CN-121504531-B - Farmer agricultural resource demand prediction method and system based on time sequence behavior analysis

CN121504531BCN 121504531 BCN121504531 BCN 121504531BCN-121504531-B

Abstract

The invention discloses a method and a system for forecasting peasant household agricultural demands based on time sequence behavior analysis, which relate to the field of agricultural digital management, aim at the situation that a county store network and a cooperative agent all adopt and finish order submission before an agricultural node and are constrained by an early supply period, a packaging step length, a minimum order quantity and a quality guarantee period together, solve the problem that an order main body is difficult to form a recalculation demand process depiction under a discrete contact event, generate an order main body granularity continuous time demand signal sequence by taking an order main body identifier and a target agricultural identifier as main keys, jointly solve the order submission time and the integral order quantity by on-line estimation of a demand arrival intensity function and an uncertainty parameter sequence, and realize rolling calibration according to transaction data, unfilled data and loss reporting data.

Inventors

RUI HONG
WENG YUANKAI

Assignees

安徽直农科技服务有限公司

Dates

Publication Date: 20260512
Application Date: 20260109

Claims (10)

1. The method for predicting the agricultural demand of the farmer based on the time sequence behavior analysis is characterized by comprising the following steps: S1, collecting target agricultural stock and supply advance of an order body identifier corresponding to a target agricultural identifier, extracting a peasant household behavior event time stamp, executing consistency verification, and generating an order body granularity continuous time demand signal sequence according to a behavior event contribution coefficient and time attenuation transfer; S2, forming time intervals according to adjacent sampling time stamps in a supply advance period coverage window based on a continuous time demand signal sequence of the granularity of an order main body, and counting interval arrival counts, wherein the interval arrival counts correspond to the number of time intervals in which the payment completion time of an order for the delivery falls; s3, solving order submitting time and integer order quantity under the constraints of packaging step length, minimum order quantity and quality guarantee period according to a discretization sequence of a demand arrival intensity function and an uncertainty parameter sequence by taking expected total cost corresponding to the unit price of loss of stock, unit price of holding cost and unit price of overdue discount as targets, and outputting an order instruction; and S4, collecting the data of the success and failure of the stock and the loss report after the execution of the order, establishing a cost decomposition account book, updating the demand arrival intensity function discretization sequence, the uncertainty parameter sequence and three types of unit cost according to the account book error, wherein the three types of unit cost are the stock and loss unit price, the holding cost unit price and the outdated discount unit price, and outputting a calibration result set.
2. The method for predicting agricultural demands of farmers based on time series behavior analysis as recited in claim 1, wherein: In step S1, a one-to-one mapping table of original material codes and target agricultural material identifiers is established according to commodity main data, and consultation records, price inquiring records, order placing records, payment completion records and return records which are successfully mapped are extracted as a peasant household behavior event time stamp sequence, a behavior event type takes contact point enumeration table to fix category codes, and peasant household behavior event time stamp takes service occurrence time.
3. The method for predicting agricultural demands of farmers based on time series behavior analysis as recited in claim 2, wherein: In step S1, the coverage map value is determined by the main body field of the last order in the supply lead time window before the order calculation starting time, the coverage map value is determined by the main body field of the client main data when the main body field of the coverage map is lost, the duplication removal is performed by the event source pipelining main key and the earliest writing record is reserved, the time sequence verification is performed according to the constraint that the payment completion time stamp is not earlier than the order time stamp and the return time stamp is not earlier than the payment completion time stamp, and the peasant behavior event time stamp sequence arranged in ascending order of the time stamps is formed.
4. A method for predicting agricultural needs of a farmer based on time sequential behavior analysis as recited in claim 3, wherein: in step S2, forming time intervals by using adjacent sampling time stamps in a supply advance period coverage window, counting the interval arrival counts, wherein the interval arrival counts correspond to the quantity of time intervals in which the payment completion time of the order form falls, generating a prior mean value of a demand arrival intensity function by using exponential mapping according to the value of a continuous time demand signal sequence of the granularity of an order main body at the starting point of the time intervals, and estimating a target agricultural intensity reference coefficient and a target agricultural signal sensitivity coefficient by using Poisson log likelihood iteration.
5. The method for predicting agricultural demands of farmers based on time series behavior analysis as recited in claim 4, wherein: In step S2, constructing the mixed likelihood of the conventional poisson component and the heavy tail component according to the interval arrival count, calculating a temperature factor according to the posterior probability of the conventional poisson component, wherein the temperature factor is used for measuring the trusted proportion of the single time interval observation entering update, the value range is larger than zero and not larger than one, the maximum update mixing coefficient is expected, then executing conjugate update on the shape parameter sequence and the rate parameter sequence by the temperature factor, outputting the uncertainty parameter sequence, and taking the temperature factor and the mixing coefficient as audit fields to be output together.
6. The method for predicting agricultural demands of farmers based on time series behavior analysis as recited in claim 5, wherein: In step S3, the arrival time is determined by the order submitting time and the supply lead time, and the supply lead time coverage window time interval index set and the quality guarantee period coverage window time interval index set are generated according to the arrival time distribution of the demand in each time interval is determined according to the shape parameter sequence and the rate parameter sequence, and the supply lead time coverage window demand distribution and the quality guarantee period coverage window demand distribution are calculated by adopting discrete convolution recurrence between areas, and the recurrence process cuts off limited support by numerical precision parameters.
7. The method for predicting agricultural demands of farmers based on time series behavior analysis as recited in claim 6, wherein: In step S3, an integer order quantity set is constructed according to a packing step multiple rule and a minimum order quantity segmentation rule, and expected total cost is formed according to an order loss unit price, a holding cost unit price and an expiration discount unit price, candidate order submission time and integer order quantity are enumerated after the upper limit of the order quantity is determined according to the accumulated probability and cost ratio of the window demand quantity distribution covered by the expiration date, an order instruction corresponding to the minimum expected total cost is selected, and three types of expected amount decomposition fields are output.
8. The method for predicting agricultural demands of farmers based on time-series behavior analysis as recited in claim 7, wherein: In step S4, a time interval sequence number is formed along with a sampling time stamp sequence, the interval actual delivery number, the interval lack-of-stock unsatisfied number, the interval actual demand number and the interval loss reporting number are summarized according to the time interval sequence number, the interval actual demand number is the sum of the interval actual delivery number and the interval lack-of-stock unsatisfied number, the window total amount is formed on the supply lead period coverage window and the quality guarantee period coverage window, and the cost resolution account book record expected number, actual number and amount caliber error are established.
9. The method for predicting agricultural demands of farmers based on time series behavior analysis as recited in claim 8, wherein: In step S4, a section residual error is constructed according to the difference between the actual number of sections and the predicted number of sections, an updating amplitude factor is calculated by decomposing the total error and the sum scale by cost, an exponential mapping updating shape parameter sequence is adopted, a demand arrival intensity function discretization sequence is recalculated by the shape parameter sequence and the rate parameter sequence, meanwhile, the positive value of logarithmic difference driving is executed for the unit price of the loss of stock, the unit price of the holding cost and the unit price of the outdated discount according to three types of actual loss amounts, and a calibration result set is output for the next period to read.
10. A peasant household agricultural demand prediction system based on time sequence behavior analysis, for implementing the method of any of claims 1-9, comprising a signal generation unit, an intensity estimation unit, an order solving unit and a closed loop calibration unit; The signal generation unit acquires the target agricultural stock and the supply advance period of the agricultural material identification corresponding to the target agricultural material identification, extracts the agricultural behavior event time stamp and executes consistency verification, and generates an order body granularity continuous time demand signal sequence according to the behavior event contribution coefficient and the time attenuation transfer; Estimating a demand arrival intensity function according to a demand signal sequence of continuous time of granularity of an order body, wherein the demand arrival intensity function is expected of arrival times of the order in unit time, and updating an uncertainty parameter sequence by a temperature factor, the temperature factor is a trusted proportion for measuring the observation and the update of a single time interval calculated according to the posterior probability of a conventional poisson component, the value range is larger than zero and not larger than one, the uncertainty parameter sequence is a shape parameter sequence and a rate parameter sequence corresponding to each time interval, and the demand arrival intensity function discretization sequence and the uncertainty parameter sequence are output; the order solving unit forms expected total cost by using the unit price of loss of stock, the unit price of holding cost and the unit price of overdue discount, solves order submitting time and integer order quantity under the constraint of packaging step length, minimum order quantity and quality guarantee period according to the discretization sequence of the demand arrival intensity function and the uncertainty parameter sequence, and outputs order instructions; The closed loop calibration unit collects the delivery data, the unfit data and the loss reporting data after the order instruction is executed, establishes a cost decomposition account book, and achieves an intensity function discretization sequence, an uncertainty parameter sequence and three types of unit cost according to the account book error updating requirement, wherein the three types of unit cost are the unit price of the missing goods loss, the unit price of the holding cost and the unit price of the overdue price, and outputs a calibration result set.

Description

Farmer agricultural resource demand prediction method and system based on time sequence behavior analysis Technical Field The invention relates to the field of agricultural digitization, in particular to a method and a system for forecasting agricultural demands of farmers based on time sequence behavior analysis. Background In the scenes of county agricultural material distribution, cooperative system acquisition, supply and marketing system store network, platform agricultural material service and the like, goods ordering often uses a multi-stage supply chain formed by stores, warehouses and suppliers as parameters to organize goods sources, and the goods preparation, goods arrival inspection and allocation distribution are required to be completed before agricultural time nodes such as sowing, topdressing, pest control and the like. Because agricultural products cover fertilizers, pesticides, seeds, auxiliary agents and the like, the agricultural products have various products and specifications, and particularly have obvious specification and concentration differences, and are greatly influenced by the fluctuation of the supply period, the packaging, the minimum ordering amount, the batch-to-goods capability, the quality guarantee period, the storage conditions and other operation constraints. The existing business system usually deposits historical sales, orders, prices and stock data in the stock or order system, forms an estimated value of future demand by means of statistical prediction or machine learning prediction, calculates order advice by means of rules such as safety stock, replenishment points, service level coefficients or economic order quantity, and selects a predicted value by means of probability distribution or quantile prediction and combining cost preference in some applications. Meanwhile, in reality, the agricultural material demand is determined by historical transaction, and can be displayed in advance by the sequential triggering of behaviors and events such as consultation, browsing, price inquiring, agricultural technology service record, land crop structure, weather disturbance and the like of farmers, and the characteristics of strong seasonality, strong burstiness, unstable annual repeat purchase interval and the like are also presented. Because the demand value output by the prediction module is often inconsistent with the discrete constraint of the purchase execution list, the ordering personnel still need to convert, split and combine the prediction result, the existing quantity, the in-transit quantity and the supply limit according to experience to form an available ordering process which is difficult to directly execute. However, in the prior art, it is difficult to coordinate the demand prediction result obtained based on the historical transaction and the behavior signal with the supply advance period and the discrete operation constraints such as packaging, ordering constraint, quality guarantee period and the like under the condition of not depending on a large number of artificial thresholds and experience correction, so that the prediction output can be stably and repeatedly converted into the order submitting time and the order quantity which can be directly executed, and the reason is that the prediction model mostly takes the continuous demand quantity as the output, the order execution is constrained by the discrete batch and the arrival rhythm, and the demand in the advance period can generate time-varying uncertainty due to the agricultural time window, weather and pest events, so that the fixed order supplement point or the static safety stock rule is difficult to cover the changeable situation. When the coordination is absent, the ordering decision is easy to deviate in an agricultural time window, and the ordering decision is represented by key goods arrival lag, mispairing specification or excessive stock, thereby causing the consequences of lack of stock, stock backlog and overdue price, rising of fund occupation, increase of cross-store allocation and emergency replenishment frequency and the like. Aiming at the business requirements and constraint conditions, a method and a system for predicting the agricultural requirements of farmers based on time sequence behavior analysis are provided. Disclosure of Invention The embodiment of the invention provides a method and a system for forecasting peasant household agricultural demand based on time sequence behavior analysis, wherein the method takes an order main body identifier and a target agricultural identifier as main keys, collects the existing quantity and the early supply period of the target agricultural, extracts event time stamps of consultation, price inquiry, order ordering, payment completion and the like, generates an order main body granularity continuous time demand signal sequence, estimates a demand arrival intensity function and an uncertainty parameter in an early supply period covera