Parallel computing and policy evolution simulation method and platform for massive macro data
By matching the current macroeconomic state with historical states and performing continuation calculations and corrections, the problems of unified state expression and inheritance-based continuation calculations in existing technologies are solved, achieving efficient and reliable policy evolution simulation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI NORMAL UNIVERSITY
- Filing Date
- 2026-04-07
- Publication Date
- 2026-06-09
Smart Images
Figure CN122175473A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer simulation technology, and in particular relates to a parallel computing and policy evolution simulation method and platform for massive macro data. Background Technology
[0002] Macroeconomic analysis and policy effectiveness evaluation have long relied on statistical calculations, econometric models, scenario simulation programs, and data platforms oriented towards business analysis. Existing technologies typically use aggregated macroeconomic indicators as input to predict, compare, or calculate the impact of established policy tools. However, when simulating the continuous evolution of massive policy scenarios, several tightly coupled and long-standing problems remain unresolved. First, existing calculation systems often treat the current macroeconomic state as a one-time input, lacking a processing mechanism to unify the current state with historical evolution results into a single expression space. This makes direct comparison between the current state and historical trajectories difficult, and even if historical results are preserved, it is hard to accurately determine which segments can be inherited. Second, while existing parallel computing platforms can distribute and accelerate multiple tasks, they typically treat different policy scenarios as independent tasks, lacking a successor-based calculation mechanism for policy evolution scenarios. Therefore, when faced with a large number of similar policy scenarios, it is still necessary to repeatedly calculate a large number of identical or similar evolutionary segments. Third, even if existing reuse technologies can utilize historical intermediate results, they often remain at the level of simple hits and direct splicing, lacking a dedicated correction process to address the coordination requirements between aggregate, structural, and temporal relationships in the macro-state. This leads to situations where the continuation results are locally continuous but overall unbalanced, thus affecting the credibility of policy projections. Precisely because there is a lack of a dedicated technical link for macro-policy simulation between current state representation, inheritance segment identification, continuation execution, and result correction, existing technologies struggle to simultaneously balance the efficiency of large-scale parallel solutions with the reliability of policy evolution results. Summary of the Invention
[0003] This invention discloses a parallel computing and policy evolution simulation method and platform for massive macroeconomic data, in order to solve the problems mentioned in the background art.
[0004] To achieve the above objectives, the first aspect of the present invention provides a parallel computing and policy evolution simulation method for massive macroeconomic data, the method comprising:
[0005] S100: Receive basic macroscopic state data and convert the basic macroscopic state data into an initial macroscopic state, wherein the initial macroscopic state includes a positionalized state component and its corresponding weight.
[0006] S200: Based on the initial macro state, identify inheritable evolutionary segments in the historical evolution result data that match the current state in terms of structural position and policy action mode, and generate a continuation calculation description result, which includes the historical trajectory number, the inheritance end time step, and the continuation calculation start point.
[0007] S300: Based on the continued calculation description results, extract the state sequence of the inheritance segment, and combine the differences between the current policy parameters and the historical policy parameters, perform the evolution result inheritance calculation at the continued calculation starting point to generate candidate evolution trajectories;
[0008] S400: Perform consistency correction on the candidate evolutionary trajectories based on macroscopic structural relationships and temporal continuity, and output the policy evolution simulation results.
[0009] Furthermore, the process of converting the basic macroscopic state data into an initial macroscopic state includes:
[0010] Extract the index value at the current moment to form the original state vector;
[0011] Based on the central level and fluctuation level of similar index sequences in historical evolution data, the position of the current value relative to the historical distribution is calculated, and the positional state component is generated.
[0012] Based on the range of change of the indicator within the historical window, the components with smaller ranges of change are marked as steady-state components and assigned higher weights, while the components with larger ranges of change are marked as dynamic components and assigned lower weights, thus generating the corresponding weights.
[0013] Furthermore, the identification of inheritable evolutionary segments includes:
[0014] Calculate the matching score between the current initial macro state and the starting state of each historical trajectory. The matching score includes a state difference item and a policy difference item. Select the historical trajectory with the lowest matching score as the candidate trajectory.
[0015] The candidate trajectory is scanned along the time dimension, and the segment score at each time step is calculated. The segment score includes a time adjustment term that increases with the time step.
[0016] When the score of a segment is consistently below the preset threshold, the corresponding time step is included in the inherited segment; when the score of a segment exceeds the preset threshold for the first time, the previous time step is recorded as the end position of the inherited segment, and the current time step is used as the starting point for the continuation calculation.
[0017] Furthermore, the time adjustment term introduced when calculating the segment score at each time step makes the inherited segments more concentrated in the early part of the trajectory, giving priority to the use of segments with strong commonalities in the early stage of policy evolution.
[0018] Furthermore, the inheritance and continuation of the execution evolution result includes:
[0019] At the starting point of the continuation calculation, the historical trend increment is calculated using the state difference between two adjacent time steps at the end of the inherited segment.
[0020] The incremental policy difference is calculated using the difference between current policy parameters and historical policy parameters;
[0021] Calculate the boundary traction increment using the deviation between the current state and the inherited boundary state;
[0022] Based on the historical trend increment, policy difference increment, and boundary traction increment, a new state of the continued calculation segment is generated recursively until the target simulation length is reached.
[0023] Furthermore, the injection coefficient of the policy difference increment increases over time, while the traction coefficient of the boundary traction increment decreases over time, ensuring the continuity of the trajectory at the boundary in the early stage of the continuation calculation, and making the influence of the current policy dominant in the middle and later stages.
[0024] Furthermore, the generation of candidate evolutionary trajectories includes:
[0025] The state sequence of the identified inheritable evolutionary segment is spliced with the new state sequence generated by the continued calculation segment in chronological order to form a candidate evolutionary trajectory.
[0026] Furthermore, the performance of consistency correction includes:
[0027] Read the constraint matrix and reference vector containing the macroscopic structural coordination relationships;
[0028] Calculate the constraint residuals between the state vector and the constraint matrix of the candidate evolution trajectory at each time step;
[0029] Introduce a continuous time term to calculate the difference between the current state and the corrected state of the previous time step;
[0030] Based on the constraint residuals and time continuity terms, the state vectors of the candidate evolution trajectories are corrected to generate the final evolution trajectory.
[0031] Furthermore, the correction coefficient of the time continuity term decays exponentially with time step, which suppresses sudden jumps near the inheritance and continuation boundary during the initial correction of the trajectory, and makes the structural constraint term dominant during the mid-to-late correction of the trajectory.
[0032] A second aspect of the invention provides a parallel computing and policy evolution simulation platform for massive macroeconomic data, the platform comprising:
[0033] The state generation module is used to receive basic macroscopic state data and convert it into an initial macroscopic state, wherein the initial macroscopic state includes positionalized state components and their corresponding weights.
[0034] The segment identification module is used to identify inheritable evolutionary segments in the historical evolution result data that match the current state in terms of structural position and policy action, based on the initial macro state, and generate a continuation description result.
[0035] The continuation calculation execution module is used to extract the state sequence of the inheritance segment based on the continuation calculation description result, and combine the differences between the current policy parameters and historical policy parameters to perform evolution result inheritance calculation at the continuation calculation starting point to generate candidate evolution trajectories;
[0036] The result correction module is used to perform consistency correction on the candidate evolution trajectory based on macroscopic structural relationships and temporal continuity relationships, and output the policy evolution simulation results.
[0037] The beneficial technical effects of the present invention are at least as follows:
[0038] To address the aforementioned issues, this invention provides a parallel computing and policy evolution simulation method and platform for massive macroeconomic data. First, the current macroeconomic state is transformed into a unified state expression that can be directly compared with historical evolution results, establishing a determinable and inheritable connection between the current task's macroeconomic starting point and historical trajectories. Second, based on this unified state expression, inheritable evolutionary segments are identified by combining historical trajectory states and corresponding policy conditions. The identification results are directly converted into continuation calculation description information, enabling the system to clearly know which historical trajectory to inherit, to which time point, and from where to continue calculation. Then, in the continuation calculation phase, the local trends at the end of the inherited segment serve as the basis for continuation, gradually injecting the differences between the current policy and historical policies into the subsequent state update process. This ensures that the newly generated trajectory retains the continuity of historical evolution while reflecting the actual action path of the current policy. Finally, candidate evolutionary trajectories undergo consistency correction based on macroeconomic structural relationships and temporal continuity, yielding formal simulation results that can be directly used for policy comparison, policy sandbox simulation, and platform demonstration. Through the above overall design, this invention transforms the problem of repetitive solution under massive policy scenarios into the problem of inheritable segment identification and differential segment continuation calculation, and further solves the structural imbalance problem that is prone to occur after reuse and continuation calculation through consistency correction, thereby achieving the unity of parallel efficiency and simulation reliability within the same scheme. Attached Figure Description
[0039] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.
[0040] Figure 1 This is a flowchart of the parallel computing and policy evolution simulation method for massive macro data according to the present invention.
[0041] Figure 2 This is a framework diagram of the parallel computing and policy evolution simulation platform for massive macro data of this invention. Detailed Implementation
[0042] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0043] In one or more embodiments, such as Figure 1 As shown, a parallel computing and policy evolution simulation method for massive macroeconomic data is disclosed, the method comprising the following:
[0044] S100: Receive basic macroscopic state data and convert the basic macroscopic state data into an initial macroscopic state, wherein the initial macroscopic state includes positionalized state components and their corresponding weights.
[0045] Specifically, step one involves receiving basic macroscopic state data. This is then transformed into the initial macroeconomic state that can be directly used in subsequent policy evolution simulations. . The aggregated macroeconomic indicator data source is typically pushed to the simulation platform from statistical databases, thematic indicator libraries, or business data platforms at fixed intervals during actual deployment. The platform reads data snapshots at the current point in time through periodic extraction tasks and then assembles them into current state records according to a pre-agreed field order. The key here is not to perform general cleaning, but to organize indicators from different sources and with different fluctuation characteristics into the same set of horizontally comparable state components, so that subsequent steps can determine which historical evolution segments are suitable for inheritance based on the current economic starting point. In specific implementation, the system first starts from... Extract the index value at the current moment to form the original state vector. For example, in a scenario simulating regional policy implementation, the system reads aggregate indicators, price indicators, and structural indicators from the database at a certain point in time and writes them sequentially into the same vector; if the ratio values of the three indicators read at the moment are respectively , and ,but It can be represented as These values all come from fields that have already been aggregated in the business system, so there is no need to split them here. The system only needs to keep the field order stable to facilitate subsequent one-to-one correspondence with the same positions in the historical trajectory.
[0046] get Then, the system utilizes historical evolution data. The sequence of similar indicators in the dataset is used to perform positional transformation on each positional component, thereby generating the initial macroscopic state. In the project implementation, the platform will first select from the indicator numbers... Extract data records from the corresponding historical window, such as reading the same indicator sequence from several past periods, then calculate the central level and volatility level of the sequence, and finally convert the current value into its position relative to the historical distribution. The calculation method is as follows:
[0047] ;
[0048] in, It is the original state vector The Middle The current value of each indicator is directly derived from the input data. The current time point record is obtained; It is data resulting from historical evolution. The Middle The average level of an indicator within the selected historical window can be obtained by the system reading the historical records corresponding to that indicator, summing them one by one, and then dividing by the number of records. It represents the fluctuation level of the indicator within the same historical window, which the system can obtain by statistically analyzing the deviation between historical records and the average level. This is a fixed small value in the platform configuration file, written into memory during system initialization to ensure the stability of the calculation process. Taking the above formula as an example, if the current value of a certain indicator... Corresponding to the historical average level Historical fluctuation level ,Pick Then we can get This indicates that the indicator is currently positioned slightly above its historical average. Processing all indicators in the same way yields the complete result. At this point, all components are in a unified and comparable space, so when judging the degree of matching between the historical trajectory and the current state, it will not be dominated by a single high-value indicator.
[0049] In formation Afterwards, the system continued to rely on The magnitude of change of each indicator within the historical window, for The internal dimensions are structured and labeled, and weights for subsequent matching are generated for each component according to a preset mapping relationship. In practice, the platform reads the historical values of each indicator from several past periods, calculates the difference between the maximum and minimum values of the indicator within a window, and compares this range of change with a preset threshold. Components with smaller ranges of change are marked as steady-state components and given higher weights, while components with larger ranges of change are marked as dynamic components and given lower weights. For example, a certain structural ratio indicator may consistently remain within a certain range over multiple periods. arrive Between these points, the system will label them as steady-state components and map them to higher values. If a price indicator shows more significant fluctuations between adjacent periods, it will be marked as a dynamic component and mapped to a lower value. At the same time, the platform will obtain the results from step one. , and Solidify it into the state transition configuration, and for The subsequent historical states involved in matching and calculation are positionally processed using the same expression, so that in step two... , and in step three , All with They reside in the same state representation space. The labeling results, weight mapping results, and state transition configurations are consistent with... Binding storage forms the final output. The final result is... It is a state vector that has been positionalized and carries structural labels. It retains the original correspondence of the input indicators and transforms the current macroscopic state into a unified expression suitable for matching with historical trajectories. Subsequent steps can directly use this to perform evolutionary result inheritance and calculation.
[0050] S200: Based on the initial macro state, identify inheritable evolutionary segments in the historical evolution result data that match the current state in terms of structural position and policy action mode, and generate a continuation calculation description result, which includes the historical trajectory number, the inheritance end time step, and the continuation calculation start point.
[0051] Specifically, the initial macroscopic state has already been obtained in step one. This step then revolves around a core operation: analyzing historical evolutionary data. Find the current state Historical trajectory segments that match both "structural location" and "policy action mode" are then converted into subsequent recalculation description results that can be directly accessed. Step one has already transformed the raw macroscopic data into a state representation with a unified scale and completed the comparability processing between different indicators within the same vector. Therefore, step two can directly perform matching and filtering within this unified representation space without involving the original data processing. In a practical system, The data is stored in time series format. Each historical trajectory consists of a state vector of several consecutive time steps, and is associated with a corresponding historical policy parameter record through a trajectory number. These policy parameters are linked to the current policy scenario data to be evaluated. Maintaining consistency in field structure allows for dimension-by-dimensional comparison.
[0052] The system first targets each historical trajectory The initial state is used for matching calculation. This calculation is an extension of the classic vector distance model, and its basic source can be traced back to the dimension-by-dimensional difference idea of Euclidean distance or Manhattan distance. This step introduces two types of improvements on this basis: one is to utilize the structural weights already formed in step one. The approach involves adjusting the importance of different indicators and introducing the differences between policy parameters as an additional term, ensuring that the matching results simultaneously reflect both state consistency and policy consistency. The specific calculation is expressed as follows:
[0053] ;
[0054] in, For the first The matching score of each historical trajectory; the smaller the value, the closer it is to the current simulation conditions. The dimension of the state vector is given by the output of step one. The length is determined directly; The first in the current initial macroscopic state The values of each component; For historical evolution results data The Middle The trajectory at the starting time step corresponds to the first... Each component, the value of which is obtained by reading the trajectory status table according to the trajectory number; For the first The weights of each component are generated from the division of steady-state and dynamic components in step one. In implementation, this can be accomplished by storing the marker bits in the database and mapping them to the weight values. This is an adjustment coefficient used to balance the impact of state difference items and policy difference items; this coefficient is set in the system configuration. As a policy parameter dimension, it is derived from policy scenario data. The number of fields is determined; For the current policy scenario to be evaluated in the first The values at each parameter position come directly from the input data. ; For the first The historical policy parameters corresponding to each historical trajectory are obtained by linking the trajectory number to the policy record table.
[0055] In the specific calculation process, this can be achieved as follows: The system reads the starting states and corresponding policy parameters of several historical trajectories from the database, loads them into memory to form two matrices, one for states and one for policies; then, it performs dimension-wise interpolation and weighted summation operations on each row. For example, if the current... The initial state of a certain historical trajectory is Weight Then the state item is Divide by the dimension to get approximately If the current policy parameters Historical policy parameters The policy difference item is: ,Pick The policy item is approximately ,final The system calculates all trajectories. Then, sort by numerical value and select several trajectories corresponding to the minimum value as a candidate set.
[0056] After obtaining candidate trajectories, the system further identifies inheritable evolutionary segments along the time dimension. This process unfolds step-by-step based on time series analysis, introducing a time position adjustment term on top of the matching scores. This approach stems from the principle of prioritizing the stability of the early stages of a time series in time series analysis. This step adds an adjustment factor that increases with each time step, concentrating the inheritance segments more in the early stages of the trajectory, thus better reflecting the commonalities prevalent in the early stages of policy evolution. The specific calculation is expressed as follows:
[0057] ;
[0058] in, Indicates the first Candidate trajectories at time step The score for the section; For this trajectory at time step The Each state component is obtained by sequentially reading the corresponding record in the trajectory state table; The current scanned time step number is given by the program loop variable; The total time length of the trajectory is obtained from the trajectory index table; This is the time adjustment coefficient, set in the system configuration. The second term in this formula... The increase in time makes it more difficult for later segments to meet the inheritance conditions, thus enabling priority utilization of earlier stable intervals at the algorithm level.
[0059] During implementation, the system performs a sequential scan of each candidate trajectory: from time step... Start calculating step by step And compare it with a preset threshold. When When the value remains below the threshold, the system will include the corresponding time step in the inheritance segment; when it first appears... When the threshold is exceeded, the system records... This is the end position of the inherited section, and... As the starting point for subsequent calculations. For example, in a certain calculation, the state differences of a trajectory in the first four time steps are as follows: ,Pick , The corresponding time adjustment items are as follows: The obtained segment score is If the threshold is set to If the first three time steps are included in the inheritance segment, the fourth time step will serve as the starting point for the continuation calculation. After completing the segment identification of all candidate trajectories, the system will organize the results into a continuation calculation description result. In implementation, The data is stored as a set of records, each containing a current policy scenario index, trajectory number, inheritance end time step, and continuation start position. This data comes directly from the aforementioned calculation process and is written to the task scheduling table. The current policy scenario index is used to directly locate the corresponding policy parameter record in step three, the trajectory number is used to read the inherited historical trajectory status, and the inheritance end time step and continuation start position are used to determine the inheritance segment boundary and the continuation start position. Through this processing, This is mapped to a specific set of execution instructions: "policy scenario index + trajectory number + time location," enabling subsequent steps to be directly based on... Initiate parallel computation to transform the unified state representation constructed in step one into an executable evolution starting point and inheritance segment.
[0060] S300: Based on the continued calculation description results, extract the state sequence of the inheritance segment, and combine the differences between the current policy parameters and historical policy parameters, perform the evolution result inheritance calculation at the continued calculation starting point to generate candidate evolution trajectories.
[0061] Specifically, step three directly uses the continuation description result generated in step two. As the entry point for execution. The system has already provided the historical trajectory number, inheritance end time step, and continuation start point for each policy scenario to be evaluated. Therefore, the system will not perform matching judgments at this stage, but will instead... This is interpreted as a set of executable continuation tasks. In practice, the scheduling module first reads each task one by one. The system retrieves the task records, then uses the trajectory numbers to extract the state sequences corresponding to the inherited segments from the historical evolution results data table, and writes these sequences into the trajectory cache of the current task in chronological order. Next, it uses the policy scenario index in the same record to retrieve the current policy parameters from the policy data table, and then uses the trajectory number to retrieve the policy parameters corresponding to the historical trajectory from the historical policy record table. In this way, when the system enters the continuation calculation, it simultaneously possesses three sets of data directly used for calculation: one set is the state at the end of the inherited segment and its state at the previous moment, which gives the historical evolution direction near the boundary; one set is the current policy parameters, which give the actual policy effects to be applied in this simulation; and the third set is the historical policy parameters, which give the policy background when the inherited trajectory was formed. Step two outputs... This serves to bind together the three things: "which historical trajectory, where to start the continuation calculation, and which set of policy parameters to use during the continuation calculation." Therefore, the continuation calculation process obtained in step three naturally inherits the judgment result of the previous step, rather than reconstructing a new trajectory that is detached from the inheritance segment.
[0062] The basic idea of continuation calculation comes from the discrete-time differential update formula in mathematics, which states that "the state at the next time step equals the current state plus several increment terms." In the scenario of macroeconomic policy evolution, the new state after the boundary must not only continue the local trend formed at the end of the inherited segment, but also gradually reflect the difference between the current policy and historical policies, while ensuring a smooth transition between the newly generated first few steps and the inherited segment at the boundary. Based on this idea, the system decomposes the increment in the discrete update formula into three parts: the first part is the historical trend increment, which consists of the state difference between two adjacent time steps at the end of the inherited segment; the second part is the policy difference increment, which consists of the difference between the current policy parameters and the historical policy parameters; and the third part is the boundary traction increment, which is used to control the smooth development of the initial state around the inherited boundary. After combining these three parts, the candidate evolution trajectory at each time step... The Each state component is generated by the following formula:
[0063] ;
[0064] in, It is the first time step generated by the current task in the next time step. Each state component is generated and immediately written into the candidate trajectory cache; It is the first step of the current time. Each state component is taken from the state record corresponding to the end time step of the inheritance at the start of the continuation calculation, and in subsequent steps it is taken from the state just generated in the previous round. It was the first step in the previous period of time. Each state component is taken from the historical state record before the end of the inheritance time step in the first step of the continuation calculation, and then updated as the recursion process proceeds. It is the trend continuation coefficient, which is given by the task configuration table and is used to control the degree to which historical local trends are retained in the continuation calculation; Is with the first The policy difference corresponding to each state component is actually calculated by subtracting the historical policy parameter from the current policy parameter. The field correspondence comes from the policy parameter mapping table. It is the first The policy injection coefficient for each time step of the continuation calculation is pre-generated according to the task configuration, which is a coefficient sequence that increases over time, so that the policy difference enters smoothly in the early stage of the continuation calculation and is gradually released in subsequent time steps; It is the first The boundary traction coefficient of each continuation time step is pre-generated according to the task configuration as a coefficient sequence that decreases over time, so that the state of the few steps near the boundary remains continuous with the inherited section, and then is gradually released. It is the first step at the end of the inheritance time. Each state component comes from the state records of the historical trajectory at the boundary positions. The derivation of this update formula can be understood as follows: first, from the classic discrete update framework... Let's go, and then... It is broken down into three parts: "trend continuation term + policy injection term - boundary traction term". The trend continuation term uses the difference between adjacent time steps to represent the historical change direction already formed near the boundary. The policy injection term uses the difference between the current policy and historical policies to represent the new effect of the current task relative to the inherited trajectory. The boundary traction term uses the deviation between the current state and the boundary state to suppress abrupt jumps in the first few steps of the continuation calculation. Because... , and All of these originate from the unified state representation processed in step one. It is also converted to the same computational scale through a mapping table. , , All of these are proportionality coefficients, so all terms in the above formula can be directly added together, and the calculation process remains consistent.
[0065] At the system implementation level, and This isn't a temporary estimate; rather, before entering the time loop, two coefficient sequences of the same length as the number of continuation steps are generated from the task configuration table and written into memory. The generation method uses an exponential timetable, derived from the classic laws of exponential growth and decay: policy effects gradually enter after the boundary, while boundary pull gradually weakens over time. Before each task begins, the platform generates an increasing timetable based on the continuation start point and the target simulation length. Sequence and a decreasing one The sequence is then processed, and values are directly retrieved by time step index within the loop. The advantage of this design is that in the early stages of computation, the system prioritizes ensuring the continuity of the trajectory at the boundaries, while in the later stages, the influence of the current policy becomes dominant, aligning with the actual pattern that the effects of macroeconomic policies typically require several time steps to gradually propagate. Taking a certain component as an example, the state inherited at the end of the time step is... The previous moment was The difference between current and historical policies in this component for If at a certain step , , Then the trend continuation term is Policy injection items are The boundary traction term is in the first step of the continuation calculation because And for The new state is now If the system reads new data in the next step... , By recursively applying the results from the previous step, a subsequent state that is continuous near the boundary and gradually reflects policy differences can be obtained. This calculation can be performed directly through a loop: first, read the boundary state, then substitute the configured coefficients and the policy difference of the current task, calculate a new state, and immediately write it to the trajectory cache, then proceed to the next time step, until the number of termination steps specified in the task table is reached.
[0066] After all time steps have been calculated, the system concatenates the state sequence of the inherited segment with the newly generated state sequence of the continued segment in chronological order to form a candidate evolution trajectory. This information is then written to a candidate trajectory table or memory buffer for further processing. It has a very clear structure: the first part comes from the inheritable evolutionary segment already confirmed in step two, and the middle and latter parts come from the recursive calculation results under the current policy conditions. The two are naturally connected through boundary states. Here, step three takes the descriptive results output from step two. It is truly transformed into a candidate evolutionary trajectory that can be directly observed and further corrected. Therefore, the inheritance mechanism, parallel task scheduling, and subsequent consistency processing can all be carried out around the same trajectory object.
[0067] S400: Perform consistency correction on the candidate evolutionary trajectories based on macroscopic structural relationships and temporal continuity, and output the policy evolution simulation results.
[0068] Specifically, step four directly uses the candidate evolutionary trajectories generated in step three. To process the object, each state vector within it is modified along the time sequence to a final state that satisfies macroscopic constraints, thus forming the final policy evolution simulation result. The processing here is based on a combination of two classic methods: one is the mathematical idea of projection correction, which involves pulling a vector back along the constraint direction when it deviates from the constraint set, bringing it back to the acceptable region; the other is the idea of discrete-time series smoothing, which involves controlling the jumps in state changes between adjacent time steps to keep the trajectory continuous in time. Step three yields... The concatenation results of the inherited and continued calculation segments have already been demonstrated. Therefore, step four no longer changes the trajectory length and time step position. Instead, it makes small, directional adjustments to the state components at each time step, ensuring that the trajectory retains the evolutionary trend formed by policy effects while also meeting the coordination requirements of the macro-structure at each time step. During system implementation, the constraint matrix is first read from the rule configuration table, and then from the historical evolution result data... Read the corresponding reference vector from the stable section, and then... The correction is performed step by step at each time step, and the corrected state is written to the result cache, ultimately forming the trajectory. .
[0069] The constraint matrix used here is derived from the structured expression of business rules. In specific implementation, the platform first selects a set of indicators that are stable over the long term and should maintain coordination across different policy scenarios. Examples include the linear consistency between total and structural components, the combined balance of several core components, and the continuity relationship where abnormal oscillations should not occur between adjacent time steps. Each linear relationship can be written as "several state components added together by fixed coefficients to approximate a certain benchmark value." The system organizes these coefficients into a matrix. And organize the corresponding baseline values into a vector. .matrix Each row in the matrix represents a constraint, and the matrix elements are given by the rule configuration table; vector Then based on historical evolution data The baseline value for a constraint is obtained by statistically analyzing stable segments of trajectories of the same category as the current task. For example, by reading the state records of the latter part of several historical trajectories and averaging the corresponding linear combinations. Since the state representation in steps one through three is always at a uniform scale, The components and matrices in The target and reference vector Since they all reside in the same representation space, subsequent matrix operations can be performed directly. The system manages time steps... Candidate state vector When performing correction, first calculate its constraint residuals. Then introduce the continuous time term ,in This represents the state after correction at the previous time step. The constraint residual reflects the degree of deviation of the current state from the macroscopic structural relationship, while the time continuity term reflects the degree of deviation between the current state and the previous time step. Combining the two allows us to simultaneously address the two requirements that must be met in the macroscopic trajectory: "structural consistency" and "time smoothness."
[0070] Based on this, the system uses the following correction formula to calculate the final state step by step:
[0071] ;
[0072] in, Indicates time step The corrected state vector will be written into the final trajectory. ; Representing candidate evolutionary trajectories At time step The original state vector is given directly from step three; It is a constraint matrix, derived from predefined linear coordination relationships in the rule configuration table; It is a matrix The transpose of is used to reproject the constraint residuals back into the state space; It is a baseline vector, derived from historical evolution data. The statistics for the stable section were obtained, and the state transition configuration was changed to match the one established in step one. The same state representation space; It is the structural correction coefficient, used to control the magnitude of constraint correction; It is a time step The corresponding time smoothing coefficient is used to control the magnitude of the continuity correction between adjacent time steps; It is the state vector that has been corrected in the previous time step, and can be taken from the first correction step of the time series. The previous state or the end state of the inherited segment. The derivation of this formula comes from the superposition of the projection correction idea and the discrete-time smoothing idea: First, the residual Describe the deviation of the current state from the constraint set; then, multiply the residual on the left. This maps the deviation direction in the constraint space back to the state space, thus obtaining a set of correction values that can be directly applied to the state vector; then multiply by the scaling factor. This yields the structural correction term. Meanwhile, the state vector should remain continuous in time; therefore, a smoothing correction term is introduced. Its direction of action always points to the correction result of the previous time step. Ultimately, the above formula can be understood as "original candidate state minus structural deviation, then minus temporal mutation". All terms in the formula belong to vector representations in the same state space, so they can be subtracted component by component.
[0073] To ensure the correction intensity converges naturally with each time step, the system further employs a time-varying smoothing coefficient to control the correction rhythm in the initial and subsequent stages. This design is derived from the classic exponential decay law: in the early stages of the trajectory, the inherited and continued calculation segments have just been spliced together, and temporal continuity is more important; as time progresses, the trajectory gradually moves away from the splicing boundary, and structural constraints become more suitable as the dominant factor. Therefore, the system expresses the time smoothing coefficient in exponential decay form and updates it at each time step:
[0074] ;
[0075] in, Indicates time step The corresponding smoothing coefficient will be substituted into the previous formula. use; It is the initial smoothing coefficient, given by the task configuration table; It is the decay rate, determined by the parameter tuning record table or offline simulation results; It is the current time step number. The logical relationship between this formula and the previous formula is: first, based on the time step... The smoothing coefficient required for this step Then Substituting these terms into the state correction formula yields the final state at the current time step. Thus, the smoothing term has a stronger effect during the initial correction of the trajectory, effectively suppressing sudden jumps near the boundaries of inheritance and continuation calculations; during the later correction of the trajectory, the smoothing term gradually weakens, allowing the structural constraint term to gradually become dominant, thereby more accurately reflecting the impact of the current policy scenario on the mid-to-late stage evolution.
[0076] In the specific implementation process, the following specific numerical calculation example is given. Assume the candidate state vector at a certain time step is... The state vector after correction at the previous time step is The constraint matrix is taken ; Reference vector is taken Structural correction coefficient is taken as The initial smoothing coefficient is taken as The decay rate is taken Take the current time step First, calculate using the second formula. Next, calculate the constraint residuals: ,therefore Multiply by the left again get Therefore, the structural correction term is In the continuous time term, the difference between the current state and the previous corrected state is... multiplied by have to Subtracting the two corrections from the original state yields... The system then writes this correction state into the final trajectory and uses it as the basis for calculations in the next time step. Process them in the same order. After completing all time steps, the complete final evolutionary trajectory can be obtained. .
[0077] After the system completes the correction for all time steps, the final trajectory will be... The results are written into the results table along with the corresponding policy scenario identifier, task number, and simulation time range. After reading these results, the presentation layer can directly generate multi-policy scenario comparison curves, time-series change graphs of key state components, and scenario difference analysis views. Step four is responsible for "upgrading candidate trajectories into formal simulation results that can be directly used for policy comparison and result presentation": Step two provides the starting point for the continuation calculation, Step three generates candidate trajectories, and Step four converges the trajectories to the final output that satisfies the macro-coordination relationship through structural correction and time smoothing. Therefore, the entire scheme forms a closed loop here.
[0078] In one or more embodiments, such as Figure 2 As shown, a parallel computing and policy evolution simulation platform for massive macroeconomic data is disclosed, the platform comprising:
[0079] The state generation module is used to receive basic macroscopic state data and convert it into an initial macroscopic state, wherein the initial macroscopic state includes positionalized state components and their corresponding weights.
[0080] The segment identification module is used to identify inheritable evolutionary segments in the historical evolution result data that match the current state in terms of structural position and policy action, based on the initial macro state, and generate a continuation description result.
[0081] The continuation calculation execution module is used to extract the state sequence of the inheritance segment based on the continuation calculation description result, and combine the differences between the current policy parameters and historical policy parameters to perform evolution result inheritance calculation at the continuation calculation starting point to generate candidate evolution trajectories;
[0082] The result correction module is used to perform consistency correction on the candidate evolution trajectory based on macroscopic structural relationships and temporal continuity relationships, and output the policy evolution simulation results.
[0083] It is worth noting that the specific workflow of the parallel computing and policy evolution simulation platform for massive macro data provided in this embodiment of the invention is the same as that of the parallel computing and policy evolution simulation method for massive macro data described in the above embodiments, and will not be repeated here.
[0084] This invention also provides a parallel computing and policy evolution simulation device for massive macroeconomic data, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps described in the above embodiments of the parallel computing and policy evolution simulation method for massive macroeconomic data, for example... Figure 1 The steps S1 to S4 described above; or, when the processor executes the computer program, it implements the functions of each module in the above platform embodiments.
[0085] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the parallel computing and policy evolution simulation device for massive macroscopic data.
[0086] The parallel computing and policy evolution simulation device for massive macroeconomic data can be a desktop computer, laptop, handheld computer, or cloud server, etc. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the device may also include input / output devices, network access devices, buses, etc.
[0087] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the parallel computing and policy evolution simulation device for massive macroscopic data, connecting all parts of the device via various interfaces and lines.
[0088] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the parallel computing and policy evolution simulation device for massive macro data. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating platform, at least one application program required for a function, etc.; the data storage area may store data created according to the operation of the controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0089] The module integrated with the parallel computing and policy evolution simulation equipment for massive macroeconomic data, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.
[0090] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0091] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A parallel computing and policy evolution simulation method for massive macroeconomic data, characterized in that, The method includes: S100: Receive basic macroscopic state data and convert the basic macroscopic state data into an initial macroscopic state, wherein the initial macroscopic state includes a positionalized state component and its corresponding weight. S200: Based on the initial macro state, identify inheritable evolutionary segments in the historical evolution result data that match the current state in terms of structural position and policy action mode, and generate a continuation calculation description result, which includes the historical trajectory number, the inheritance end time step, and the continuation calculation start point. S300: Based on the continued calculation description results, extract the state sequence of the inheritance segment, and combine the differences between the current policy parameters and the historical policy parameters, perform the evolution result inheritance calculation at the continued calculation starting point to generate candidate evolution trajectories; S400: Perform consistency correction on the candidate evolutionary trajectories based on macroscopic structural relationships and temporal continuity, and output the policy evolution simulation results.
2. The parallel computing and policy evolution simulation method for massive macroeconomic data as described in claim 1, characterized in that, The process of converting basic macroscopic state data into initial macroscopic state includes: Extract the index value at the current moment to form the original state vector; Based on the central level and fluctuation level of similar index sequences in historical evolution data, the position of the current value relative to the historical distribution is calculated, and the positional state component is generated. Based on the range of change of the indicator within the historical window, the components with smaller ranges of change are marked as steady-state components and assigned higher weights, while the components with larger ranges of change are marked as dynamic components and assigned lower weights, thus generating the corresponding weights.
3. The parallel computing and policy evolution simulation method for massive macroeconomic data as described in claim 1, characterized in that, The identification of inheritable evolutionary segments includes: Calculate the matching score between the current initial macro state and the starting state of each historical trajectory. The matching score includes a state difference item and a policy difference item. Select the historical trajectory with the lowest matching score as the candidate trajectory. The candidate trajectory is scanned along the time dimension, and the segment score at each time step is calculated. The segment score includes a time adjustment term that increases with the time step. When the score of a segment is consistently below the preset threshold, the corresponding time step is included in the inherited segment; when the score of a segment exceeds the preset threshold for the first time, the previous time step is recorded as the end position of the inherited segment, and the current time step is used as the starting point for the continuation calculation.
4. The parallel computing and policy evolution simulation method for massive macroeconomic data as described in claim 3, characterized in that, The time adjustment term introduced when calculating the segment score at each time step makes the inherited segments more concentrated in the early part of the trajectory, giving priority to the use of segments with strong commonalities in the early stage of policy evolution.
5. The parallel computing and policy evolution simulation method for massive macroeconomic data as described in claim 1, characterized in that, The inheritance and continuation of the execution evolution results include: At the starting point of the continuation calculation, the historical trend increment is calculated using the state difference between two adjacent time steps at the end of the inherited segment. The incremental policy difference is calculated using the difference between current policy parameters and historical policy parameters; Calculate the boundary traction increment using the deviation between the current state and the inherited boundary state; Based on the historical trend increment, policy difference increment, and boundary traction increment, a new state of the continued calculation segment is generated recursively until the target simulation length is reached.
6. The parallel computing and policy evolution simulation method for massive macroeconomic data according to claim 5, characterized in that, The injection coefficient of the policy difference increment increases over time, while the traction coefficient of the boundary traction increment decreases over time. This ensures that the continuity of the trajectory at the boundary is prioritized in the initial stage of the continuation calculation, while the influence of the current policy becomes dominant in the middle and later stages.
7. The parallel computing and policy evolution simulation method for massive macroeconomic data according to claim 1, characterized in that, The generated candidate evolutionary trajectories include: The state sequence of the identified inheritable evolutionary segment is spliced with the new state sequence generated by the continued calculation segment in chronological order to form a candidate evolutionary trajectory.
8. The parallel computing and policy evolution simulation method for massive macroeconomic data according to claim 1, characterized in that, The consistency correction includes: Read the constraint matrix and reference vector containing the macroscopic structural coordination relationships; Calculate the constraint residuals between the state vector and the constraint matrix of the candidate evolution trajectory at each time step; Introduce a continuous time term to calculate the difference between the current state and the corrected state of the previous time step; Based on the constraint residuals and time continuity terms, the state vectors of the candidate evolution trajectories are corrected to generate the final evolution trajectory.
9. The parallel computing and policy evolution simulation method for massive macroeconomic data as described in claim 8, characterized in that, The correction coefficient of the time continuity term decays exponentially with time step, which suppresses sudden jumps near the inheritance and continuation boundary during the early stage of trajectory correction, and makes the structural constraint term dominant during the middle and later stages of trajectory correction.
10. A parallel computing and policy evolution simulation platform for massive macroeconomic data, characterized in that: The platform includes: The state generation module is used to receive basic macroscopic state data and convert it into an initial macroscopic state, wherein the initial macroscopic state includes positionalized state components and their corresponding weights. The segment identification module is used to identify inheritable evolutionary segments in the historical evolution result data that match the current state in terms of structural position and policy action, based on the initial macro state, and generate a continuation description result. The continuation calculation execution module is used to extract the state sequence of the inheritance segment based on the continuation calculation description result, and combine the differences between the current policy parameters and historical policy parameters to perform evolution result inheritance calculation at the continuation calculation starting point to generate candidate evolution trajectories; The result correction module is used to perform consistency correction on the candidate evolution trajectory based on macroscopic structural relationships and temporal continuity relationships, and output the policy evolution simulation results.