Farmland total primary productivity optimization simulation method based on particle filter

Abstract

The invention relates to a farmland total primary productivity optimization simulation method based on particle filtering. The method comprises the following steps: constructing a biosphere-atmosphere exchange process simulator model and acquiring a random parameter set based on preset empirical distribution; driving the biosphere-atmosphere exchange process simulator model based on a random parameter group in the random parameter set and obtaining a corresponding simulation variable; obtaining an observation variable corresponding to the analog variable, and carrying out iterative operation on the model parameters based on the observation variable; and repeating the iterative operation until the assimilation of the observation data is completed. The method does not involve complex matrix transposition and inversion operation, is simpler and more efficient, and is more suitable for optimization of a complex ecological system process model; the condition limitation that a Kalman filtering algorithm requires that a processing system must be a linear system is overcome; the method breaks through the condition requirement that a variational method requires a system state variable to be based on Gaussian distribution, and the nonlinear and non-Gaussian random mode feature enables the method to have a wide application prospect.

Description

Technical Field

[0001] The present application relates to the technical field of particle filtering, and in particular to a method for optimizing and simulating the total primary productivity of farmland based on particle filtering. Background Art

[0002] Gross Primary Productivity (GPP) is the total amount of carbon fixed by green vegetation per unit area and per unit time through photosynthesis, and is an important indicator of crop carbon sink capacity and yield. Accurately estimating the GPP of farmland ecosystems is not only conducive to understanding the changing characteristics of terrestrial ecosystems under global climate change and human activities, but also helps to avoid possible global food risks in the future. Terrestrial ecosystem models are important tools for studying the GPP of terrestrial ecosystems, which can simulate and predict the dynamic changes of GPP at global and regional scales.

[0003] In the related art, for example, Chinese patent CN116011635A provides a method and system for estimating regional crop yields by particle filter data assimilation, the method comprising: running each particle at a fixed time interval during the critical growth period of the crop until each particle runs to the crop maturity period, and obtaining the first LAI output by each particle; updating the initial weight of each particle according to the posterior probability density function, each updated first LAI and the likelihood function, and obtaining the updated weight of each particle; when the particle divergence is less than a preset value, resampling each particle in the initial particle set and updating the initial particle set; when each particle runs to the crop maturity period, obtaining the estimated value of the crop yield according to the crop yield and final weight of each particle. The present invention can improve the efficiency of sequential sampling of particles, reduce the number of resampling times, maintain the diversity of particle input parameters, and at the same time improve the speed and accuracy of crop yield prediction.

[0004] However, in the related art, the production of crops is directly simulated, optimized and updated through the particle filtering algorithm. In the actual simulation process of the particle filtering algorithm, there is a situation where the observed value is inconsistent with the predicted value, resulting in the subsequent particle filtering optimization algorithm being inconsistent with the actual observation situation, making the simulation results mismatch with the actual observation results, resulting in low accuracy of the simulation results. Summary of the invention

[0005] 1. Problem to be solved Based on this, it is necessary to provide a particle filtering-based optimization simulation method for total primary productivity of farmland, which considers the relationship between observed values ​​and simulated values ​​and improves the accuracy of simulation results, in order to address the above technical problems.

[0006] 2. Technical solution In the first aspect, the present application provides a method for optimizing and simulating the total primary productivity of farmland based on particle filtering. The method comprises: Constructing a biosphere-atmosphere exchange process simulator model and obtaining a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each random parameter group includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; Based on the parameter groups composed of random values ​​of sensitive parameters, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding simulation variables are obtained; Obtaining observed variables corresponding to the simulation variables and performing an iterative operation on the model parameters based on the observed variables, wherein the iterative operation includes: Based on the observed variables and the simulated variables corresponding to the model parameters, calculating the likelihood probability of the observed variables occurring when the simulated variables corresponding to the model parameter group occur; Adjust the weights of the sensitive parameter group based on the likelihood probability, and set the adjusted weights as update weights; Constructing an update parameter set based on the update weights and the corresponding sensitive parameter groups; Based on the updated parameter set, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding updated simulation variables are obtained; The iterative operation is repeated until the observed variables are assimilated.

[0007] In one embodiment, the likelihood probability of the observed variable occurring when the simulated variable occurs is calculated based on the observed variable and the simulated variable: ; in, is the likelihood probability of the observed variable occurring when the simulated variable occurs, represents pi; Represents the natural base; represents the standard deviation of the observations, express Moment parameter groups, express The corresponding simulation variable is express The observed variable corresponding to the moment; Based on the likelihood probability of the observed variable occurring, updating the weight of the parameter group; The updated weight of the parameter group The likelihood probability is calculated to ensure that the sum of the weights of all parameter groups at each moment is 1, and the formula is as follows: ; in, is the likelihood probability of the observed variable occurring when the simulated variable occurs, express Moment The updated weights of the parameter groups are express Moment The updated weights of the parameter groups.

[0008] In one embodiment, the estimated value of the number of valid parameter groups is calculated as follows: ; in, is an estimate of the number of valid parameter sets, For the Parameter groups in Update weight at each moment; comparing the estimated number of valid parameter groups with a preset critical threshold; If the estimated value of the number of valid parameter groups is less than a preset critical threshold, it is determined to perform parameter set resampling and obtain a resampled parameter set.

[0009] In one embodiment, determining to resample the parameter set includes: The cumulative number of replications of the statistical update parameter set is calculated using the following formula: ; in, For Before The cumulative number of replications for the update parameter group, is the rounding function, is the number of parameter groups, For the Update parameter groups in Update weight at each moment; Based on the cumulative number of replications, the number of replications for each parameter group is calculated using the following formula: = - ; in, For the Update parameter groups in The number of times a moment is replicated, For Before The cumulative number of replications for the update parameter group, For Before The cumulative number of replications for each update parameter group; Whether to resample the parameter group is determined based on the number of copies of the update parameter group.

[0010] In one embodiment, determining whether to resample the parameter group based on the number of replications of the update parameter group includes: When the estimated value of the number of valid parameter groups is less than a preset critical threshold, the update parameter group whose number of replications is not zero is resampled based on the Holden sequence and the exponential function to obtain a resampled parameter group, and the formula is as follows: ; in, is the rounding function, express Moment parameter groups, For the Update parameter groups in The number of times a moment is replicated, To update the offset of the parameter group, the formula is as follows: ; in, is obtained by resampling based on the updated parameter group Moment The offset corresponding to each parameter group, To update the number of parameter groups, Used to determine the dispersion of the sampling parameter groups.

[0011] In the second aspect, the present application also provides a device for optimizing and simulating the total primary productivity of farmland based on particle filtering. The device comprises: A random parameter construction module is used to construct a biosphere-atmosphere exchange process simulator model and obtain a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each of which includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; A simulation variable calculation module is used to drive the biosphere-atmosphere exchange process simulator model based on a random parameter set and obtain the corresponding simulation variables; A model parameter iteration module is used to obtain the observed variables corresponding to the simulation variables and perform iterative operations on the model parameters based on the observed variables, wherein the iterative operations include: calculating the likelihood probability of the observed variables occurring when the simulation variables corresponding to the model parameter group occur based on the observed variables and the simulation variables corresponding to the model parameters; adjusting the weights of the sensitive parameter group based on the likelihood probability, and setting the adjusted weights as the updated weights; constructing an updated parameter set based on the updated weights and the corresponding sensitive parameter group; and driving the biosphere-atmosphere exchange process simulator models based on the updated parameter set and obtaining the corresponding updated simulation variables; The iterative assimilation output module is used to repeat the iterative operation until the assimilation of the observed variables is completed.

[0012] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program: Constructing a biosphere-atmosphere exchange process simulator model and obtaining a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each random parameter group includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; Based on the parameter groups composed of random values ​​of sensitive parameters, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding simulation variables are obtained; Obtaining observed variables corresponding to the simulation variables and performing an iterative operation on the model parameters based on the observed variables, wherein the iterative operation includes: Based on the observed variables and the simulated variables corresponding to the model parameters, calculating the likelihood probability of the observed variables occurring when the simulated variables corresponding to the model parameter group occur; Adjust the weights of the sensitive parameter group based on the likelihood probability, and set the adjusted weights as update weights; Constructing an update parameter set based on the update weights and the corresponding sensitive parameter groups; Based on the updated parameter set, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding updated simulation variables are obtained; The iterative operation is repeated until the observed variables are assimilated.

[0013] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented: Constructing a biosphere-atmosphere exchange process simulator model and obtaining a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each random parameter group includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; Based on the parameter groups composed of random values ​​of sensitive parameters, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding simulation variables are obtained; Obtaining observed variables corresponding to the simulation variables and performing an iterative operation on the model parameters based on the observed variables, wherein the iterative operation includes: Based on the observed variables and the simulated variables corresponding to the model parameters, calculating the likelihood probability of the observed variables occurring when the simulated variables corresponding to the model parameter group occur; Adjust the weights of the sensitive parameter group based on the likelihood probability, and set the adjusted weights as update weights; Constructing an update parameter set based on the update weights and the corresponding sensitive parameter groups; Based on the updated parameter set, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding updated simulation variables are obtained; The iterative operation is repeated until the observed variables are assimilated.

[0014] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented: Constructing a biosphere-atmosphere exchange process simulator model and obtaining a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each random parameter group includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; Based on the parameter groups composed of random values ​​of sensitive parameters, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding simulation variables are obtained; Obtaining observed variables corresponding to the simulation variables and performing an iterative operation on the model parameters based on the observed variables, wherein the iterative operation includes: Based on the observed variables and the simulated variables corresponding to the model parameters, calculating the likelihood probability of the observed variables occurring when the simulated variables corresponding to the model parameter group occur; Adjust the weights of the sensitive parameter group based on the likelihood probability, and set the adjusted weights as update weights; Constructing an update parameter set based on the update weights and the corresponding sensitive parameter groups; Based on the updated parameter set, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding updated simulation variables are obtained; The iterative operation is repeated until the observed variables are assimilated.

[0015] 3. Beneficial effects Compared with the prior art, the present invention has the following beneficial effects: (1). This method uses a particle filter algorithm to obtain the simulated values ​​and observed values ​​of the BEPS model, and adjusts the weights in the model parameters through the observed values ​​to iterate the new simulated values ​​until the simulated values ​​of the BEPS model are iterated to a result that matches the observed values. It does not involve complex matrix transposition and inversion operations, nor does it require professionals to perform time-consuming model accompanying code writing. Compared with other common data assimilation algorithms, it is simpler and more efficient, and more suitable for the optimization of complex ecosystem process models. On the one hand, this method overcomes the conditional restriction that the Kalman filter algorithm requires the processing system to be a linear system; on the other hand, it also breaks through the conditional requirement of the variational method that the system state variables are based on Gaussian distribution. This nonlinear non-Gaussian random pattern feature makes it have a wide range of application prospects.

[0016] (2) This method is based on the process-based ecosystem model BEPS. The BEPS model contains multiple mechanism processes such as plant photosynthesis and respiration, and takes into account the close coupling of water, heat and carbon related processes. Therefore, when optimizing the total primary productivity of crops, this method does not directly optimize the variables, but indirectly optimizes the variables by adjusting the model control parameters, which enables this method to analyze the optimization results from a deeper and more mechanistic perspective. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a method for optimizing and simulating total primary productivity of farmland based on particle filtering in one embodiment; Figure 2 A specific flow chart of a method for optimizing and simulating total primary productivity of farmland based on particle filtering in one embodiment; Figure 3 is a flowchart for explaining a particle filtering algorithm in one embodiment; Figure 4 is a box plot of posterior parameters of different crop types in one embodiment; Figure 5 The variation trend of the optimized parameters over time in one embodiment; Figure 6 A violin plot of a posterior model evaluation indicator in one embodiment; Figure 7 A schematic diagram showing the comparison between a priori and a posteriori GPP simulations and GPP observations in one embodiment; Figure 8It is a structural block diagram of a device for optimizing the total primary productivity of farmland based on particle filtering in one embodiment; Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] The particle filtering-based optimization simulation method for farmland total primary productivity provided in the embodiment of the present application, in the field of environmental science and ecology, the Chinese name of the BEPS (Biosphere-atmosphere Exchange Process Simulator) model is the biosphere-atmosphere exchange process simulator. The BEPS model is an ecosystem process model used to simulate the material and energy exchange between terrestrial ecosystems and the atmosphere, especially the carbon, water and energy cycles. Figure 1 and Figure 2 As shown, the method comprises the following steps: Step 202: construct a BEPS model and obtain a set of random parameters based on a preset empirical distribution.

[0020] Among them, the random parameter set includes multiple random parameter groups, each random parameter group includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters, the preset empirical distribution includes different sensitive parameters, and after the sensitive parameter group is input into the BEPS model, the generated simulation value is within the preset range.

[0021] Step 204: Drive the BEPS model based on the random parameter set and obtain the corresponding simulation variables.

[0022] Among them, the random parameter groups in the random parameter set are used as particles to drive the BEPS model for simulation, and the simulation variables in this scheme are GPP simulations.

[0023] Step 206, obtaining the observed variables corresponding to the simulated variables and performing iterative operations on the model parameter group based on the observed variables.

[0024] The iterative operations include: Based on the simulation variables corresponding to the observed variables and the model parameters, the likelihood probability of the observed variables occurring when the simulation variables corresponding to the model parameter group occur is calculated; based on the likelihood probability, the weights of the sensitive parameter groups are adjusted, and the adjusted weights are set as the updated weights; based on the updated weights and the corresponding sensitive parameter groups, an updated parameter set is constructed; based on the sensitive parameter groups in the updated parameter set, the biosphere-atmosphere exchange process simulator model is driven respectively and the corresponding updated simulation variables are obtained, that is, according to the observed GPP data, the weights of the model parameter groups in the parameter set are continuously adjusted, and then the GPP of the prior model is corrected by the GPP set average simulated by the adjusted new parameter group, so as to obtain the GPP set average result of the model simulation using the new parameter set.

[0025] Step 208, repeating the iterative operation until the observed variable assimilation is completed.

[0026] In the above-mentioned particle filter-based optimization simulation method for farmland total primary productivity, this method adopts a particle filter algorithm, obtains the simulated value and observed value of the BEPS model, and adjusts the weights in the model parameter group through the observed value to iterate the new simulated value until the simulated value of the BEPS model is iterated to a result that matches the observed value. It does not involve complex matrix transposition and inversion operations, nor does it require professionals to perform time-consuming model accompanying code writing. Compared with other common data assimilation algorithms, it is simpler and more efficient, and more suitable for the optimization of complex ecosystem process models. On the one hand, this method overcomes the conditional restriction that the Kalman filter algorithm requires the processing system to be a linear system; on the other hand, it also breaks through the conditional requirement of the variational method that the system state variables are based on Gaussian distribution. This nonlinear non-Gaussian random pattern feature gives it a wide range of application prospects.

[0027] This method is based on the process-based ecosystem model BEPS. The BEPS model contains multiple mechanism processes such as plant photosynthesis and respiration, and takes into account the close coupling of water, heat and carbon related processes. Therefore, when optimizing the total primary productivity of crops, this method does not directly optimize the variables, but indirectly optimizes the variables by adjusting the model parameter group, which enables this method to analyze the optimization results from a deeper and more mechanistic perspective.

[0028] In one embodiment, the core of the particle filter assimilation method is to use the Monte Carlo method based on Sequential Importance Sample (SIS) to obtain the weight of each particle. Since SIS calculates the importance weight of particles in a recursive manner, the weights of some particles may gradually approach 0 as the number of iterations increases, and they are gradually discarded in subsequent calculations, resulting in particle degradation. At this time, if particles with small weights participate in the calculation, it will increase the unnecessary calculation time of the model. Therefore, it is necessary to deal with particle degradation, including: First, the number of copies of each particle is calculated based on the cumulative copy times (ACT), and then new particles are generated. The ACT calculation formula is as follows: ; in, For Before The number of replications of the update parameter set, is the rounding function, is the number of parameter groups, For the Update parameter groups in The updated weight at time, For the Update parameter groups in The number of times the moment is replicated, then The formula for the number of replications of a particle is: = - ; in, For the Update parameter groups in The number of times a moment is replicated, For Before The number of replications of the update parameter set, For Before The number of replications for each update parameter group; It is worth mentioning that , so = = =N; in, For the Parameter groups in The updated weight at time, For Before The number of replications for each update parameter group, is the rounding function, is the number of parameter groups, For the Update parameter groups in The number of times a moment is replicated; when hour, For the mother particle, unlike the residual resampling method that simply copies the mother particle to generate new particles, the IRR-PF method uses the Halton sequence and exponential function to generate new particles. The formula is as follows: ; in, is the rounding function, is the number of parameter groups, For the Parameter groups in The update parameter group at the moment, For the Update parameter groups in The updated weight at time, To update the offset of the parameter group, the formula is as follows: ; in, By parameter group The resampling generated Parameter groups in The offset of the update parameter group at the moment, is the number of parameter groups, It is used to determine the dispersion of the parameter groups generated by resampling. =0.01.

[0029] In one embodiment, the resampling method can effectively solve the problem of particle degradation. The core is to increase the number of particles with large weights and reduce the number of particles with small weights. Since excessive resampling will increase the amount of calculation and also lose particle diversity to a certain extent, before resampling, it is necessary to evaluate whether the algorithm currently needs to be resampled. The specific evaluation includes: Generally speaking, counting the number of effective particles is a feasible method to measure the degree of particle degradation. The calculation formula is as follows: ; in, is an estimate of the number of valid parameter sets, For the Parameter groups in Update weight at the moment; usually a threshold is set in advance ,when < In this study, resampling is performed. ,in is the number of particles.

[0030] In one embodiment, the core of the particle filter assimilation method is to use a Monte Carlo method based on Sequential Importance Sample (SIS) to obtain the weight of each particle.

[0031] The weight of each particle is calculated using the following formula: ; Among them, among them, For the Update parameter groups in The updated weight at time, For the Update parameter groups in The updated weight at time, For the The parameters are The update parameter group at the moment, for The corresponding simulation variable is For the model The observed variables at time, is the likelihood probability density function. Likelihood probability density The calculation formula is: ; in represents pi; Represents the natural base; represents the standard deviation of the observations, is the likelihood probability density function, For the model The observed variables at time, For the The parameters are The update parameter group at the moment, for The corresponding simulation variable.

[0032] In one embodiment, referring to Figure 2 , the construction process of the particle filter data assimilation algorithm in the BEPS model includes: At k=0, N sampling particles are randomly sampled from the prior probability density distribution, and the weights of all particles are 1 / N at this time; the initialization parameter group and the BEPS model drive data are used to predict the state value at the next moment; at k=k+1, the observation information is used to sample the importance of the particles, calculate the weight of each particle, and normalize it; determine whether particle degradation occurs. If degradation occurs, the particles with large weights are used to generate a new particle set according to the improved residual resampling method, and the particles with small weights are eliminated. The total number of particles remains unchanged before and after resampling. After resampling, the weights of all particles are 1 / N; the optimal estimate of the state quantity at this moment is obtained, and the particles continue to develop and repeat over time until the assimilation is completed. In this application, the particles are the sensitive parameter groups in BEPS.

[0033] This embodiment adopts the above method. From the perspective of innovation, this method is the first to propose and successfully implement an optimization simulation method for total primary productivity of farmland based on particle filtering and the BEPS model, and apply this method to a variety of farmland ecosystems including rice, soybeans, wheat, corn, etc. Finally, combined with observation data, the method is evaluated for the optimization effect of simulating the total primary productivity of farmland of the BEPS model.

[0034] From the algorithmic point of view, this method adopts the particle filter algorithm. This method does not involve complex matrix transposition and inversion operations, nor does it require professionals to perform time-consuming model accompanying code writing. Compared with other common data assimilation algorithms (including Kalman filtering, ensemble Kalman filtering, three-dimensional variational method, four-dimensional variational method, etc.), it is simpler and more efficient, and more suitable for the optimization of complex ecosystem process models. In addition, on the one hand, this method overcomes the conditional restriction that the processing system must be a linear system required by the Kalman filter algorithm; on the other hand, it also breaks through the conditional requirement that the system state variables are based on Gaussian distribution of the variational method. This nonlinear non-Gaussian random pattern feature makes it have a wide range of application prospects.

[0035] Structurally, this method is built on the process-based ecosystem model BEPS. The BEPS model contains multiple mechanism processes such as plant photosynthesis and respiration, and takes into account the close coupling of water, heat and carbon related processes. Therefore, when optimizing the total primary productivity of crops, this method does not directly optimize the variables, but indirectly optimizes the variables by adjusting the model control parameter group, which enables this method to analyze the optimization results from a deeper and more mechanistic perspective.

[0036] Judging from the results, the optimization method based on particle filtering not only significantly improved the simulation effect of the BEPS model on the total primary productivity of farmland, but also revealed the distribution range and temporal variation characteristics of key physiological parameters of different crop types (such as the maximum carboxylation rate at 25°C, etc.), which provided theoretical and practical support for studying the role of agricultural production in global change and carbon emission reduction, and also provided a reference for the subsequent construction of a high spatial resolution regional carbon assimilation system based on the particle filtering algorithm.

[0037] It is worth mentioning that the particle filter algorithm process. The particle filter algorithm includes three steps: initialization sampling, weight update and resampling, such as Figure 3 As shown in the figure, the particle filter method first randomly samples N particles of equal weight from the prior probability distribution of the state using the Monte Carlo method; according to the probability density of the importance of the sampled particles, the weights of the particles are calculated, and the weights of these particles are updated, the weights of the particles in the middle become larger, and the weights of the particles on both sides become smaller. When resampling is required, the number of particles with large weights in the middle is retained and increased, and the particles with small weights are discarded.

[0038] Spatial distribution of 32 farmland observation stations used to build and verify the particle filter assimilation system. The station data comes from FLUXNET, ICOS, AmeriFlux, ChinaFLUX and TPDC. Among them, CN_YC and CN_JR are winter wheat and summer rice rotation. US_CRT station planted soybeans in 2011-2012 and wheat in 2013. Among the farmland stations with corn crop type, US_Ne2 station has been planting corn in odd years and since 2009; US_Ne3 is a rotation that only plants corn in odd years, US-Ro1 station plants corn in odd years, and US_Ro5 station plants corn in even years.

[0039] Reference Figure 4 , Box plot of posterior parameters of different crop types after data assimilation using particle filtering. All parameters are based on the parameter values ​​at the hourly scale after optimization, blue is the median, red is the average, and red dots represent outliers. Overall, , , , , , The average values ​​of these parameters are not much different among all crop types, and are stable at around 3.0, 2.5, 0.6, 4.0, 0.5 and 0.6 respectively. , , and are relatively concentrated, while those of C3 crops are relatively dispersed. and , while the distribution of both is more discrete in C4 (corn) crops, with more outliers.

[0040] Parameters for different crop types The average values ​​after optimization were quite different. The average value is higher than other crops, at 66.22 , while corn, wheat, and rice The average values ​​are 44.39 , 48.91 , 54.54 ; The average value of other crops is 71.86 . C4 (corn) Overall, it is lower than C3 crops, among which C3 crops have a total The average value is 62.07 , 17.68 higher than C4 crops (corn) .

[0041] For the same crop type, the optimized parameters also vary between different sites, with maize having a relatively small variation. The upper and lower bounds of the box plot are 34.26 and 51.65 , the maximum and minimum values ​​are 27.07 and 62.95 . Soybean The dispersion is not much different from that of corn, with the upper and lower limits being 59.67 and 74.37 , the maximum and minimum values ​​are 42.17 and 93.33 The parameter distributions of wheat, rice and other crops are relatively discrete.

[0042] Previous estimates of C4 vegetation Range 11~75 Ground observations show that the maximum Range 105~150 , the maximum wheat is 106.47 , compared with ground-based observations Similarly, soybeans observed on the ground Range 17-190 , rice Range 17-140 In this study, the optimized soybean and rice The ranges are consistent with previous studies.

[0043] Figure 5 The graph shows the changing trend of optimized parameters over time. The horizontal axis is the month, and the vertical axis is the different parameter values; the blue dots represent the model default values, and the error bars are the maximum and minimum values ​​of the model prior range; the red dots represent the average values ​​of optimized parameters in different crop types, and the error bars are the maximum and minimum values ​​of the posterior range.

[0044] like Figure 5 As shown, the posterior , , ,and In four crops, corn, soybean, wheat, and rice, there is little change over time, and the differences between sites are small. and The nitrogen content of leaves and environmental factors (temperature, humidity and radiation, etc.) have obvious seasonal variation trends. has a great impact, so will show significant seasonal and interannual variability, and The changes in leaf nitrogen content ( ) changes similarly. Due to the differences in crop types and planting methods, the posterior and There are also significant differences in seasonal trends for corn, soybeans, wheat, and rice. Corn, soybeans, and rice crops are often planted in the summer, so they have higher posterior values ​​in July-September. Value, corn in July The maximum is reached in August, while soybean and rice are the largest in August. Wheat is planted in spring and early summer, so the posterior The value of is higher in March-June, indicating that the optimized model parameters can show significant physiological differences in photosynthesis of different crops.

[0045] For corn and wheat crops, from January to December, all stations were optimized The average values ​​are greater than the a priori 30 This result shows that for corn and wheat crops, the On the contrary, the prior values ​​for soybean and rice are Value (84.5 ), the setting is too high. In this study, the posterior Reached the maximum value of 86.38 , slightly higher than the prior value, and reaches a minimum of 48.66 in June. . Rice Posterior The average value also reaches its maximum in August, at 75.47. , reaching its minimum in April at 26.83 .

[0046] Figure 6 The figure shows the prior (blue) and posterior (red) simulation effects of the BEPS model on GPP of corn, soybean, wheat, rice and other crops before and after particle filter optimization. Overall, for all crop types, the GPP simulated by the particle filter optimized model is better in capturing the temporal trend of observed GPP (R 2 ) or the capture of observed values ​​(RMSE) showed better results than the prior model, which demonstrated the usefulness and robustness of the particle filter algorithm for GPP simulation and optimization of agricultural ecosystems.

[0047] R between the GPP simulated by the a priori model and the observed GPP for wheat, rice and other crop types 2 The values ​​are concentrated between 0.2 and 0.6, and the RMSE values ​​are concentrated between 0.2 and 0.4. After optimization, the two evaluation indicators of the model are better than the prior simulation, R 2 Increased to between 0.8 and 1.0; RMSE decreased to 0 to 0.2 The prior model simulates soybean better than wheat and rice. 2 The mean and RMSE average is 0.63 and 0.25 ; R between the GPP of the posterior model and the observation 2 The RMSE is also better than that of wheat and rice, with average values ​​of 0.94 and 0.10 respectively. For all C3 plants, the R 2 The average values ​​of and RMSE are 0.89 and 0.11 respectively. , compared with the prior model R 2 Improved by 0.3, RMSE decreased by 0.1 .

[0048] The posterior model has the most obvious effect on improving maize (C4 crops). The distribution of evaluation indicators between the GPP simulated by the prior model and the observations is the most dispersed, which indicates that the GPP simulation effects of the model at different sites vary greatly, among which R 2 The range of is 0.20~0.89, and the range of RMSE is 0.19~0.67 ; Through particle filter optimization, the posterior simulation achieved good simulation results, R 2 The range is 0.80~0.99, and the RMSE range is 0.08~0.31 .

[0049] Figure 7 The time series of daily GPP of corn, soybean, wheat, rice and other crops from January to December are shown, with the horizontal axis representing the month; blue represents the prior GPP, red represents the posterior GPP, and black represents the observed GPP. The results show that although the GPP of the prior model can capture the seasonal variation trend of flux GPP, it is seriously underestimated in all crops, especially in the growing season, and cannot reach the peak value of flux GPP in the growing season; the posterior model can not only better capture the seasonal variation trend of observed GPP, but also alleviate the deviation of the prior model. It shows that the particle filter algorithm can optimize the relevant parameters such as photosynthesis, energy, water, and stomatal conductance in the BEPS model to obtain more accurate GPP simulation results. In addition, the model before optimization cannot distinguish the difference between the peak values ​​of GPP of C3 and C4 crops well, while the optimized model effectively realizes the seasonal variation simulation of GPP of C3 and C4 crops by adjusting the relevant physiological parameters of C3 and C4.

[0050] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0051] Based on the same inventive concept, the embodiment of the present application also provides a particle filter-based farmland total primary productivity optimization simulation device for implementing the above-mentioned particle filter-based farmland total primary productivity optimization simulation method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the particle filter-based farmland total primary productivity optimization simulation device provided below can refer to the limitations of the particle filter-based farmland total primary productivity optimization simulation method above, and will not be repeated here.

[0052] In one embodiment, Figure 8 As shown, a particle filtering-based farmland total primary productivity optimization simulation device is provided, comprising: a random parameter construction module, a simulation variable calculation module, a model parameter iteration module and an iterative assimilation output module, wherein: A random parameter construction module is used to construct a biosphere-atmosphere exchange process simulator model and obtain a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each of which includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; A simulation variable calculation module is used to drive the biosphere-atmosphere exchange process simulator model based on a random parameter set and obtain the corresponding simulation variables; A model parameter iteration module is used to obtain the observed variables corresponding to the simulation variables and perform iterative operations on the model parameters based on the observed variables, wherein the iterative operations include: calculating the likelihood probability of the observed variables occurring when the simulation variables corresponding to the model parameter group occur based on the observed variables and the simulation variables corresponding to the model parameters; adjusting the weights of the sensitive parameter group based on the likelihood probability, and setting the adjusted weights as the updated weights; constructing an updated parameter set based on the updated weights and the corresponding sensitive parameter group; and driving the biosphere-atmosphere exchange process simulator models based on the updated parameter set and obtaining the corresponding updated simulation variables; The iterative assimilation output module is used to repeat the iterative operation until the assimilation of the observed variables is completed.

[0053] In one embodiment, the iterative assimilation output module is further used to: count the estimated value of the number of effective parameters, the formula is as follows: ; in, is an estimate of the number of valid parameter sets, For the Parameter groups in The update weight at the moment; comparing the estimated value of the number of valid parameter groups with the preset critical threshold; if the estimated value of the number of valid parameter groups is less than the preset critical threshold, resampling the update parameter set whose number of replications is not zero based on the Holden sequence and the exponential function to obtain the resampled parameter set; repeating the iterative operation until the assimilation of the observed variables is completed.

[0054] In one embodiment, the iterative assimilation output module is further used to calculate the number of replications of each update parameter group based on the cumulative number of replications. The cumulative number of replications is calculated using the following formula: ;in, For Before The number of replications of the update parameter set, is the rounding function, is the number of parameter groups, For the Parameter groups in The updated weight at time, For the Update parameter groups in The number of replications at a moment is calculated as follows: = - , obtaining a resampling parameter group based on the number of copies of the update parameter group.

[0055] In one embodiment, the iterative assimilation output module is further used to: when the number of replications of the update parameter set is greater than a preset value, resample the update parameter set whose number of replications is not zero based on the Holden sequence and the exponential function to obtain a resampled parameter set, and the formula is as follows: , , , ,in, is the rounding function, is the number of parameter groups, For the The parameters are The update parameter group at the moment, For the Update parameter groups in The updated weight at time, To update the offset of the parameter group, the formula is as follows: ;in, Reason The resampling generated Parameter groups in The offset of the update parameter set at the moment, is the number of parameter groups, Used to determine the dispersion of the parameter groups generated by resampling.

[0056] In one embodiment, the random parameter construction module is further used to: construct a biosphere-atmosphere exchange process simulator model and obtain a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each random parameter group includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; Each module in the above-mentioned particle filter-based farmland total primary productivity optimization simulation device can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0057] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for optimizing the total primary productivity of farmland based on particle filtering is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0058] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0059] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0060] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0061] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0062] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0063] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0064] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A particle filtering-based optimization simulation method for total primary productivity of farmland, characterized in that: The method comprises: Constructing a biosphere-atmosphere exchange process simulator model and obtaining a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each random parameter group includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; Based on the parameter groups composed of random values ​​of sensitive parameters, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding simulation variables are obtained; Obtaining observed variables corresponding to the simulation variables and performing an iterative operation on the model parameters based on the observed variables, the iterative operation comprising: Based on the observed variables and the simulated variables corresponding to the model parameters, calculating the likelihood probability of the observed variables occurring when the simulated variables corresponding to the model parameter group occur; Adjust the weights of the sensitive parameter group based on the likelihood probability, and set the adjusted weights as update weights; Constructing an update parameter set based on the update weights and the corresponding sensitive parameter groups; Based on the updated parameter set, the biosphere-atmosphere exchange process simulator models are driven respectively and the corresponding updated simulation variables are obtained; The iterative operation is repeated until the observed variables are assimilated.

2. The particle filtering-based optimization simulation method for total primary productivity of farmland according to claim 1, characterized in that: The method further comprises: Based on the observed variables and simulated variables, the likelihood probability of the observed variable occurring when the simulated variable occurs is calculated: ； in, is the likelihood probability of the observed variable occurring when the simulated variable occurs, represents pi; Represents the natural base; represents the standard deviation of the observations, express Moment parameter groups, express The corresponding simulation variable is express The observed variable corresponding to the moment; Based on the likelihood probability of the observed variable occurring, updating the weight of the parameter group; The updated weight of the parameter group The likelihood probability is calculated to ensure that the sum of the weights of all parameter groups at each moment is 1, and the formula is as follows: ； in, is the likelihood probability of the observed variable occurring when the simulated variable occurs, express Moment The updated weights of the parameter groups are express Moment The updated weights of the parameter groups.

3. The particle filtering-based optimization simulation method for total primary productivity of farmland according to claim 1, characterized in that: The method further comprises: The estimated value of the number of statistically effective parameter groups is as follows: ； in, is an estimate of the number of valid parameter sets, For the Parameter groups in Update weight at each moment; comparing the estimated number of valid parameter groups with a preset critical threshold; If the estimated value of the number of valid parameter groups is less than a preset critical threshold, it is determined to perform parameter set resampling and obtain a resampled parameter set.

4. The particle filtering-based optimization simulation method for total primary productivity of farmland according to claim 3, characterized in that: The determining to perform parameter set resampling comprises: The cumulative number of replications of the statistical update parameter set is calculated using the following formula: ； in, For Before The cumulative number of replications for the update parameter group, is the rounding function, is the number of parameter groups, For the Update parameter groups in Update weight at each moment; Based on the cumulative number of replications, the number of replications for each parameter group is calculated using the following formula: = - ； in, For the Update parameter groups in The number of times a moment is replicated, For Before The cumulative number of replications for the update parameter group, For Before The cumulative number of replications for each update parameter group; Whether to resample the parameter group is determined based on the number of copies of the update parameter group.

5. The particle filtering-based optimization simulation method for total primary productivity of farmland according to claim 4, characterized in that: The determining whether to resample the parameter group based on the number of replications of the update parameter group includes: When the estimated value of the number of valid parameter groups is less than a preset critical threshold, the update parameter group whose number of replications is not zero is resampled based on the Holden sequence and the exponential function to obtain a resampled parameter group, and the formula is as follows: ； in, is the rounding function, express Moment parameter groups, For the Update parameter groups in The number of times a moment is replicated, To update the offset of the parameter group, the formula is as follows: ； in, is obtained by resampling based on the updated parameter group Moment The offset corresponding to each parameter group, To update the number of parameter groups, Used to determine the dispersion of the sampling parameter groups.

6. A particle filtering-based farmland total primary productivity optimization simulation device, characterized in that: The device comprises: A random parameter construction module is used to construct a biosphere-atmosphere exchange process simulator model and obtain a random parameter set based on a preset empirical distribution, wherein the random parameter set includes a plurality of random parameter groups, each of which includes sensitive parameters corresponding to different biosphere-atmosphere exchange process simulator models and initial weights corresponding to the sensitive parameters; A simulation variable calculation module is used to drive the biosphere-atmosphere exchange process simulator model based on a random parameter set and obtain the corresponding simulation variables; A model parameter iteration module is used to obtain the observed variables corresponding to the simulation variables and perform iterative operations on the model parameters based on the observed variables, wherein the iterative operations include: calculating the likelihood probability of the observed variables occurring when the simulation variables corresponding to the model parameter group occur based on the observed variables and the simulation variables corresponding to the model parameters; adjusting the weights of the sensitive parameter group based on the likelihood probability, and setting the adjusted weights as the updated weights; constructing an updated parameter set based on the updated weights and the corresponding sensitive parameter group; and driving the biosphere-atmosphere exchange process simulator models based on the updated parameter set and obtaining the corresponding updated simulation variables; The iterative assimilation output module is used to repeat the iterative operation until the assimilation of the observed variables is completed.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images