Search

CN-122022150-A - Carbon market policy impact causal deduction platform based on double difference method

CN122022150ACN 122022150 ACN122022150 ACN 122022150ACN-122022150-A

Abstract

The invention belongs to the field of data processing, and discloses a carbon market policy impact causal deduction platform based on a double difference method; the method comprises the steps of collecting multidimensional characteristic data of a carbon market test point area and a non-test point area, carrying out standardized processing on the data, designing a multiple interpolation algorithm based on machine learning, combining an area economic level and an auxiliary variable of an energy structure, filling a missing value, and marking low-reliability data and reducing the weight of the low-reliability data in subsequent analysis if the deviation between the filled value and an actual observed value exceeds 10%. And generating a sample weight matrix according to the characteristic difference of the test point region and the non-test point region, and enabling the non-test point region to be more matched with the test point region in the characteristic space through weight distribution. And (3) adopting a sliding window mechanism, automatically adjusting weight distribution and outputting a parallel trend test report if the trend difference exceeds a preset threshold value, so as to realize the carbon market policy impact causal deduction based on a double difference method.

Inventors

  • LV LIJUAN

Assignees

  • 山东青年政治学院

Dates

Publication Date
20260512
Application Date
20260127

Claims (8)

  1. 1. A carbon market policy impact causal deduction platform based on a double difference method, comprising: The data preprocessing and feature extraction sub-module is used for collecting multidimensional feature data of the carbon market test point area and the non-test point area, carrying out standardized processing on the data, extracting main feature vectors through principal component analysis and reducing data dimension; the non-random data missing filling sub-module is used for designing a multiple interpolation algorithm based on machine learning, combining with auxiliary variables of regional economic level and energy structure, filling the missing value, carrying out consistency test on the filled data, marking the filled data as low-reliability data and reducing the weight of the filled data in subsequent analysis if the deviation between the filled value and an actual observed value exceeds 10 percent; The self-adaptive weight distribution sub-module is used for constructing a weight generation algorithm based on a gradient lifting tree, and generating a sample weight matrix according to the characteristic difference of the test point region and the non-test point region; the dynamic parallel trend test sub-module is used for calculating the carbon emission intensity trend difference of the test point area and the weighted non-test point area by adopting a sliding window mechanism in a time window before policy implementation, automatically adjusting weight distribution until the trend difference converges if the trend difference exceeds a preset threshold value, and outputting a parallel trend test report comprising a trend difference curve, a confidence interval and a significance test result; And the policy effect estimation sub-module is used for constructing a double differential model based on the adjusted weight matrix and estimating the net effect of policy implementation on the carbon emission intensity of the test point region.
  2. 2. The carbon market policy impact causal deduction platform of claim 1, wherein the double differencing method comprises: constructing a dynamic space weight matrix based on three dimensions of inter-provincial technology flow, energy source transmission topology and geographic proximity; the spatial Dubin double-difference joint dynamic spatial weight matrix is embedded into the DID model, and the direct effect of the policy on the test point area and the indirect effect on the surrounding area are estimated; and decomposing the total emission reduction effect into a direct effect and an indirect effect, and outputting an effect decomposition report.
  3. 3. The carbon market policy impact causal deduction platform based on the double differencing method of claim 2, wherein the spatial weight matrix comprises: Constructing a long and short-time memory neural network, predicting a future carbon price sequence, and embedding the future carbon price sequence as a time-varying covariant into a DID model; and expanding the DID model, adding interactive items of carbon price fluctuation, and outputting a dynamic risk report comprising the change trend and the confidence interval of the policy effect under different carbon price scenes.
  4. 4. The carbon market policy impact causal deduction platform based on the double differencing method according to claim 3, wherein the DID model comprises: the method comprises the steps of accessing four types of data sources, including macroscopic data, microscopic data, real-time data and space data, designing a block chain anchoring check channel, triggering data consistency verification through an intelligent contract, and automatically starting manual auditing if the deviation between provincial data and satellite inversion values exceeds 15%; and (3) adopting a multi-mode data alignment algorithm to normalize the data with different time-space resolutions into a unified format, and storing the data in a distributed data lake.
  5. 5. The dual differencing based carbon market policy impact causal deduction platform of claim 4, wherein the distributed data lake comprises: the policy context library is internally provided with 12 policy combinations including carbon tax, quota auction, green certificate transaction and the like; based on the DID model, simulating a carbon emission path when the policy is not implemented, and calculating a policy net effect and a confidence interval; Three-dimensional evaluation reports are generated, including policy net effect, enterprise emission reduction cost and benefit simulation and regional industry structure transformation paths.
  6. 6. The carbon market policy impact causal deduction platform based on the double differencing method according to claim 5, wherein the policy context library comprises: Constructing multi-subject models of power enterprises, government regulatory authorities, carbon traders and the like, and training a Q-learning strategy based on historical data; Taking the emission reduction effect output by the DID model as a reward function, inputting the reward function into the MAS model, and optimizing a main decision; and automatically generating policy optimization suggestions according to the simulation result.
  7. 7. The carbon market policy impact causal deduction platform based on the double differencing method of claim 6, wherein the multi-principal model comprises: The method comprises the steps of using a hierarchical parallel architecture, adopting Spark clusters to calculate K_O DID (direct digital) coefficient in parallel by a first layer, using a GPU to accelerate matrix operation by a second layer, and realizing real-time visualization of a differential result by a third layer through a D3.js engine; and supporting dynamic resource allocation, and automatically expanding the computing nodes when the data size exceeds a preset threshold.
  8. 8. The carbon market policy impact causal deduction platform based on the double differencing method of claim 7, wherein the hierarchical parallel architecture comprises: the thermodynamic diagram layer displays the change rate of the intensity of each province carbon in a color gradient; The network layer represents a direct effect value by the size of a node, the edge width represents the overflow effect intensity, and the time-space evolution of the retrospective 2011-2025 year is supported; the user views the detailed effect analysis report at a specific time point or region by dragging the time axis or clicking on the node.

Description

Carbon market policy impact causal deduction platform based on double difference method Technical Field The invention belongs to the field of data processing, and particularly relates to a carbon market policy impact causal deduction platform based on a double difference method. Background The traditional double difference method (DID) relies on the parallel trend assumption, but there is a significant difference between the pilot and non-pilot regions in the actual carbon market before policy implementation. For example, studies have shown that carbon emission reduction effects (11.5%) in the eastern economically developed region far exceed those in the western region (4.2%), and that conventional DID models have difficulty capturing this regional heterogeneity. Meanwhile, the problem of non-random missing of carbon emission data exists, and especially the missing rate of partial year data in the western region is up to 30%, so that estimation deviation is obvious. Carbon markets are affected by multidimensional factors including energy price fluctuations (e.g., 2024 increases 2.5% of global natural gas consumption), climate anomalies (e.g., extremely high temperature frequent) and technical innovations (e.g., novel hydrogen production technologies), but existing models lack the ability to quantitatively analyze policy-market-technology dynamic coupling mechanisms. The radiation effect (such as joint emission reduction caused by technical diffusion) of the carbon transaction test point area on the peripheral area needs to depend on a space metering model, but the prior art does not realize deep coupling of DID and SDM, and cannot simultaneously identify the policy net effect and the space overflow effect. Aiming at the problems, the patent provides a carbon market policy deduction platform integrating a multi-mode causal deduction engine. Disclosure of Invention In order to overcome the defects in the prior art and achieve the purposes, the invention provides a carbon market policy impact causal deduction platform based on a double difference method, which comprises a data preprocessing and feature extraction sub-module, a data processing and feature extraction sub-module and a data processing and feature extraction sub-module, wherein the data preprocessing and feature extraction sub-module is used for collecting multidimensional feature data of a carbon market test point area and a non-test point area, carrying out standardized processing on the data, extracting main feature vectors through principal component analysis, and reducing data dimension. The non-random data missing filling sub-module is used for designing a multiple interpolation algorithm based on machine learning, combining with auxiliary variables of regional economy level and energy structure, filling missing values, carrying out consistency test on filled data, and marking the filled data as low-reliability data and reducing the weight of the filled data in subsequent analysis if the deviation between the filled value and an actual observed value exceeds 10%. The self-adaptive weight distribution sub-module is used for constructing a weight generation algorithm based on a gradient lifting tree, generating a sample weight matrix according to the characteristic difference of the test point region and the non-test point region, and enabling the non-test point region to be more matched with the test point region in the characteristic space through weight distribution. The dynamic parallel trend test sub-module is used for calculating the carbon emission intensity trend difference of the test point area and the weighted non-test point area by adopting a sliding window mechanism in a time window before policy implementation, automatically adjusting weight distribution until the trend difference is converged if the trend difference exceeds a preset threshold value, and outputting a parallel trend test report comprising a trend difference curve, a confidence interval and a significance test result. And the policy effect estimation sub-module is used for constructing a double differential model based on the adjusted weight matrix and estimating the net effect of policy implementation on the carbon emission intensity of the test point region. Preferably, the double differentiation method comprises: constructing a dynamic space weight matrix based on three dimensions of inter-provincial technology flow, energy source transmission topology and geographic proximity; the spatial Dubin double-difference joint dynamic spatial weight matrix is embedded into the DID model, and the direct effect of the policy on the test point area and the indirect effect on the surrounding area are estimated; and decomposing the total emission reduction effect into a direct effect and an indirect effect, and outputting an effect decomposition report. Preferably, the spatial weight matrix includes: Constructing a long and short-time memory neural network, predicting a futur