application, For compressor efficiency after data application, For optimal efficiency. For the output after data application, To achieve optimal output, These are weighting coefficients used to balance the importance of different objectives; particle positions are initialized during the particle swarm optimization phase. With speed The updated formula is: ； ； in, The velocities of the i-th particle in the j-th dimension at time t+1 and t are respectively. For inertial weights, As a learning factor, A random number in the range [0,1]. Let be the optimal position of the i-th particle in the j-th dimension. Let be the globally optimal position in the j-th dimension. Let be the positions of the i-th particle in the j-th dimension at time t+1 and t, respectively; when the number of particle swarm optimization iterations reaches 60% of the total number of iterations, switch to the gradient descent algorithm, and the gradient update formula is: ； in, For learning rate, For the objective function in The gradient at point is iterated until the objective function value is less than 1. Stop at the appropriate time and output the optimal value. .

8. The compressor data asset valuation method based on optimization algorithm according to claim 1, characterized in that: S2 Outlier removal employs a method based on local outlier factors, calculating the local outlier factor for each data point. The formula is: ； in, For the number of nearest neighbors, For the data points to be evaluated, for Nearest neighbor data points, for The set of k nearest neighbors, for arrive The reachable distance, for arrive The reachable distance, for and Euclidean distance, for The distance to its k-th nearest neighbor; when Values ​​identified as outliers are removed; standardization preprocessing uses Z-score standardization, with the formula: ； in, This represents the value of the j-th parameter at the i-th sampling time after standardization. This represents the original value of the j-th parameter at the i-th sampling time before standardization. Let j be the mean of the j-th parameter. Let be the standard deviation of the j-th parameter; the data preprocessing process can remove invalid interference data and unify the parameter dimensions, ensuring the data quality of subsequent S3 modeling and S4 analysis, and avoiding deviations in value assessment results caused by abnormal data.

9. The compressor data asset valuation method based on optimization algorithm according to claim 7, characterized in that: S5 Medium compressor energy efficiency attenuation coefficient Calculate using the following formula: ； in, For the evaluation period, For the efficiency at time t, For initial efficiency, The integral value of efficiency over the evaluation period; capacity fluctuation coefficient. The calculation is as follows: ； in, For the rated output, Let be the output at time t. To assess average yield over the evaluation period; loss quantification data Including heat loss Mechanical loss Leakage loss The calculation formula is: ； ； ； in, For heat dissipation coefficient, For heat dissipation area, The average temperature. The ambient temperature; For mechanical loss coefficient, To average operating pressure, Average flow rate; Leakage coefficient, The leakage area; Total loss; By calculating the energy efficiency decay coefficient, the capacity fluctuation coefficient, and loss quantification data, the calculation results of energy consumption differences can be corrected, and the interference of non-data factors on value assessment can be eliminated.

10. The compressor data asset valuation method based on optimization algorithm according to claim 7, characterized in that: When determining the value of data assets in S5, energy consumption differences are used as a basis. The unit price of energy type Annual operating time of equipment Equipment depreciation factor Construct a value calculation formula: in, For the value of data assets, Annual cost of the data acquisition system The annual maintenance cost of the algorithm model; when the energy type is steam, It needs to be converted into equivalent electrical energy. The conversion formula is: in, To account for the difference in equivalent electrical energy consumption, Due to differences in steam consumption, This refers to the enthalpy of vapor. For vapor density, For boiler efficiency, For turbine efficiency; the value calculation result is rounded to two decimal places; The value calculation formula transforms "energy consumption difference" into "annual net income" and deducts data-related costs to obtain the actual value of data assets.

Images