CN-121980782-A - Compatibility constraint-oriented automatic generation method for simultaneous modeling of standing tree total, overground and underground biomass

CN121980782ACN 121980782 ACN121980782 ACN 121980782ACN-121980782-A

Abstract

The invention provides a compatibility constraint-oriented automatic generation method for simultaneous modeling of standing tree total, overground and underground biomass, and relates to the technical field of forestry metrology. The method comprises the steps of constructing an observation mask and a sample weight descriptor, and explicitly modeling the addition relation between the total biomass and the above-ground and underground biomass into a compatibility constraint calculation diagram, and automatically generating a candidate simultaneous model set under the unified topology constraint. Further, the candidate simultaneous models are quantitatively evaluated and optimized by constructing a simultaneous objective function constrained by compatibility and executing uniform parameter estimation, and finally a deployable model product comprising a model structure, a parameter file, an input field specification and a prediction interface definition is generated. The method can ensure the consistency of the prediction result in the numerical value and physical significance, improve the stability, the automation degree and the engineering applicability of the standing tree biomass modeling, and is suitable for application scenes such as forest resource monitoring, carbon reserve evaluation and the like.

Inventors

MA FENGFENG
PENG PAI
PI BING
XIAO YAQIN
ZHANG XIE
LIU ZHENHUA
SONG QINGAN
WU XIAOLI

Assignees

湖南省林业科学院
湖南省森林草原防火监测调度评估中心

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (10)

1. The automatic generation method of the combined modeling of the total, overground and underground biomass of the standing tree facing the compatibility constraint is characterized by comprising the following steps: constructing a modeling sample set based on a standing tree sample data set, performing unit and dimension unification on the modeling sample set to obtain the modeling sample set after the unification, and extracting a missing measurement state of a dependent variable observation field set in the modeling sample set, wherein the modeling sample set comprises an independent variable field set and a dependent variable observation field set; generating an observation mask according to the missing state, and determining a sample weight descriptor according to a preset weight generation rule; constructing a compatibility constraint calculation diagram according to the added compatibility constraint, and setting total output nodes, overground output nodes and underground output nodes according to the compatibility constraint calculation diagram; Under the topological constraint of the compatibility constraint calculation graph, generating a candidate simultaneous model set based on a preset function family template library, and establishing an overground sub-model structure and an underground sub-model structure for each candidate simultaneous model in the candidate simultaneous model set, and generating a total sub-model structure according to the compatibility constraint calculation graph, wherein the overground sub-model structure and the output end of the underground sub-model structure adopt a monotonic positive mapping function as output mapping; Constructing a simultaneous objective function for each candidate simultaneous model in the candidate simultaneous model set based on the observation mask, the sample weight descriptor and the compatibility constraint calculation map, and performing parameter solution on the simultaneous objective function by adopting a simultaneous parameter estimation algorithm to obtain a candidate parameter set corresponding to each candidate simultaneous model one by one; Calculating model evaluation index values for candidate parameter sets of the candidate simultaneous models according to a preset evaluation index system, determining an optimal simultaneous model according to a preset decision rule, and generating a deployable model product, wherein the deployable model product comprises model structure description, parameter files, input field specifications and prediction interface definition of the optimal simultaneous model.
2. The method for automatically generating a compatibility constraint-oriented simultaneous modeling of aggregate, overground and underground biomass of standing tree of claim 1, wherein the set of dependent variable observation fields comprises: total biomass, aboveground biomass, and underground biomass.
3. The method for automatically generating a compatibility constraint-oriented simultaneous modeling of aggregate, overground and underground biomass in accordance with claim 1, wherein said set of independent variable fields comprises: chest diameter and tree height.
4. The method for automatically generating the simultaneous modeling of the total, overground and underground biomass for the compatibility constraint according to claim 1, wherein the steps of constructing a modeling sample set based on a standing tree sample data set, performing unit and dimension unification processing on the modeling sample set to obtain the modeling sample set after the unification processing, extracting the missing measurement state of a dependent variable observation field set in the modeling sample set include: Obtaining sample records in the standing tree sample data set, and analyzing each sample record to obtain an independent variable field set and a dependent variable observation field set; Constructing the modeling sample set based on the set of independent variable fields and the set of dependent variable observation fields; Performing validity screening treatment on the modeling sample set to obtain a screened modeling sample set; obtaining a preset unit mapping rule and a preset dimension mapping rule, uniformly converting units of each field in the screened modeling sample set based on the preset unit mapping rule, and uniformly converting dimension representations of each field in the screened modeling sample set based on the preset dimension mapping rule to obtain a uniform modeling sample set; According to the modeling sample set after the unification processing, executing observation state judgment processing on the dependent variable observation field set of each sample record, wherein the observation state judgment processing comprises the steps of respectively judging whether a valid observation value exists for a total biomass field, an overground biomass field and an underground biomass field; based on the result of the observation state judgment processing, generating a dependent variable missing state identifier for each sample record in the modeling sample set after the unification processing to obtain missing states of a dependent variable observation field set in the modeling sample set, wherein the dependent variable missing state identifier is used for representing observed states and missing states of the total biomass field, the overground biomass field and the underground biomass field.
5. The method for automatically generating a compatibility constraint-oriented simultaneous modeling of total, overground, and underground biomass according to claim 1, wherein the generating an observation mask according to the absence of measurement state and determining a sample weight descriptor according to a preset weight generation rule comprises: Acquiring a modeling sample set after the consistency processing and a missing measurement state of a dependent variable observation field set in the modeling sample set; Generating an observation mask for the modeling sample set based on the missing state, wherein the observation mask is aligned to a sample dimension of the modeling sample set and a field dimension of the set of dependent variable observation fields, and generating an observation mark on each of the dependent variable observation fields for each sample record, the observation mark being used for indicating that a corresponding dependent variable observation field is observed or missing; Determining a sample weight coefficient for each sample record in the modeling sample set according to the preset weight generation rule and in combination with the observation mask; And organizing the sample weight coefficients into sample weight descriptors in a manner consistent with the sample order of the modeling sample set.
6. The automatic generation method of the standing tree total, overground and underground biomass simultaneous modeling for compatibility constraint according to claim 1, wherein the steps of constructing a compatibility constraint calculation graph according to the added compatibility constraint, and setting total output nodes, overground output nodes and underground output nodes according to the compatibility constraint calculation graph include: obtaining a summation compatibility constraint, wherein the summation compatibility constraint defines a predicted total biomass from a sum of a predicted above-ground biomass and a predicted below-ground biomass; obtaining a graph structure description of a compatibility constraint calculation graph based on the addition of the compatibility constraint, wherein the graph structure description comprises a node set and a connection relation set; Setting total output nodes, overground output nodes and underground output nodes in the node set, and respectively configuring field identifications for the total output nodes, overground output nodes and underground output nodes so as to correspond to a total biomass field, an overground biomass field and an underground biomass field; setting a summation operation node in the node set, and establishing a connection relation from the overground output node to the underground output node and a connection relation from the summation operation node to the total output node in the connection relation set so as to obtain the compatibility constraint calculation graph; the compatibility constraint calculation map has the expression: ; Wherein, the To predict total biomass; to predict aboveground biomass; To predict subsurface biomass.
7. The method for automatically generating the simultaneous modeling of the aggregate, the overground and the underground biomass for the compatibility constraint according to claim 1, wherein the generating the set of candidate simultaneous models based on a preset function family template library under the topology constraint of the compatibility constraint calculation graph, and establishing an overground sub-model structure and an underground sub-model structure for each candidate simultaneous model in the set of candidate simultaneous models, generating a total sub-model structure according to the compatibility constraint calculation graph comprises: Obtaining a compatibility constraint calculation diagram and a preset function family template library; Obtaining a candidate simultaneous model set according to the function family template library and the topological constraint of the compatibility constraint calculation graph, wherein each candidate simultaneous model in the candidate simultaneous model set comprises an overground sub-model structure corresponding to overground output nodes, an underground sub-model structure corresponding to underground output nodes and a total quantity sub-model structure corresponding to total quantity output nodes; based on each candidate simultaneous model in the candidate simultaneous model set, building an above-ground sub-model structure according to the function family template library and connecting an output end of the above-ground sub-model structure to the above-ground output node; Establishing an underground submodel structure according to the function family template library and connecting an output end of the underground submodel structure to the underground output node; Setting a summation operation structure between the output end of the above-ground sub-model structure and the output end of the underground sub-model structure according to the compatibility constraint calculation diagram, and connecting the output end of the summation operation structure to the total output node to generate the total sub-model structure; The total submodel structure expression is: ; Wherein, the To predict aboveground biomass; To predict subsurface biomass; To predict total biomass; is chest diameter; Is tree-height; is a monotonic positive mapping function; The first model parameter, the second model parameter and the third model parameter are respectively; The first subsurface sub-model parameter, the second subsurface sub-model parameter, and the third subsurface sub-model parameter, respectively.
8. The method of claim 1, wherein the constructing a simultaneous objective function for each candidate simultaneous model in the set of candidate simultaneous models based on the observation mask, the sample weight descriptor, and the compatibility constraint computation graph, and performing a simultaneous parameter estimation algorithm on the simultaneous objective function to obtain a set of candidate parameters that are in one-to-one correspondence with each candidate simultaneous model, comprises: Based on each candidate simultaneous model in the candidate simultaneous model set, obtaining prediction output corresponding to a total output node, an overground output node and an underground output node respectively according to the compatibility constraint calculation diagram; determining a residual accounting set of the predicted output according to the observation mask; Constructing a simultaneous objective function according to the residual calculation set and the sample weight descriptor, wherein the simultaneous objective function comprises weighted residual items respectively corresponding to the total output node, the overground output node and the underground output node, and the weighted residual items only indicate the residual calculation set to generate and calculate a sample record and a dependent variable observation field which participate in residual calculation; Performing parameter solving on the simultaneous objective function by adopting a simultaneous parameter estimation algorithm to obtain candidate parameters corresponding to the candidate simultaneous models; Collecting the candidate parameters corresponding to each candidate simultaneous model to obtain a candidate parameter set corresponding to each candidate simultaneous model one by one; the expression of the simultaneous objective function is: ; Wherein, the Is a simultaneous objective function; is a simultaneous model parameter set; Is the number of samples; Is the first Sample weight coefficients for the bar samples; Is an observation mask; Is the first Strip sample in dependent variable field An observation value on the table; Is the first Strip sample in dependent variable field Predicted values of the above; Respectively corresponding to total biomass, overground biomass and underground biomass.
9. The method for automatically generating the compatibility constraint-oriented aggregate, overground and underground biomass simultaneous modeling according to claim 1, wherein the calculating model evaluation index values for candidate parameter sets of the candidate simultaneous models according to a preset evaluation index system and determining an optimal simultaneous model according to a preset decision rule to generate a deployable model product comprises: acquiring a candidate simultaneous model set and candidate parameter sets corresponding to the candidate simultaneous models one by one, and acquiring a modeling sample set and an observation mask after the unification treatment; According to each candidate simultaneous model in the candidate simultaneous model set, configuring corresponding candidate parameters into the candidate simultaneous model, and generating prediction output of the current candidate simultaneous model based on the modeling sample set after the unification treatment, wherein the prediction output comprises prediction values corresponding to total output nodes, overground output nodes and underground output nodes respectively; Determining a model evaluation index value from a dependent variable observation field in the modeling sample set after prediction output and unification processing according to a preset evaluation index system and the observation mask, wherein the observation mask is used for limiting a sample record and a dependent variable observation field which participate in model evaluation index value calculation; comparing the model evaluation index values of the candidate simultaneous models according to a preset decision rule, determining an optimal simultaneous model, and determining an optimal parameter set corresponding to the optimal simultaneous model; Generating a deployable model product based on the optimal simultaneous model and the optimal parameter set, wherein the deployable model product comprises a model structure description, a parameter file, an input field specification and a prediction interface definition of the optimal simultaneous model.
10. The compatibility constraint-oriented method of simultaneous modeling of aggregate, overground, and subsurface biomass of claim 1, wherein the deployable model product is configured to output predicted values of aggregate, overground, and subsurface biomass, respectively, via the aggregate submodel structure, the overground submodel structure, and the subsurface submodel structure upon receipt of the set of independent variable fields.

Description

Compatibility constraint-oriented automatic generation method for simultaneous modeling of standing tree total, overground and underground biomass Technical Field The invention relates to the technical field of forestry metrology, in particular to a compatibility constraint-oriented automatic generation method for simultaneous modeling of standing tree total, overground and underground biomass. Background In the biomass measurement and monitoring work of standing trees, a method for establishing an empirical or semi-empirical regression model based on tree measurement factors such as breast diameter, tree height, crown width and the like of the sample wood is generally adopted to realize the estimation of biomass. The existing method comprises the steps of independently fitting the above-ground biomass and the underground biomass respectively, introducing a total biomass model, and checking consistency of the results of the above-ground biomass model, the underground biomass model and the total biomass model after modeling so as to meet the requirements of production application and resource monitoring. Along with the continuous investigation of forest resources and the continuous expansion of the scale of sample plot investigation data, the model updating frequency is obviously improved, and the modeling work presents standardized and engineering development trend. The industry gradually forms a unified candidate model set, an evaluation index system and screening rules, and hopes to shift the model production process from relying on personal experience to a reproducible, traceable and batch-deployable automatic process so as to support rapid iterative application of cross-tree species, cross-region and multi-period data. While the prior art emphasizes the summation consistency between the total biomass and the above-ground biomass and the underground biomass, the actual modeling is still highly dependent on manual model structure selection, initial value setting, outlier processing and weight adjustment, multi-model comparison and achievement solidification, which results in low efficiency and poor repeatability. Especially when the sample has missing measurement or large noise difference, the common practice is easy to have the problems that the compatibility is difficult to stably maintain in the prediction stage, the predicted value is unreasonable and nonnegative, and a large amount of reworking is required for cross-data set migration, so that the application requirements of large-scale continuous updating and quick deployment are difficult to meet. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide the automatic generation method of the simultaneous modeling of the total, overground and underground biomass of the standing tree facing the compatibility constraint, and solves the problems that the automation degree of the simultaneous modeling is low and the compatibility constraint is difficult to stably maintain in the prior art. In order to achieve the above object, the present invention provides the following solutions: an automatic generation method of a standing tree total, overground and underground biomass simultaneous modeling facing compatibility constraint comprises the following steps: constructing a modeling sample set based on a standing tree sample data set, performing unit and dimension unification on the modeling sample set to obtain the modeling sample set after the unification, and extracting a missing measurement state of a dependent variable observation field set in the modeling sample set, wherein the modeling sample set comprises an independent variable field set and a dependent variable observation field set; generating an observation mask according to the missing state, and determining a sample weight descriptor according to a preset weight generation rule; constructing a compatibility constraint calculation diagram according to the added compatibility constraint, and setting total output nodes, overground output nodes and underground output nodes according to the compatibility constraint calculation diagram; Under the topological constraint of the compatibility constraint calculation graph, generating a candidate simultaneous model set based on a preset function family template library, and establishing an overground sub-model structure and an underground sub-model structure for each candidate simultaneous model in the candidate simultaneous model set, and generating a total sub-model structure according to the compatibility constraint calculation graph, wherein the overground sub-model structure and the output end of the underground sub-model structure adopt a monotonic positive mapping function as output mapping; Constructing a simultaneous objective function for each candidate simultaneous model in the candidate simultaneous model set based on the observation mask, the sample weight descriptor and the compatibility cons