Pesticide production order and inventory collaborative management method

Abstract

The invention relates to the technical field of production management, and discloses a pesticide production order and inventory collaborative management method, and the method comprises the steps: obtaining a multi-source heterogeneous data set of pesticides, and carrying out the feature extraction; carrying out demand prediction by adopting a multi-level feature fusion network model; obtaining current finished product inventory data, and carrying out inventory state multi-dimensional evaluation and demand gap calculation; calculating a production priority score, determining a production batch, and arranging a production time sequence; calculating various raw material demanded quantities, and evaluating the raw material supply capability; a hierarchical optimization strategy is adopted for collaborative decision making; using a disturbance type discrimination method to identify a disturbance event, using an influence propagation analysis method to evaluate disturbance influence, generating a coping scheme, and updating a production plan and an order delivery plan; according to the method, the limitation of independent decision making of the production plan and the inventory control in the traditional method is broken through by establishing the collaborative decision-making model of production inventory bidirectional coupling.

Description

Technical Field

[0001] This invention relates to the field of production management technology, and more specifically, to a method for collaborative management of pesticide production orders and inventory. Background Technology

[0002] As a crucial material for ensuring agricultural production, pesticides exhibit seasonal fluctuations, regional differences, and product diversification in market demand. Pesticide manufacturers face multiple challenges, including difficulty in forecasting order demand, a lack of coordination between production planning and inventory control, and sluggish raw material supply chain responses. Traditional order management and inventory management systems often operate independently, lacking effective coordination mechanisms, leading to frequent adjustments to production plans, frequent inventory backlogs or stockouts, and persistently high production costs.

[0003] Currently, pesticide manufacturers face the following prominent problems in order and inventory management: First, order demand forecasting lacks the ability to deeply mine multi-source heterogeneous data. Traditional methods mainly rely on simple statistical analysis of historical sales data, making it difficult to capture the complex impact of multiple factors such as agricultural production cycles, climate change, pest outbreaks, and policy adjustments on pesticide demand. Forecast accuracy is generally low, especially for sudden surges in demand caused by pest outbreaks, which are almost impossible to predict in advance. Second, there is a lack of dynamic coordination between production scheduling and inventory allocation. Production departments often schedule production based on fixed production cycles and batch sizes, while inventory management departments make replenishment decisions based on safety stock thresholds. This information asymmetry and asynchronous decision-making between the two leads to insufficient capacity and delivery delays during peak order periods, and inventory backlogs and capital tied up during off-peak periods. Third, there is insufficient coordinated optimization between raw material and finished product inventory. Pesticide production involves the proportioning and reaction processes of various chemical raw materials. The procurement cycle, storage conditions, and proportions of raw materials are closely related to the production plan and inventory strategy of finished products, but existing systems struggle to achieve coordinated optimization between raw material and finished product inventory.

[0004] Furthermore, due to the unique characteristics of pesticide products—including their hazardous chemical properties, strict shelf-life limitations, and stringent environmental regulations—the collaborative management of orders and inventory is particularly important and complex. In actual production, urgent orders and order changes occur frequently, especially when pest outbreaks occur in a region, leading to a surge in pesticide orders within a short period. Companies need to respond quickly, but existing systems primarily rely on manual experience and localized adjustments to handle dynamic disturbances, lacking systematic support and making it difficult to respond rapidly to urgent needs while ensuring the delivery of normal orders. Therefore, a collaborative management method for pesticide production orders and inventory is needed to address these technical challenges. Summary of the Invention

[0005] This invention provides a method for collaborative management of pesticide production orders and inventory, which solves the technical problems of inaccurate forecasting, insufficient coordination, and weak responsiveness in pesticide production order and inventory management in related technologies.

[0006] This invention provides a method for collaborative management of pesticide production orders and inventory, comprising: A standardized feature vector set is obtained by merging multi-source heterogeneous datasets of pesticides using data interface calling technology and extracting features. Demand prediction is performed using a multi-level feature fusion network model based on a standardized feature vector set, resulting in a fine-grained demand prediction matrix. Obtain current finished product inventory data, combine it with fine-grained demand forecasting matrix to conduct multi-dimensional assessment of inventory status and calculate demand gap, and obtain the inventory status and demand gap value of each pesticide variety. Based on inventory status and demand gap, calculate production priority score, determine production batch size, and use reverse scheduling method to arrange production sequence to obtain preliminary production plan; Based on the preliminary production plan, the bill of materials expansion method is used to calculate the demand for various raw materials, the time matching analysis method is used to assess the raw material supply capacity, and a raw material procurement plan is generated. Based on the preliminary production plan and raw material procurement plan, a hierarchical optimization strategy is adopted to make collaborative decisions and obtain a collaboratively optimized production plan. Based on the collaboratively optimized production plan, identify disturbance events, use impact propagation analysis to assess the impact of disturbances, generate response plans, and update the production plan and order delivery plan.

[0007] In a preferred embodiment, the feature extraction includes: We acquired agricultural production cycle datasets and pest and disease monitoring datasets, and used a growth stage mapping method to divide the agricultural production cycle datasets into stages, namely, sowing period, tillering period, jointing period, heading period, grain filling period, and maturity period. A demand correlation matrix between growth stages and pesticide varieties was established based on the stage segmentation results. Risk quantification scoring method is used to assess the risk of pest and disease monitoring dataset. The pest and disease risk index is calculated by weighted summation based on the degree of temperature deviation, the degree of humidity satisfaction, and the number of consecutive rainfall days, resulting in a standardized feature vector set.

[0008] In a preferred embodiment, the demand forecasting includes: Based on the standardized feature vector set, a time series analysis is performed on the standardized feature vector set using a differential autoregressive moving average model. The autoregressive term captures the short-term autocorrelation of demand, the moving average term captures the impact of random disturbances, and the differential operation eliminates non-stationary trends to obtain the detrended prediction component. A periodic superposition model is used to perform periodic analysis on the standardized feature vector set. Based on the identified principal period parameters, a trigonometric function fitting method is used to construct a periodic fluctuation function to obtain the periodic prediction component. The basic demand forecast value is obtained by adding the detrended forecast component and the cyclical forecast component.

[0009] In a preferred embodiment, the demand forecasting further includes: Based on the basic demand forecast, a regional differentiation adjustment model is used to adjust the basic demand forecast regionally. The baseline demand distribution matrix is ​​obtained by performing matrix multiplication between the periodic correlation matrix and the regional demand time window distribution. The risk-driven demand adjustment vector is used as the adjustment coefficient to adjust the baseline demand distribution matrix. For high-risk spatiotemporal regions, the baseline demand value of this region for this time period is multiplied by the risk adjustment coefficient. The cross-regional aggregation method is used to sum the demand values ​​of each region according to the pesticide variety dimension to obtain the demand adjustment value after considering regional differences and meteorological risks.

[0010] In a preferred embodiment, the calculation of the demand gap value includes: Based on a fine-grained demand forecasting matrix and forecast confidence, a dynamic safety stock calculation method is used to measure the uncertainty of forecast confidence. The safety stock level is set as the service level coefficient multiplied by the demand standard deviation multiplied by the square root of the replenishment cycle and then multiplied by the confidence adjustment factor. The confidence adjustment factor is equal to 1 plus the confidence adjustment coefficient multiplied by 1 minus the forecast confidence. When the forecast confidence is high, the safety stock is calculated according to the traditional formula; when the forecast confidence is low, the safety stock is increased accordingly. The demand gap is set as the cumulative demand forecast for the coming weeks plus safety stock minus effective inventory.

[0011] In a preferred embodiment, the multidimensional assessment of inventory status includes: Based on current finished goods inventory data, an inventory validity assessment method is used to classify the finished goods inventory data. Based on the remaining shelf life, inventory with a longer remaining shelf life is classified as normal inventory, inventory with a moderate remaining shelf life is classified as near-expiry inventory, and inventory with a short remaining shelf life is classified as slow-moving inventory. The effective inventory level is set as the sum of normal inventory and near-expiry inventory. A turnover risk identification method is used to assess the risk of inventory data. If the weighted average inventory age exceeds the historical average inventory age and the inventory turnover rate is lower than the historical average turnover rate, then this product is considered to have turnover risk.

[0012] In a preferred embodiment, the formulation of the preliminary production plan includes: Based on the demand gap value and inventory allocation priority, a multi-factor comprehensive scoring method is used to prioritize the demand gap value. The production priority score is calculated by weighted summation of demand gap normalization, time urgency, switching cost, and profit margin. The production priority score is optimized by using an economic production batch model. The optimal production batch is determined based on the demand gap and production cost parameters. If the demand gap is small, the actual batch is equal to the demand gap. If the demand gap is moderate, the actual batch is equal to the economic batch. If the demand gap is large, the demand gap is split into multiple batches.

[0013] In a preferred embodiment, the generation of the raw material procurement plan includes: Based on the preliminary production plan and product formula data, the bill of materials expansion method is used to calculate the raw material requirements of the preliminary production plan. The planned production quantity of each product is multiplied by the proportion coefficient of each raw material in the formula and summed according to the type of raw material to obtain the total demand of each type of raw material. The time-matching analysis method is used to assess the supply capacity of the total demand for raw materials. The latest order time for raw materials is calculated as the planned start time of the earliest production task using this raw material minus the supply cycle and the raw material warehousing and inspection time. If the regular supply cycle exceeds the time difference but the shortest supply cycle can meet the demand, there is a supply risk and expedited supply is required. If the shortest supply cycle still exceeds the time difference, the raw materials cannot be supplied on time and the preliminary production plan needs to be adjusted.

[0014] In a preferred embodiment, the updating of the production plan and order delivery plan includes: Based on the collaboratively optimized production plan and real-time production data, a disturbance impact assessment method is used to identify disturbances in the real-time production data. The scope of tasks that need to be adjusted is determined by calculating the degree of impact of disturbance events on each production task. An adaptive scheduling method is used to optimize the scheduling based on the disturbance impact assessment results. The corresponding scheduling strategy is selected according to the disturbance type and impact range. For equipment failure disturbances, a task reassignment strategy is adopted; for raw material shortage disturbances, a task postponement strategy is adopted; and for urgent order disturbances, a task insertion strategy is adopted. The production plan and order delivery plan are updated accordingly.

[0015] In a preferred embodiment, a computer-readable storage medium is provided for storing computer-readable instructions that, when read by a computer, enable the execution of a pesticide production order and inventory collaborative management method as described above.

[0016] The beneficial effects of this invention are as follows: By constructing a demand forecasting mechanism that integrates multi-source data fusion, it integrates multi-dimensional information such as historical order data, agricultural production cycle data, weather forecast data, and pest and disease monitoring data. It employs a multi-level feature fusion network for deep fusion and nonlinear mapping, establishing a complex correlation between order demand and multi-dimensional influencing factors, thereby improving the accuracy of demand forecasting. In particular, by introducing pest and disease risk assessment and sudden demand identification mechanisms, it effectively solves the technical problem of traditional methods' difficulty in predicting sudden demand. Simultaneously, it innovatively proposes an uncertainty quantification method based on prediction confidence, providing risk measurement indicators for subsequent decision-making. By establishing a collaborative decision-making model that couples production and inventory, this approach breaks through the limitations of independent decision-making between production planning and inventory control in traditional methods, achieving integrated optimization of orders, production, finished goods inventory, and raw material inventory. It innovatively incorporates forecast confidence into safety stock calculation, establishing a confidence-driven dynamic safety stock model that allows safety stock levels to adaptively adjust based on forecast uncertainty, balancing inventory costs and service levels. A hierarchical optimization strategy decomposes the complex multi-objective optimization problem into three sub-problems: production batch timing optimization, inventory replenishment strategy optimization, and raw material procurement batch optimization. Global collaboration is achieved through iterative coordination, effectively reducing computational complexity and ensuring the algorithm's real-time performance. Simultaneously, it fully considers practical constraints such as the shelf-life constraints of pesticide products, the storage capacity limitations of hazardous chemicals, and production line switchover costs, improving the feasibility and economy of the solution. Attached Figure Description

[0017] Figure 1 This is a flowchart of a pesticide production order and inventory collaborative management method according to the present invention. Detailed Implementation

[0018] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a method for collaborative management of pesticide production orders and inventory, such as... Figure 1 As shown, it includes: Step 1: Use data interface calling technology to obtain multi-source heterogeneous datasets of pesticides, merge them for feature extraction, and obtain a standardized feature vector set; Based on the enterprise order management system, agricultural information platform, meteorological data interface, and plant protection monitoring platform, and using data interface calling technology and web crawling technology, historical order datasets, agricultural production cycle datasets, meteorological forecast datasets, and pest and disease monitoring datasets are obtained to obtain the original multi-source heterogeneous data set.

[0020] Based on order records from the past three years stored in the enterprise order management system, database queries were used to extract fields such as order time, pesticide variety code, order quantity, customer region, and delivery cycle to obtain a historical order data table. Based on this historical order data table, a time window sliding method was used to statistically analyze the historical demand for each pesticide variety on a weekly basis, resulting in a time-series demand sequence. Based on this time-series demand sequence, a seasonal decomposition method was used to decompose the sequence into trend, seasonal, and random components, yielding trend and periodic feature vectors. The trend feature vector reflects the long-term growth or decline trend of demand, while the periodic feature vector reflects the seasonal fluctuation pattern of demand. Based on the periodic feature vector, a fast Fourier transform was used to identify the main periodic components, obtaining the principal period parameters of demand. For imidacloprid insecticides, the identified principal period is the peak demand period from May to July each year, corresponding to the critical period for pest control from the rice tillering to heading stage.

[0021] Based on planting plan data released by the Ministry of Agriculture and Rural Affairs and agricultural information platforms of various provinces, an API interface was used to obtain information on rice planting area, main varieties, sowing time distribution, and expected harvest time in provinces along the Yangtze River and in South China, resulting in an agricultural production cycle data table. Based on this data table and an agronomic knowledge base, a growth stage mapping method was used to divide the rice growth cycle into six stages: sowing, tillering, jointing, heading, grain filling, and maturity. The start and end times of each stage and the corresponding major pests and diseases were labeled, resulting in a growth stage labeled sequence. Based on the growth stage labeled sequence and a pesticide variety functional attribute table, an association rule mining method was used to establish a demand correlation matrix between growth stages and pesticide varieties. This matrix characterizes the demand intensity coefficient of each pesticide variety at each growth stage. Based on the differences in sowing time among provinces, a time offset overlay method was used to overlay the growth stage sequences of each province along the time axis, resulting in a regionally differentiated demand time window distribution map. This distribution map reflects the staggered peak demand characteristics of different regions, providing a basis for the timing of production planning.

[0022] Based on the medium-term weather forecast data released by the meteorological bureau, a meteorological data interface was used to obtain the forecast values ​​of meteorological elements such as weekly average temperature, weekly cumulative rainfall, and weekly average relative humidity for major rice-producing areas over the next eight weeks, resulting in a meteorological forecast data matrix. Based on the meteorological forecast data matrix and a threshold table for meteorological conditions for pest and disease occurrence, a threshold discrimination method was used to identify spatiotemporal regions that meet the conditions for the occurrence of specific pests and diseases. The threshold table for meteorological conditions for pest and disease occurrence was established based on the knowledge of plant protection experts and historical records of pest and disease outbreaks. For example, the high incidence conditions for brown planthoppers are a temperature of 25 to 28 degrees Celsius and a relative humidity greater than 80%, while the high incidence conditions for rice blast are a temperature of 20 to 25 degrees Celsius and more than three consecutive days of overcast and rainy weather. Based on the identified high-risk spatiotemporal regions, a risk quantification scoring method was used to calculate the pest and disease risk index for each region and time period.

[0023] The risk index is calculated as follows: a deviation penalty score is assigned to the degree of temperature deviation from the optimum temperature; a humidity weighted score is assigned to the degree of humidity meeting high humidity conditions; and a persistence weighted score is assigned to the number of consecutive rainfall days. The comprehensive risk index is obtained by weighted summing of the three scores. The higher the index, the greater the risk of pest and disease outbreaks. During the calculation process, each meteorological parameter needs to be standardized preprocessed: temperature deviation is normalized by dividing the absolute value of the difference between the actual temperature and the optimum temperature by the temperature tolerance range, with a value range of [0, 1]; humidity satisfaction is normalized by comparing the actual humidity value with the high humidity threshold, dividing the portion exceeding the threshold by the humidity range, with a value range of [0, 1]; and rainfall persistence is normalized by dividing the number of consecutive rainfall days by the maximum number of observed days, with a value range of [0, 1]. The comprehensive risk index is calculated by weighted summing of the three standardized scores. The weight coefficients are determined based on the importance of each meteorological factor to the occurrence of pests and diseases, and satisfy the constraint that the weight sum is 1. Based on the pest and disease risk index and the correspondence between pest and disease types and pesticide varieties, the demand surge mapping method is used to convert the high-risk index into the potential demand surge coefficient of the corresponding pesticide variety, thus obtaining the risk-driven demand adjustment vector.

[0024] Based on weekly pest and disease monitoring reports and early warning information released by plant protection departments in various provinces, text information extraction and structured parsing technologies were used to extract key information such as pest and disease type, occurrence area, affected area, and severity level, resulting in a structured data table for pest and disease monitoring. Based on this data table, a severity grading and quantification method was used to quantify four levels—mild occurrence, moderate occurrence, severe occurrence, and major occurrence—as numerical values ​​1, 2, 3, and 4, respectively, yielding quantified severity values. This encoding method uses ordinal encoding, converting categorical data into numerical data according to the natural order of severity, maintaining the relative relationship between levels, and facilitating subsequent numerical calculations. Based on the quantified severity values ​​and affected area, an impact range assessment method was used to calculate the impact scale index of pest and disease events. The impact scale index is equal to the product of the quantified severity value and the affected area. Based on the impact scale index, a threshold discrimination method was used to identify major pest and disease events with an impact scale exceeding a preset threshold, marking them as sudden demand-triggered events. Based on the correspondence between sudden demand triggering events and pest / disease types and pesticide varieties, an emergency demand estimation method is adopted. Based on parameters such as affected area, recommended application rate, and control coverage rate, the sudden demand is calculated to obtain the sudden demand vector.

[0025] Based on trend and periodic feature vectors, periodic correlation matrices, regional demand time window distribution maps, risk-driven demand adjustment vectors, and sudden demand vectors, a feature standardization method is employed to normalize various features, mapping features of different dimensions to the same numerical range, resulting in a standardized feature vector set. The standardization process uses the Min-Max normalization method, scaling each feature value to the zero-to-one interval. The specific calculation steps are: determining the maximum and minimum values ​​of the feature across all samples, subtracting the minimum value from the original feature value to obtain the difference, and then dividing this difference by the difference between the maximum and minimum values ​​to obtain the standardized feature value. For time-series features, normalization is performed based on the global maximum and minimum values ​​of the time series; for spatial features, normalization is performed based on the maximum and minimum values ​​within the regional range; and for scoring features such as risk indices and sudden demand, normalization is performed based on their defined domain range.

[0026] During feature construction, the raw data undergoes quality checks. Based on the original multi-source data, a missing value detection algorithm is used to identify null fields in the data records. For missing values ​​in time-series data, a linear interpolation method is used to estimate and impute them based on the values ​​before and after the missing point. For missing values ​​in spatial data, a neighborhood mean method is used to calculate the average value of the values ​​in neighboring regions. Based on the imputed data, an outlier detection algorithm is used to identify abnormal data that deviates from the normal range. The outlier discrimination criteria are as follows: for time-series data, if a value exceeds the range of the series mean plus or minus three standard deviations, it is determined to be an outlier; for spatial data, if the difference between a value and the values ​​in neighboring regions exceeds three times the standard deviation of the neighboring regions, it is determined to be an outlier. For the identified outliers, a median substitution method is used for correction. Based on the corrected data, a data quality scoring mechanism is used to assign a quality score to each data source based on three dimensions: completeness, timeliness, and accuracy. The quality score ranges from zero to one, with higher scores indicating better data quality. This quality score will be used as a weighting coefficient in subsequent feature fusion.

[0027] This step outputs a standardized feature vector set, including time-series feature vectors, periodic feature vectors, periodic correlation matrices, regional demand time window distribution, risk-driven demand adjustment vectors, sudden demand vectors, and data quality scores corresponding to each feature.

[0028] Step 2: Based on the standardized feature vector set, a multi-level feature fusion network model is used to predict demand and obtain a fine-grained demand prediction matrix. Based on the standardized feature vector set and data quality score output in step 1, a multi-level feature fusion network model is used to perform deep fusion and nonlinear mapping on features from different sources and dimensions to obtain the demand forecast value and prediction confidence of each pesticide variety in each future time window.

[0029] Based on the time-series and periodic feature vectors output from step 1, a time-series analysis model is used for the first layer of feature fusion, primarily addressing the time dimension features of demand. The time-series feature vectors are input into a differential autoregressive moving average model. This model captures the short-term autocorrelation of demand through an autoregressive term, captures the impact of random disturbances through a moving average term, and eliminates non-stationary trends through differencing to obtain a detrended predictive component. The periodic feature vectors are input into a periodic superposition model. This model, based on the identified principal period parameters, uses a trigonometric function fitting method to construct a periodic fluctuation function, obtaining the periodic predictive component.

[0030] Based on the detrended forecast component and the periodic forecast component, an additive combination method is used to add the two components together to obtain the basic demand forecast value. The specific calculation steps are as follows: add the values ​​of the detrended forecast component of the t-th time window to the periodic forecast component of the t-th time window to obtain the basic demand forecast value of the t-th time window.

[0031] In this calculation process, the base demand forecast value represents the base demand forecast value for the t-th time window; the detrended forecast component represents the detrended forecast component for the t-th time window; and the cyclical forecast component represents the cyclical forecast component for the t-th time window.

[0032] Based on the periodic correlation matrix, regional demand time window distribution, and risk-driven demand adjustment vector output from step 1, a regional differentiated adjustment model is used for the second layer of feature fusion, primarily addressing the spatial and risk dimensions of demand. The periodic correlation matrix and the regional demand time window distribution are multiplied to obtain the baseline demand distribution matrix for each region and time window. This matrix reflects the demand distribution of each region under normal meteorological conditions.

[0033] The risk-driven demand adjustment vector is used as an adjustment coefficient to adjust the baseline demand distribution matrix. The adjustment method is as follows: for high-risk spatiotemporal regions, the baseline demand value for that region during that time period is multiplied by a risk adjustment coefficient, which is calculated based on the risk index. Based on the adjusted regional demand distribution matrix, a cross-regional aggregation method is used to sum the demand values ​​of each region according to the pesticide variety dimension, obtaining the adjusted demand value considering regional differences and meteorological risks. The specific calculation steps are as follows: for each region r and time window t, the product of the baseline demand value for that region during that time window and the risk adjustment factor is calculated. The risk adjustment factor is obtained by adding the product of a factor and the risk sensitivity coefficient of the risk impact on demand, and the risk index of that region during that time window. The adjusted demand values ​​of all regions during the t-th time window are summed to obtain the adjusted demand value for the t-th time window.

[0034] In this calculation process, the demand adjustment value represents the demand adjustment value for the t-th time window; the total number of service areas represents the number of areas covered; the baseline demand value represents the baseline demand value for the r-th region in the t-th time window; the risk index represents the risk index for the r-th region in the t-th time window; and the impact sensitivity coefficient represents the sensitivity coefficient of the impact of risk on demand.

[0035] Based on the basic demand forecast, the adjusted demand, and the sudden demand vector output from step 1, an integrated fusion strategy is employed for the third layer of fusion, primarily addressing the impact of sudden events on demand. First, the basic demand forecast and the adjusted demand are weighted and averaged, with the weights dynamically determined based on the historical forecast errors of the two forecasting methods. Methods with smaller historical forecast errors are assigned higher weights. The weighted average yields the conventional demand forecast.

[0036] Secondly, the vector of sudden demand is superimposed on the forecast of regular demand. The temporal distribution of sudden demand adopts a decay-diffusion model: demand is highest at the moment of the sudden event, and then decreases exponentially week by week, with the decay period depending on the pest and disease control cycle. The decay-diffusion model ensures that sudden demand is not concentrated in a single time window, but is dispersed over several consecutive weeks, which is consistent with the continuous nature of the actual pest and disease control process.

[0037] The calculation steps for the final demand forecast are as follows: Calculate the basic forecast by multiplying the basic demand forecast by its corresponding fusion weight; calculate the adjusted forecast by multiplying the adjusted demand by its corresponding fusion weight; then calculate the sudden demand component by multiplying the peak sudden demand triggered by each sudden event by the exponential decay function value, which is obtained by raising the product of the negative decay coefficient of the natural exponential function to the power of the time difference, where the time difference is the difference between the current time window and the moment the sudden event occurs; sum the contributions of all sudden events; finally, add the basic forecast, the adjusted forecast, and the sudden demand component to obtain the final demand forecast for the t-th time window.

[0038] In this calculation process, the final demand forecast value represents the final demand forecast value for the t-th time window; the base forecast weight and the adjusted forecast weight represent the combined weights of the base forecast and the adjusted forecast, respectively, and their sum equals one; the number of emergencies represents the total number of emergencies considered; the peak emergencies demand represents the peak emergencies demand triggered by the e-th emergencies; the occurrence time of the emergencies represents the occurrence time of the e-th emergencies; the decay coefficient represents the parameter controlling the decay rate; the exponential decay function is used to simulate the decay process of emergencies demand over time. The natural exponential function decreases exponentially with the increase of the time difference when the time window is larger than the occurrence time of the emergencies. The decay rate is controlled by the decay coefficient. The larger the decay coefficient value, the faster the decay. This function ensures that the emergencies demand reaches its peak at the occurrence time of the event and then gradually decays to zero.

[0039] Based on the prediction results of different fusion layers, a prediction consistency analysis method is used to evaluate the prediction confidence. The standard deviation of the prediction results for each fusion layer is calculated; the smaller the standard deviation, the more consistent the results of the prediction methods, and the higher the prediction confidence. The calculation steps for prediction confidence are as follows: calculate the standard deviation of the prediction results of each fusion layer in the t-th time window, divide this standard deviation by the sum of the final demand prediction value for the t-th time window and a small constant to prevent the denominator from being zero, and finally subtract the quotient from one to obtain the prediction confidence for the t-th time window.

[0040] In this calculation process, the prediction confidence represents the prediction confidence of the t-th time window; the prediction result standard deviation represents the standard deviation of the prediction results of each fusion layer in the t-th time window; the final demand prediction value represents the final demand prediction value in the t-th time window; and the small constant represents the small constant to prevent the denominator from being zero.

[0041] Based on the forecast confidence level, an uncertainty quantification method is used to calculate the forecast interval. The forecast interval reflects the possible fluctuation range of demand and is used for subsequent risk decision-making. The calculation steps for the upper bound of the forecast interval are as follows: Calculate the confidence adjustment factor by subtracting the forecast confidence level from one, multiplying the result by the confidence adjustment coefficient, and then adding one; calculate the volatility by multiplying the standard normal quantile corresponding to the confidence level by the standard deviation of the forecast result, and then multiplying it by the confidence adjustment factor; finally, add the final demand forecast value to the volatility to obtain the upper bound of the forecast interval.

[0042] In this calculation process, the upper bound of the prediction interval represents the upper bound of the prediction interval for the t-th time window; the final demand prediction value represents the final demand prediction value for the t-th time window; the standard normal quantile represents the standard normal quantile corresponding to the confidence level; the standard deviation of the prediction result represents the standard deviation of the prediction results of each fusion layer for the t-th time window; the confidence adjustment coefficient represents the confidence adjustment coefficient; and the prediction confidence represents the prediction confidence for the t-th time window.

[0043] The steps for calculating the lower bound of the forecast interval are as follows: Calculate the confidence adjustment factor by subtracting the forecast confidence level from one, multiplying the result by the confidence adjustment coefficient, and then adding one; calculate the volatility by multiplying the standard normal quantile corresponding to the confidence level by the standard deviation of the forecast result and then multiplying it by the confidence adjustment factor; finally, subtract the volatility from the final demand forecast value to obtain the lower bound of the forecast interval.

[0044] In this calculation process, the lower bound of the prediction interval represents the lower bound of the prediction interval for the t-th time window; the final demand prediction value represents the final demand prediction value for the t-th time window; the standard normal quantile represents the standard normal quantile corresponding to the confidence level; the standard deviation of the prediction result represents the standard deviation of the prediction results of each fusion layer for the t-th time window; the confidence adjustment coefficient represents the confidence adjustment coefficient; and the prediction confidence represents the prediction confidence for the t-th time window.

[0045] Based on the final demand forecast and prediction confidence level, demand forecasts and confidence levels are calculated for each pesticide variety, each target region, and the next 8-week time window, and organized into a fine-grained prediction matrix. The rows of this matrix correspond to pesticide varieties, the columns correspond to time windows, and the matrix elements contain four attributes: demand forecast, prediction confidence level, and upper and lower bounds of the prediction interval.

[0046] Establish an online model update mechanism. Model parameters are updated weekly based on the latest actual demand data. A sliding window strategy is adopted, retaining historical data from the last two years as the training set and discarding outdated data older than two years. Based on the latest training set, model parameters, including trend coefficients, cycle parameters, risk sensitivity coefficients, and decay coefficients, are re-estimated. Simultaneously, the weight coefficients of each fusion layer are dynamically adjusted based on the prediction error over the last four weeks. The weights of fusion layers with smaller prediction errors are increased, while the weights of fusion layers with larger prediction errors are decreased. The weight update uses an exponentially weighted moving average method, assigning higher weight to recent errors, enabling the model to quickly adapt to changes in demand patterns.

[0047] This step outputs a fine-grained demand forecast matrix, which includes the forecast values ​​for the demand of each pesticide variety for each week of the next 8 weeks, the forecast confidence level, the upper and lower bounds of the forecast interval, and the updated forecast model parameters.

[0048] Furthermore, to address the potential data sparsity and cold-start issues in demand forecasting, a demand forecasting method based on transfer learning from similar varieties can be employed. For new or niche varieties with limited historical order data, directly building a forecasting model based on their own historical data will result in low prediction accuracy due to insufficient samples. To address this issue, a variety similarity calculation method is used. Based on attributes such as the active ingredient, target pest, and application period of pesticides, the similarity between the new variety and existing mature varieties is calculated. Several mature varieties with the highest similarity are selected as reference varieties, and their historical demand patterns are transferred to the new variety. Specifically, the demand cycle characteristics, seasonal fluctuation characteristics, and regional distribution characteristics of the reference varieties are extracted and weighted according to similarity to construct the initial demand pattern for the new variety. After accumulating a certain amount of actual order data for the new variety, the weight of the transferred pattern is gradually reduced, while the weight of the new variety's own data is increased, achieving a smooth transition from transfer learning to autonomous learning. This method effectively solves the cold-start problem in new variety forecasting and improves the forecasting capability covering all varieties.

[0049] Step 3: Obtain the current finished product inventory data, and combine it with the fine-grained demand forecast matrix to perform multi-dimensional assessment of inventory status and calculation of demand gap, so as to obtain the inventory status and demand gap value of each pesticide variety. Based on the fine-grained demand forecast matrix output in step 2, and combined with the current finished goods inventory data, a multi-dimensional inventory assessment method and a dynamic safety stock calculation method are used to obtain the effective inventory, demand gap value, and safety stock level of each pesticide variety.

[0050] Based on the enterprise warehouse management system, a real-time database query method is used to obtain the current inventory data of each pesticide variety, including inventory batch number, production date, warehousing date, inventory quantity, storage location, etc., to obtain a finished product inventory detail table. Based on the production date information in the finished product inventory detail table and the shelf life standard of each pesticide variety, the remaining shelf life calculation method is used to calculate the remaining shelf life of each batch. The remaining shelf life is equal to the total shelf life minus the production time.

[0051] Based on the remaining shelf life, an inventory validity assessment method was used to classify the validity of each batch of inventory. The criteria were as follows: inventory with a remaining shelf life greater than 6 months was classified as normal inventory and could be used for order fulfillment; inventory with a remaining shelf life between 3 and 6 months was classified as near-expiration inventory and should be prioritized for order allocation to accelerate turnover; inventory with a remaining shelf life less than 3 months was classified as slow-moving inventory, difficult to sell within its shelf life, and required promotional or disposal measures. Based on the validity classification results, the normal inventory, near-expiration inventory, and slow-moving inventory were calculated separately for each pesticide variety. The effective inventory was calculated as the sum of the normal inventory and the near-expiration inventory; slow-moving inventory was not included in the effective inventory.

[0052] Based on the finished goods inventory details, an inventory batch structure analysis method is used to assess the time distribution characteristics of the inventory. The weighted average inventory age is calculated, which is equal to the sum of the products of each batch's inventory quantity and its age, divided by the total inventory quantity. The inventory age is the current date minus the date of receipt. The weighted average inventory age reflects the freshness of the inventory; a longer inventory age indicates a longer period of inventory accumulation.

[0053] Based on the weighted average inventory age and historical inventory turnover rate, a turnover risk identification method is used to assess the inventory turnover status of each product. Inventory turnover rate equals the total outbound volume of the past quarter divided by the quarterly average inventory level. If the weighted average inventory age exceeds 1.5 times the historical average inventory age, and the inventory turnover rate is less than 0.7 times the historical average turnover rate, the product is considered to have turnover risk and is marked as a slow-turnover product. Slow-turnover products need to be de-prioritized in subsequent production plans to avoid further exacerbating inventory backlog.

[0054] Based on the proportion of near-expiry inventory, a method for quantifying expiry risk is used to calculate the expiry risk value. The expiry risk value equals the near-expiry inventory multiplied by the expiry probability, and then multiplied by the unit product cost. The expiry probability is estimated based on the remaining shelf life and historical sales speed; the shorter the remaining shelf life and the slower the sales speed, the higher the expiry probability. The expiry risk value is used in the objective function of subsequent collaborative optimization to guide decisions and avoid expiry losses.

[0055] Based on the demand forecast and forecast confidence level output from step 2, a dynamic safety stock calculation method is adopted to determine the safety stock level for each pesticide variety according to the forecast uncertainty and service level requirements. Traditional safety stock models assume that demand follows a normal distribution and the standard deviation is known, but this method innovatively uses the forecast confidence level as a measure of uncertainty to establish a confidence-driven safety stock model.

[0056] The steps for calculating dynamic safety stock are as follows: Calculate the confidence adjustment factor by subtracting the confidence level of the demand forecast for the i-th pesticide from 1, multiplying the result by the confidence adjustment coefficient, and then adding 1; Calculate the basic safety stock by multiplying the service level coefficient by the standard deviation of the demand for the i-th pesticide and then by the square root of the replenishment cycle of the i-th pesticide; Finally, multiply the basic safety stock by the confidence adjustment factor to obtain the safety stock quantity of the i-th pesticide.

[0057] In this calculation, safety stock represents the safety stock of pesticide i; service level coefficient represents the service level coefficient, which takes the value of 1.65 when the service level is 95% and 2.33 when the service level is 99%; demand standard deviation represents the demand standard deviation of pesticide i; replenishment cycle represents the replenishment cycle of pesticide i; confidence adjustment coefficient represents the confidence adjustment coefficient; and demand forecast confidence represents the demand forecast confidence of pesticide i.

[0058] The innovation of this formula lies in the introduction of a confidence adjustment term. When the forecast confidence is high, the adjustment term approaches zero, and the safety stock is calculated using the traditional formula; when the forecast confidence is low, the adjustment term increases, and the safety stock is increased accordingly to cope with greater demand uncertainty. This adaptive adjustment mechanism balances inventory costs and service levels.

[0059] Based on the demand forecasts for the next 8 weeks output in step 2, the total demand for the next 8 weeks is calculated using the cumulative demand calculation method. Based on effective inventory and safety stock levels, the demand gap calculation method is used to determine the demand gap for each pesticide variety. Based on safety stock levels, the production demand for each pesticide variety is determined using a demand gap calculation method. The calculation steps for the demand gap are as follows: calculate the cumulative demand forecast for the next 8 weeks, and add the demand forecasts for pesticide i from week 1 to week 8 one by one; add the cumulative demand forecast to the safety stock level of pesticide i to obtain the total demand; finally, subtract the effective stock level of pesticide i from the total demand to obtain the demand gap for pesticide i.

[0060] In this calculation process, the demand gap represents the demand gap for the i-th pesticide; the demand forecast represents the demand forecast for the i-th pesticide in week t; the safety stock represents the safety stock of the i-th pesticide; and the effective stock represents the effective stock of the i-th pesticide.

[0061] A demand gap greater than zero indicates insufficient inventory requiring production, while a demand gap less than zero indicates sufficient inventory requiring no production. Products with a demand gap greater than zero are added to the list of products to be produced, and the gap value serves as the basis for production planning.

[0062] Based on the capacity constraints and current inventory occupancy of the enterprise's hazardous chemical warehouse, an inventory capacity constraint check method is used to assess the availability of inventory space. The total capacity of the hazardous chemical warehouse is 1200 tons, with 800 tons currently occupied by finished goods inventory and 300 tons by raw material inventory, leaving 100 tons of available capacity. Based on the production demand calculated from the demand gap, the space occupied by the additional finished goods inventory if all production is based on the gap is estimated. If the space occupied by the additional inventory exceeds the remaining available capacity, an inventory capacity constraint exists, and batch production and priority adjustments need to be considered in the production plan.

[0063] Based on inventory capacity constraints, an inventory allocation priority strategy is adopted to determine the priority of inventory space allocation for each product. The priority considers three factors: the size of the demand gap, the urgency of demand, and the product profit margin. Products with large demand gaps and urgent demand are given priority in inventory allocation, while products with higher profit margins have higher priority under the same conditions. The inventory allocation priority will be used as a ranking basis in subsequent production planning.

[0064] This step outputs the inventory status and demand gap value for each pesticide variety. The inventory status includes the effective inventory, near-expiry inventory, slow-moving inventory, dynamic safety stock level, turnover risk indicator, expiry risk value, and inventory allocation priority for each pesticide variety.

[0065] Furthermore, to address the potential inventory expiration issue caused by the mismatch between pesticide product shelf-life constraints and seasonal demand, an inventory early warning method based on matching remaining shelf-life with demand windows can be employed. For each batch of inventory, the number of demand peaks it can experience within its remaining shelf-life is calculated. Future demand peak windows are identified based on demand forecasts, determining whether the batch of inventory can be sold within at least one demand peak window. If the remaining shelf-life of a batch of inventory cannot cover the next demand peak window, an inventory expiration warning is triggered. The warning information is pushed to the sales and production planning departments. The sales department can implement promotional measures to accelerate the sale of this batch of inventory, while the production planning department can reduce the recent production volume of this product to avoid the overlap of newly produced products with near-expiration inventory, which would further exacerbate the expiration risk. This method dynamically matches shelf-life constraints with demand time windows, achieving inventory risk management oriented towards seasonal demand characteristics.

[0066] Step 4: Calculate the production priority score based on inventory status and demand gap value, determine the production batch, and use the reverse scheduling method to arrange the production sequence to obtain a preliminary production plan; Based on the demand gap value and inventory allocation priority output in step 3, combined with production line capacity information and production process parameters, a preliminary production plan is obtained by using a heuristic priority sorting method and capacity allocation algorithm, which includes the production batches, production quantities, planned production times, and production line allocations for each pesticide variety.

[0067] Based on the demand gap value output in step 3, a candidate set of production tasks is established for products with a demand gap greater than zero. A multi-factor comprehensive scoring method is used to calculate the production priority score for each product, based on the demand gap value, inventory allocation priority, forecast confidence level, and historical order fulfillment status.

[0068] The steps for calculating the production priority score are as follows: Calculate the demand gap normalization term by dividing the demand gap of pesticide i by the economic production quantity of pesticide i, and then multiplying by the demand gap weighting coefficient; calculate the time urgency term by dividing the demand gap by the time window urgency of pesticide i, and then multiplying by the time urgency weighting coefficient; next, calculate the switching cost term by subtracting the normalized switching cost between pesticide i and the current production variety from the normalized value of the normalized switching cost, and then multiplying by the switching cost weighting coefficient; then calculate the profit margin term by multiplying the normalized profit margin of pesticide i by the profit margin weighting coefficient; finally, add the four terms together to obtain the production priority score for pesticide i.

[0069] In this calculation process, the production priority score represents the production priority score of the i-th pesticide; the demand gap represents the demand gap of the i-th pesticide; the economic production batch size represents the economic production batch size of the i-th pesticide; the demand time window urgency represents the demand time window urgency of the i-th pesticide; the normalized switching cost value represents the normalized switching cost value between the i-th pesticide and the current production variety; the normalized product profit margin value represents the normalized product profit margin value of the i-th pesticide; the demand gap weight coefficient, time urgency weight coefficient, switching cost weight coefficient, and profit margin weight coefficient represent the weight coefficients of the demand gap, time urgency, switching cost, and profit margin, respectively. During the calculation process, data preprocessing is required for each parameter: the normalized value of switching cost is obtained by dividing the original switching cost by the maximum switching cost, with a value ranging from zero to one; the normalized value of product profit margin is obtained by subtracting the minimum profit margin from the original profit margin and then dividing by the profit margin range, with a value ranging from zero to one; the weight coefficients must meet the normalization constraint condition, that is, the sum of the demand gap weight coefficient, time urgency weight coefficient, switching cost weight coefficient, and profit margin weight coefficient is equal to one, to ensure the reasonable allocation of the weights of each factor.

[0070] The demand gap normalization item reflects the quantity demand for production; the larger the gap, the higher the priority. The time urgency item reflects the time pressure on production; the closer the demand window, the higher the priority. The switching cost item reflects the economics of production; the lower the switching cost with the current product, the higher the priority. The profit margin item reflects the economic efficiency of production; the higher the profit margin, the higher the priority. The weighted average of these four items yields the production priority score.

[0071] Based on production priority scores, production tasks are sorted in descending order of their scores to obtain a production task priority sequence. This sequence serves as the basis for subsequent capacity allocation.

[0072] Based on the demand gap and production cost parameters of each product in the production task priority sequence, the optimal production batch size for each product is determined using the economic production batch size (EPB) model. The EPB is the batch size that minimizes the sum of production setup costs and inventory holding costs.

[0073] The steps for calculating the economic production batch size are as follows: Calculate the numerator by multiplying two by the demand gap of the i-th pesticide and then multiplying by the production preparation cost of the i-th pesticide; calculate the denominator by multiplying the unit inventory holding cost of the i-th pesticide by one and subtracting the ratio of demand gap to production capacity, where the ratio of demand gap to production capacity is equal to the demand gap of the i-th pesticide divided by the product of the production rate of the i-th pesticide and the planning period of the i-th pesticide; finally, divide the numerator by the denominator and take the square root to obtain the economic production batch size of the i-th pesticide.

[0074] In this calculation process, the economic production batch size represents the economic production batch size of the i-th pesticide; the demand gap represents the demand gap for the i-th pesticide; the production setup cost represents the production setup cost for the i-th pesticide; the unit inventory holding cost represents the unit inventory holding cost for the i-th pesticide; the production rate represents the production rate of the i-th pesticide; and the planning period represents the planning period for the i-th pesticide.

[0075] However, the economic order quantity (EOQ) may not match the actual demand gap. Based on the EOQ and the demand gap, a batch adjustment strategy is adopted to determine the actual production batch size. The adjustment rules are as follows: if the demand gap is less than 80% of the EOQ, the actual batch size equals the demand gap to avoid overproduction and inventory backlog; if the demand gap is between 80% and 120% of the EOQ, the actual batch size equals the EOQ to enjoy economies of scale; if the demand gap is greater than 120% of the EOQ, the demand gap is split into multiple batches, each batch having a batch size equal to the EOQ, with the last batch representing the remaining quantity.

[0076] Based on the actual production batch size and the single-batch capacity limit of the production line, a batch division method is adopted to determine the number of production batches for each product and the production volume of each batch. If the actual production batch size exceeds the single-batch capacity limit, it needs to be split into multiple batches for sequential production.

[0077] Based on the demand time window distribution of each product in the demand forecast matrix output in step 2, a reverse calculation method is used to determine the latest start time of each production task. The latest start time equals the demand window start time minus the production cycle time, and then minus the quality inspection and warehousing time. The production cycle time is determined according to the production process of each product, including the time for raw material feeding, chemical reaction, crystallization separation, drying and packaging, etc. The quality inspection time is 1 day, and the warehousing time is 1 day.

[0078] Based on the latest start time of each production task, a reverse scheduling method is used to arrange production periods backward from the latest start time. For multiple products with similar demand windows, they are arranged in order of priority, with higher priority products occupying production periods first.

[0079] In the scheduling, production line changeover constraints are considered. The production of different pesticide varieties requires cleaning the production line and adjusting parameters, with a changeover time of approximately two days. To reduce the number of changeovers and the changeover time, a variety clustering method is used, grouping varieties with similar raw material formulations and production processes together, and scheduling similar varieties for production in consecutive time slots as much as possible. Variety similarity is calculated based on the overlap of main raw materials and the proximity of process parameters such as production temperature and pressure.

[0080] Based on the company's three production lines and their respective capacity information, a capacity allocation algorithm is used to distribute production tasks to each production line. Production line A is mainly used for pesticide production, with a monthly capacity of 220 tons; production line B is mainly used for fungicide and herbicide production, with a monthly capacity of 200 tons; production line C is a general-purpose production line that can produce various pesticides, with a monthly capacity of 180 tons.

[0081] A production line adaptability assessment method is used to calculate the adaptability between each production task and each production line. Adaptability comprehensively considers three factors: the production line's historical production varieties, the matching degree of process parameters, and the current capacity load. A production line with a history of producing a certain variety has a high adaptability; a high matching degree of process parameters results in a short changeover and commissioning time; and a low current capacity load provides more available time, further contributing to a high adaptability.

[0082] Based on the suitability assessment results, a greedy allocation strategy is adopted to select the production line with the highest suitability and available capacity for each production task in sequence. The capacity utilization of each production line is updated in real time during the allocation process. If a production task cannot be completed within the available capacity of a single production line, it is split into multiple sub-tasks and assigned to different production lines.

[0083] To prevent one production line from being overloaded while others are idle, a load balancing check method is used to calculate the capacity utilization rate of each production line. Capacity utilization rate equals the time allotted for assigned tasks divided by the total available time. If the standard deviation of the capacity utilization rate of each production line exceeds a threshold, a load balancing adjustment is triggered, transferring some tasks from the high-load production line to the low-load production line to ensure balanced capacity utilization.

[0084] This step outputs a preliminary production plan, including the production batch quantity for each pesticide variety, the production volume of each batch, the planned start time, the planned completion time, and the assigned production line number.

[0085] Step 5: Based on the preliminary production plan, calculate the demand for various raw materials using the bill of materials expansion method, assess the raw material supply capacity using the time matching analysis method, and generate a raw material procurement plan. Based on the preliminary production plan output in step 4, combined with pesticide formulation data and raw material inventory data, the formula expansion calculation method and supply time matching analysis method are used to obtain the demand for various raw materials, raw material gaps, raw material procurement plans, and production feasibility assessment results.

[0086] Based on the production batches and quantities of each pesticide variety in the preliminary production plan output in step 4, obtain the raw material formula tables for each pesticide variety from the enterprise's product formula database. The formula tables record information such as the types of raw materials required for the production of each variety, their proportions, and the order of addition. For example, the formula for imidacloprid insecticide includes imidacloprid technical, adjuvants, solvents, and fillers, with a proportion of 10% imidacloprid technical, 5% adjuvants, 20% solvent, and 65% fillers.

[0087] Based on the production plan and formulation table, the total demand for each type of raw material is calculated using the bill of materials (BOM) expansion method. For each pesticide variety, its planned production quantity is multiplied by the proportion coefficient of each raw material in the formulation to obtain the demand for each raw material for that variety. The raw material demands for all varieties are summed by raw material type to obtain the total demand for each type of raw material. Based on the raw material requirements obtained in step 1, the total raw material requirements are determined using a raw material requirement aggregation calculation method. The calculation steps are as follows: For the j-th raw material, iterate through all pesticide varieties planned for production, multiply the planned production quantity of each pesticide by the proportion coefficient of the j-th raw material in the pesticide formula, and sum the requirements of all pesticide varieties for the j-th raw material to obtain the total requirement of the j-th raw material.

[0088] In this calculation process, the total demand represents the total demand for the j-th raw material; the planned number of pesticide varieties to be produced represents the total number of different pesticide varieties to be produced; the planned production volume represents the planned production volume of the i-th pesticide; and the ratio coefficient represents the ratio coefficient of the j-th raw material in the i-th pesticide formulation.

[0089] Considering raw material losses and the removal of substandard products during the production process, a loss rate adjustment method is adopted to adjust the raw material demand upwards. The adjusted raw material demand equals the theoretical demand multiplied by the loss adjustment coefficient. The loss adjustment coefficient is calculated based on historical production records and is generally between 1.03 and 1.05, representing an increase of 3% to 5% in loss margin.

[0090] Based on the enterprise raw material warehousing management system, a database query method is used to obtain current raw material inventory data, including the inventory quantity, batch number, warehousing time, and supplier information of various raw materials. Based on the raw material inventory data, the current available inventory quantity of various raw materials is calculated.

[0091] Based on the demand for various raw materials and the current available inventory, a gap calculation method is used to determine the gap for each type of raw material. The calculation steps are as follows: subtract the current inventory of raw material j from the demand for raw material j to obtain the gap for raw material j.

[0092] In this calculation, the shortage quantity represents the shortage quantity of the j-th raw material; the demand quantity represents the demand quantity of the j-th raw material; and the current inventory quantity represents the current inventory quantity of the j-th raw material.

[0093] A shortage of raw materials greater than zero indicates a need for procurement, while a shortage of raw materials less than or equal to zero indicates a sufficient supply and no need for procurement. Raw materials with a shortage greater than zero should be added to the list of raw materials to be procured.

[0094] Simultaneously, query the in-transit purchase order data to obtain information on raw material purchase orders that have been placed but not yet received, including raw material type, purchase quantity, and estimated arrival time. Based on the in-transit purchase orders, calculate the in-transit supply and estimated arrival time of various raw materials. Include the in-transit supply in the calculation of available raw materials and update the raw material gap. If the in-transit supply can cover the gap, no new purchase is needed; if the in-transit supply is insufficient to cover the gap, new purchase is required, and the new purchase quantity equals the gap minus the in-transit supply.

[0095] Based on the raw materials in the list of materials to be procured, obtain supplier information from the supplier management system, including supplier name, type of raw material supplied, regular delivery cycle, shortest delivery cycle, and supply capacity. The regular delivery cycle is the time from order placement to delivery under normal circumstances, while the shortest delivery cycle is the fastest possible delivery time under expedited circumstances.

[0096] Based on the supply cycles of various raw materials and the planned start times of each product in the production plan, a time-matching analysis method is used to assess the feasibility of raw material supply. For each raw material to be procured, its latest order time is calculated. The latest order time is equal to the planned start time of the earliest production task using that raw material minus the supply cycle and then minus the raw material warehousing and inspection time. If the latest order time has already passed or is very shortly after the current moment, it indicates that the procurement time for that raw material is tight and expedited measures are required.

[0097] Based on the time matching analysis results, the feasibility of the production plan is assessed. The judgment rule is as follows: if the regular supply cycle of a certain raw material exceeds the time difference between the latest order time and the current time, but the shortest supply cycle can meet the demand, then the raw material has a supply risk but is feasible, and it is necessary to negotiate with the supplier for expedited supply; if the shortest supply cycle of a certain raw material still exceeds the time difference, then the raw material cannot be supplied on time, the production task relying on the raw material is not feasible, and the preliminary production plan needs to be adjusted.

[0098] For production tasks that are not feasible, a plan adjustment strategy is adopted. Strategy one is to postpone the planned start time of the task and wait for the raw materials to arrive before starting production, but it is necessary to check whether the postponement will affect the order delivery commitment; strategy two is to reduce the production batch size of the task, prioritize the production of some batches, and postpone the remaining batches to partially meet the demand; strategy three is to adjust the priority of production tasks, downgrade the task, and prioritize the production of other tasks with sufficient raw materials.

[0099] Based on the list of raw materials to be procured and the new procurement quantities of various raw materials, and based on the supply cycle analysis results, a procurement batch optimization method is adopted to determine the procurement batch and timing of various raw materials.

[0100] For raw materials with ample procurement time, the Economic Order Quantity (EOQ) model is used to determine the optimal procurement quantity by comprehensively considering ordering costs and inventory costs. The calculation steps for the EOQ are as follows: Calculate the numerator by multiplying by the annual demand for the j-th raw material and then by the ordering cost of the j-th raw material; divide the numerator by the unit inventory holding cost of the j-th raw material; finally, take the square root of the quotient to obtain the EOQ for the j-th raw material.

[0101] In this calculation, the economic order quantity (EOQ) represents the economic order quantity for raw material j; the annual demand represents the annual demand for raw material j; the ordering cost represents the ordering cost for raw material j; and the unit inventory holding cost represents the unit inventory holding cost for raw material j.

[0102] However, the economic order quantity (EOQ) may not match the current procurement demand. A batch adjustment strategy can be adopted: if the current procurement demand is less than half the EOQ, then the procurement batch should be equal to the EOQ, allowing for advance stockpiling of raw materials to benefit from economies of scale; if the current procurement demand is close to the EOQ, then the procurement batch should be equal to the EOQ; if the current procurement demand is greater than the EOQ, then the procurement batch should be equal to the current demand, avoiding excessive capital and inventory space occupation.

[0103] For raw materials with tight procurement timelines, an expedited procurement strategy is adopted, negotiating with suppliers to shorten delivery cycles. Expedited procurement typically incurs additional expedited fees. A cost comparison is made between the expedited fees and the production delay losses due to raw material shortages. If the expedited fees are lower than the delay losses, expedited procurement is chosen; otherwise, the production plan is postponed.

[0104] Based on the determined purchase volume and timing, a raw material purchase plan is generated, which includes information such as the purchase quantity of various raw materials, target suppliers, order placement time, estimated delivery time, and whether expedited processing is required.

[0105] This step outputs the raw material procurement plan, including the demand, inventory, shortage, and in-transit supply of various raw materials, a procurement plan table, production feasibility assessment results, and a list of production tasks that need to be adjusted.

[0106] Furthermore, to address the issue of frequent production plan adjustments due to unstable supply of certain key raw materials, a safety stock strategy based on raw material supply risk assessment can be adopted. For each raw material, a raw material supply risk index is calculated based on factors such as the supplier's historical on-time delivery rate, the number of suppliers, the procurement cycle length, and the strategic importance of the raw material. A higher risk index indicates greater uncertainty in the supply of that raw material. Based on the supply risk index, the safety stock level of the raw materials is dynamically adjusted. Higher safety stock is set for high-risk raw materials to buffer supply fluctuations; lower safety stock is set for low-risk raw materials to reduce inventory costs. The safety stock level is determined comprehensively based on the risk index and demand volatility. The calculation method is to multiply the demand standard deviation term in the traditional safety stock formula by a risk adjustment coefficient; the higher the risk index, the larger the adjustment coefficient. This strategy achieves a balance between raw material supply uncertainty and inventory costs, improving the stability and executability of production plans.

[0107] Step 6: Based on the preliminary production plan and raw material procurement plan, a hierarchical optimization strategy is adopted to make collaborative decisions and obtain a collaboratively optimized production plan; Based on the production feasibility assessment results and raw material procurement plan output in step 5, combined with the preliminary production plan in step 4 and the inventory status in step 3, a two-way coupled optimization model and a hierarchical solution algorithm are used to obtain the collaboratively optimized production plan, inventory replenishment strategy, and raw material procurement plan.

[0108] Based on the production task list requiring adjustment output from step 5, the preliminary production plan from step 4 is adjusted. For production tasks that are not feasible due to insufficient raw material supply, the planned start time of the task is postponed according to the estimated arrival time of the raw materials. The postponed start time is equal to the raw material arrival time plus the raw material inspection and warehousing time.

[0109] Based on the adjusted production schedule, re-examine whether it will affect order delivery. If the delayed estimated completion time is still earlier than the order's required delivery date, the adjustment is feasible; if it is later than the delivery date, there is a risk to order delivery, and further optimization is needed. Optimization methods include: first, further negotiating with suppliers to shorten the raw material supply cycle; second, arranging for production lines to work overtime to shorten the production cycle; third, allocating inventory from other factories or partners; and fourth, negotiating with customers to postpone delivery.

[0110] Based on the order delivery risk and the costs of various optimization methods, a cost-benefit analysis is used to select the optimal solution. If the cost of all solutions is too high or cannot be implemented, the order is marked as a delivery risk order and submitted to management for decision-making.

[0111] Based on the adjusted feasible production plan, a hierarchical optimization strategy is adopted for the first level of optimization: production batch and timing optimization. The optimization objective is to minimize production changeover costs and in-transit inventory holding costs while meeting demand and delivery deadlines.

[0112] Production changeover costs include production line cleaning costs, equipment debugging costs, and capacity loss costs during the changeover period. Changeover costs vary between different product varieties; varieties with significant differences in raw material formulations and process parameters have higher changeover costs. In-transit inventory holding costs refer to the costs incurred in holding inventory after production is completed but before order delivery. Premature production leads to longer in-transit inventory periods and higher holding costs.

[0113] Dynamic programming is used to solve the production sequence optimization problem. The state in dynamic programming is defined by the current time and the current product type, the decision is the next product type to be produced, and the state transition cost is the switching cost plus the cost of in-transit inventory. Through recursive dynamic programming, the minimum cost path from the initial state to the final state is found, and this path corresponds to the optimal production sequence.

[0114] In the dynamic programming solution process, a production time window constraint is considered, meaning that each task must be scheduled before its latest start time; otherwise, the delivery deadline cannot be met. Constraint pruning reduces the search space and improves algorithm efficiency.

[0115] The optimized production sequence may differ from the initial plan. The production order of product varieties has been adjusted based on changeover costs and time constraints, resulting in lower overall costs.

[0116] Based on the production timing obtained from the first-level optimization, the second-level optimization is performed: inventory replenishment strategy optimization. The optimization objective is to minimize inventory holding costs and stockout costs while meeting service level requirements.

[0117] A dynamic inventory strategy model is adopted, which considers the time-varying nature of demand and the periodicity of replenishment. The model assumes that for each pesticide variety, replenishment is triggered when the inventory level falls below the reorder point, and the replenishment quantity restores the inventory to the target inventory level. The reorder point and the target inventory level are the decision variables that need to be optimized.

[0118] Determining the reorder point requires considering the demand and safety stock within the replenishment cycle. Based on the demand forecast results from step 2, the average demand and standard deviation of demand within the replenishment cycle are calculated. The steps for calculating the reorder point are as follows: multiply the average demand rate of pesticide i by the replenishment cycle of pesticide i, and add the safety stock of pesticide i to obtain the reorder point of pesticide i.

[0119] In this calculation, the reorder point represents the reorder point of the i-th pesticide; the average demand rate represents the average demand rate of the i-th pesticide; the replenishment cycle represents the replenishment cycle of the i-th pesticide; and the safety stock represents the safety stock of the i-th pesticide.

[0120] Determining the target inventory level requires balancing inventory holding costs and stockout costs. Too high a target inventory level leads to high holding costs, while too low a target inventory level leads to a high risk of stockouts. A cost-trade approach is used to find the target inventory level that minimizes total costs.

[0121] Considering the seasonality of demand, a time-varying inventory strategy is adopted, which involves increasing the target inventory level before peak demand periods and decreasing the target inventory level during trough demand periods. Based on the demand forecast matrix output in step 2, the peak and trough demand periods for each product are identified, and the target inventory level is dynamically adjusted.

[0122] Based on the production plan and inventory strategy obtained from the first and second layers of optimization, a third layer of optimization is performed: raw material procurement batch optimization. The optimization objective is to minimize raw material procurement costs and raw material inventory costs.

[0123] Raw material procurement costs include the unit price of raw materials and ordering costs. Some suppliers offer quantity discounts, with lower unit prices for larger purchase orders. Raw material inventory costs include inventory holding costs and inventory space occupancy costs.

[0124] A multi-product joint procurement optimization model is adopted, which takes into account the correlation between different raw materials. Some raw materials may come from the same supplier, and joint procurement can enjoy the benefits of transportation cost sharing and bulk discounts. Some raw materials are used simultaneously in the formula, so it is necessary to maintain a coordinated inventory ratio.

[0125] The constraints of the model include: raw material demand satisfaction constraint, the purchase quantity should meet the raw material demand of the production plan; inventory capacity constraint, the raw material inventory shall not exceed the raw material warehouse capacity; and capital budget constraint, the total purchase cost shall not exceed the purchase budget.

[0126] This optimization problem is solved using a mixed-integer programming approach. Due to the large problem size, a heuristic algorithm is employed for fast solution. The heuristic rules include: prioritizing the purchase of raw materials with high demand and quantity discounts; combining purchases of multiple raw materials from the same supplier whenever possible; and minimizing the number of purchase batches while still meeting demand.

[0127] Based on the production timing, inventory strategy, and procurement plan obtained from the first, second, and third layers of optimization, an iterative coordination method is used for integration. Since the three layers of optimization are interconnected, adjustments to the production timing in the first layer will affect the inventory demand in the second layer, and adjustments to the inventory strategy in the second layer will affect the raw material demand in the third layer. Therefore, multiple rounds of iterative coordination are required.

[0128] The iterative coordination process is as follows: In the first round, the three-layer optimization is executed in the above order to obtain a preliminary coordination plan; in the second round, the raw material procurement plan obtained from the third-layer optimization is fed back to the first layer to check whether there are raw material constraints that cause the production sequence to need to be adjusted. If adjustments are needed, the first-layer optimization is re-executed to obtain an updated production sequence; in the third round, the updated production sequence is fed back to the second and third layers to re-execute the inventory strategy optimization and procurement batch optimization.

[0129] The iteration terminates when the change in optimization results over two consecutive rounds is less than a set threshold, or when the maximum number of iterations is reached. Convergence typically occurs after 2 to 3 iterations. The converged result is the co-optimized production plan, inventory strategy, and raw material procurement plan.

[0130] Based on the results of collaborative optimization, the optimized total cost is calculated, including production costs, changeover costs, inventory holding costs, stockout costs, procurement costs, and raw material inventory costs. This cost is then compared with the initially planned total cost to evaluate the optimization effect. Typically, collaborative optimization can reduce total costs by 8% to 15%.

[0131] This step outputs the collaboratively optimized production plan (including production batches, production volume, production sequence, and production line allocation for each product), dynamic inventory strategy (including reorder points and target inventory levels for each product), optimized raw material procurement plan, total cost, and cost composition.

[0132] Step 7: Identify disturbance events based on the collaboratively optimized production plan, assess the impact of the disturbances using the impact propagation analysis method, generate response plans, and update the production plan and order delivery plan; Based on the collaboratively optimized production plan and inventory strategy output in step 6, and combined with real-time production data, including real-time order data and production progress data, a disturbance impact propagation assessment model and a multi-solution generation algorithm are used to obtain a disturbance impact assessment report, a set of response solutions, an updated production plan, and an order delivery plan.

[0133] Based on the enterprise order management system and production execution system, a real-time data acquisition interface is used to obtain the latest order change information and production progress information. Order change information includes new orders, order cancellations, order quantity adjustments, and delivery date adjustments. Production progress information includes the actual start time, current progress, and estimated completion time of each production task.

[0134] Based on order change information, a disturbance type discrimination method is used to identify the type and parameters of disturbance events. Disturbance types include: urgent order insertion, regular order cancellation, order quantity increase, order quantity decrease, delivery date advance, and delivery date delay. For urgent order insertion, parameters such as pesticide type, quantity, required delivery date, and customer importance are extracted. For order quantity adjustments, the adjustment amount and adjustment ratio are calculated.

[0135] Based on production schedule information, a schedule deviation detection method is used to identify abnormal situations such as production delays or early completion. The deviation between actual and planned progress is calculated; if the deviation exceeds a threshold (e.g., exceeding 10% of the planned time), it is identified as a production disturbance event. Parameters such as the disrupted production task, delay duration, or early completion time are extracted.

[0136] Based on the identified disturbance events, an impact propagation analysis method is used to assess the impact of the disturbances on subsequent production plans and order delivery. Impact propagation is divided into three levels: direct impact, indirect impact, and cascading impact.

[0137] Direct Impact Analysis: For urgent order insertions, the direct impact is the production capacity and time required for the order. Based on the order type and quantity, the production cycle of the product is queried to calculate the required production time. Based on the current production line load, it is determined which production lines can handle the urgent order. If all production lines are already at full capacity, time needs to be squeezed from the existing schedule.

[0138] Indirect Impact Analysis: If urgent orders require additional production time from existing plans, the affected production tasks will need to be postponed. Based on the task dependencies in the production plan, an impact propagation path tracing method is used to identify subsequent tasks affected by the postponed tasks. Task dependencies include: time-series dependencies on the same production line, resource dependencies due to raw material constraints, and delivery dependencies due to order combinations. The impact is propagated level by level along the dependency chain, and the delay duration of each affected task is calculated.

[0139] Chain reaction analysis: Delayed production tasks may postpone raw material consumption, thus affecting raw material procurement plans. Adjustments to raw material procurement plans may impact supplier supply arrangements, further affecting other production tasks using the raw material. A correlation analysis approach is used to identify the chain reaction effects arising from raw material constraints. Simultaneously, delayed production tasks may lead to insufficient finished goods inventory during specific periods, affecting inventory allocation for other orders. Inventory time-series analysis is used to identify the chain reaction effects arising from inventory constraints.

[0140] Based on three levels of impact analysis, an impact propagation path diagram is generated. This diagram takes the disturbance event as the starting point, the affected orders and tasks as nodes, and the impact relationships as edges, intuitively showing the propagation process of the disturbance.

[0141] Based on the impact propagation path diagram, a disturbance impact assessment method is used to calculate the degree of impact caused by the disturbance. Impact indicators include: the number of affected orders, the total order delivery delay time, the increase in additional costs, and the impact score on customer satisfaction.

[0142] The total delay time for order delivery is calculated as follows: For all affected orders, calculate the difference between their estimated delivery time and the original promised delivery time. The difference is the delay time. Sum the delay times of all orders to obtain the total delay time.

[0143] Additional cost increases include: production line changeover costs (if urgent orders cause adjustments to the original product sequence, increasing the number of changeovers), overtime production costs (if overtime absorbs the disruption), logistics costs for inventory transfers (if urgent orders are met through cross-regional inventory transfers), and outsourcing costs (if urgent orders are met through outsourcing). The total additional costs are calculated based on the unit cost parameters of each type of cost and the quantity changes caused by the disruption.

[0144] Customer satisfaction scores are based on a comprehensive assessment of customer importance and order delay levels. Customer importance is scored according to factors such as historical purchase volume, length of cooperation, and strategic position, categorized into A, B, and C classes. Class A customers have the highest weight, and order delays have the greatest impact on them. Order delay levels are scored based on the proportion of delay time to the original delivery cycle; the higher the delay percentage, the more points are deducted.

[0145] Based on a comprehensive score of impact indicators, a risk level assessment method is used to classify the impact of disturbances into three levels: low risk, medium risk, and high risk. The assessment rules are as follows: if the number of affected orders is less than 5 and the total delay time is less than 10 days, and there are no delayed orders from Category A customers, it is classified as low risk; if the number of affected orders is between 5 and 15, or the total delay time is between 10 and 30 days, or there are delayed orders from Category A customers but the delay time is less than 5 days, it is classified as medium risk; if the number of affected orders is greater than 15, or the total delay time is greater than 30 days, or the delay time for Category A customers is greater than 5 days, it is classified as high risk.

[0146] Based on the disturbance impact assessment results and risk level, a multi-strategy solution generation method is adopted to generate multiple response plans for the disturbance. The plan types include: production priority adjustment plan, overtime production plan, inventory allocation plan, outsourcing production plan, and delivery date negotiation plan.

[0147] Production priority adjustment plan: Prioritize urgent orders to the highest level, adjust production sequence, insert the production tasks corresponding to urgent orders into the current production plan, and postpone some existing tasks. Based on the collaborative optimization method in step 6, re-optimize the production sequence and calculate the adjusted production plan. Evaluate the delay of normal orders and the increase in switching costs caused by this plan.

[0148] Overtime Production Plan: Maintain the original production schedule and increase capacity by scheduling overtime shifts on the production line to produce urgent orders outside of normal production hours. Overtime hours include night shifts on normal working days and weekends. Calculate the required overtime hours and overtime costs based on the overtime hours and overtime pay standards. Assess whether this plan can meet the delivery time requirements of urgent orders.

[0149] Inventory Transfer Plan: If the items for urgent orders are in stock in warehouses in other regions, inventory can be transferred from those regions to quickly fulfill the orders. Query inventory data from warehouses in other regions to identify available inventory batches and quantities. Calculate cross-regional logistics transportation time and costs. Assess whether the transfer plan can meet delivery time requirements and whether the inventory in other regions is sufficient after the transfer.

[0150] Outsourcing Production Solution: If a company's own production capacity cannot meet urgent orders and timelines are tight, it can outsource production to external partners. Research the partner company's production capacity and quality standards to assess the feasibility of outsourcing. Calculate the costs, quality risks, and delivery time of outsourcing. Evaluate the overall feasibility of the outsourcing solution.

[0151] Delivery date negotiation options: If the above options are too costly or impractical, a delivery date postponement can be negotiated with the customer. Assess the likelihood of a successful negotiation based on the customer's importance and the urgency of the order. Calculate the acceptable postponement range. Evaluate the impact of the negotiated option on the customer relationship.

[0152] For each option, evaluation indicators such as total cost, order fulfillment rate, implementation difficulty, and risk level are calculated, and a comparison table of options is generated. Total cost includes direct and indirect costs. Direct costs include quantifiable costs such as overtime pay, logistics costs, and outsourcing costs. Indirect costs include costs that are difficult to quantify precisely but need to be considered, such as loss of customer satisfaction and impact on brand reputation.

[0153] A multi-objective decision-making method is employed to comprehensively evaluate each option. An evaluation matrix is ​​established, with rows corresponding to options and columns to evaluation indicators. Matrix elements represent the scores of each option on each indicator. Evaluation indicators are normalized to eliminate the influence of dimensions. Based on the indicator weights and normalized scores, the comprehensive score of each option is calculated. The option with the highest comprehensive score is the recommended option. Simultaneously, alternative options are listed for decision-makers to choose from based on the actual situation.

[0154] Based on the response plan selected by the decision-makers, an adaptive scheduling method is adopted to update the production plan and order delivery plan. The adaptive scheduling method selects appropriate scheduling strategies according to the type and scope of the disturbance: task reallocation for equipment failure disturbances, task postponement for raw material shortage disturbances, and task insertion for urgent orders. The principle of local replanning is to adjust only the periods and tasks affected by the disturbance, maintaining the stability of the unaffected parts of the plan, and reducing the cascading effects and execution difficulty of the adjustment.

[0155] For production priority adjustment plans, regenerate the production sequence for the affected periods and update the start and finish times of production tasks. For overtime production plans, add overtime production task records to the production plan. For inventory transfer plans, update inventory transfer instructions and logistics arrangements. For outsourced production plans, generate outsourced production orders and track outsourcing progress.

[0156] Based on the updated production plan, the estimated delivery time for each order is recalculated, generating a new order delivery plan. For orders with changed delivery times, a delivery date update notification is sent to the customer, explaining the reasons for the adjustment and the new delivery time.

[0157] Establish a plan execution monitoring mechanism to track the execution status of the updated plan in real time. For critical tasks and urgent orders, set progress monitoring points, collect actual progress data daily, and compare it with the planned progress. If new deviations or disturbances are detected, trigger a new round of disturbance response procedures.

[0158] This step outputs the updated production plan, the updated order delivery plan, the disturbance impact assessment report (including the impact propagation path diagram, the quantitative results of the impact indicators, and the risk level), the response plan comparison table (including the cost, benefit, and feasibility assessment of each plan), the updated production plan, and the updated order delivery plan.

[0159] Furthermore, to address the issue of frequent small-scale disturbances potentially leading to repeated plan adjustments and execution chaos, a robust scheduling method based on a disturbance absorption buffer mechanism can be adopted. This involves reserving a certain amount of time and capacity buffer in the production plan to absorb small-scale disturbances. A time buffer refers to reserving a certain amount of idle time between critical tasks; when a task is delayed, the buffer time can absorb the delay and avoid affecting subsequent tasks. A capacity buffer means not fully utilizing production line capacity, reserving a certain amount of spare capacity. When urgent orders arise, this spare capacity can be used for a rapid response without adjusting the existing plan. The buffer settings are based on historical disturbance statistical analysis, calculating the average frequency and average impact of disturbances, and determining an appropriate buffer level based on the disturbance characteristics. Too small a buffer will not effectively absorb disturbances, while too large a buffer will reduce capacity utilization; a trade-off must be struck between stability and efficiency. This method improves the robustness of the plan by proactively reserving buffers, reduces frequent adjustments caused by small-scale disturbances, and improves the stability of production operations.

[0160] A computer-readable storage medium for storing computer-readable instructions that, when read by a computer, enable the execution of a pesticide production order and inventory collaborative management method as described above.

[0161] In one embodiment of the present invention, a specific example is provided: To verify the technical effectiveness of the method of this invention, actual production order data from a pesticide manufacturer from May to July 2024 were selected for application demonstration and effect comparison analysis. During this period, the company received 152 orders involving 15 pesticide varieties, totaling approximately 1850 tons. This period coincides with the critical period for rice pest and disease control, resulting in significant demand fluctuations. The company has three production lines and uses 32 types of raw materials, with some raw material procurement cycles lasting up to 50 days.

[0162] Based on the methods in steps 1 and 2, historical order data, agricultural production cycle data, meteorological data, and pest and disease monitoring data were collected. Examples of some multi-source data collection are shown in Table 1. Table 1: Example of multi-source data acquisition (Week 1, May 2024);

[0163] Based on the data in Table 1, the demand forecast results for the next 8 weeks are obtained after feature extraction and fusion network processing.

[0164] Based on demand forecasts and inventory status, the collaborative optimization method used in steps 3 to 6 generates an optimized production plan. Table 2 shows a comparison of key indicators before and after collaborative optimization. Table 2: Comparison of key indicators before and after collaborative optimization;

[0165] Table 2 shows that through the two-way collaborative optimization of production planning and inventory strategies, the number of production changeovers was reduced, lowering changeover costs and time losses. Average inventory levels decreased, inventory turnover significantly improved, reducing capital tied up in inventory and inventory holding costs. On-time order delivery rate increased from 87% to 96%, improving customer satisfaction. Through inventory effectiveness assessment and a dynamic safety stock mechanism, expired losses were reduced by 78%, avoiding product waste caused by inventory backlog. Overall, the combined cost savings resulted in an 11.7% reduction in total operating costs, demonstrating good economic benefits.

[0166] In the third week of June 2024, a provincial plant protection department issued a warning of a large-scale outbreak of brown planthoppers, triggering a surge in emergency orders for imidacloprid insecticide. The company received eight emergency orders totaling 95 tons within three days, requiring delivery within one week. Using the disturbance response method in step 7, a disturbance impact assessment was completed within two hours, identifying the disturbance as a high-risk disturbance that would cause delays in four regular orders, with a total delay of 28 days.

[0167] The system generated five response plans: Plan 1: Adjust production priority, increasing costs by 28,000 yuan and delaying normal orders by an average of 7 days; Plan 2: Work overtime, increasing costs by 42,000 yuan, but normal orders will not be delayed; Plan 3: Transfer 40 tons of inventory from the Guangdong factory and produce 55 tons locally through overtime, increasing costs by 35,000 yuan and delaying normal orders by an average of 2 days; Plan 4: Outsource production of 30 tons and produce 65 tons locally through overtime, increasing costs by 51,000 yuan and delaying normal orders by an average of 1 day; Plan 5: Negotiate extensions with some customers, increasing costs by 8,000 yuan, but resulting in a significant loss of customer satisfaction.

[0168] After comprehensive evaluation, Option 3 was selected, successfully completing the delivery of the emergency order within one week, while the average delay for normal orders was controlled within two days, maintaining good customer satisfaction. Traditional methods rely on manual experience and localized adjustments, typically requiring one to two days to formulate a response plan. Furthermore, they struggle to comprehensively assess the impact and compare multiple options, resulting in lower response speed and decision-making quality. This invention reduces the disturbance response time from one to two days to two hours, improving response speed by over 80%. Simultaneously, through the optimization of multiple options, the response effect is superior.

[0169] Based on the above application examples and technical effect verification, the method of this invention effectively solves the key technical problems in pesticide production order and inventory management through three major technical innovations: multi-source data fusion prediction, two-way collaborative optimization of production and inventory, and adaptive response to disturbances. It has achieved certain improvements in demand forecasting accuracy, order delivery performance, inventory optimization effect, dynamic response capability, and overall operating costs, providing pesticide production enterprises with a complete, effective, and feasible collaborative management solution for orders and inventory.

[0170] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments based on the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for collaborative management of pesticide production orders and inventory, characterized in that, include: A standardized feature vector set is obtained by merging multi-source heterogeneous datasets of pesticides using data interface calling technology and extracting features. Demand prediction is performed using a multi-level feature fusion network model based on a standardized feature vector set, resulting in a fine-grained demand prediction matrix. Obtain current finished product inventory data, combine it with fine-grained demand forecasting matrix to conduct multi-dimensional assessment of inventory status and calculate demand gap, and obtain the inventory status and demand gap value of each pesticide variety. Based on inventory status and demand gap, calculate production priority score, determine production batch size, and use reverse scheduling method to arrange production sequence to obtain preliminary production plan; Based on the preliminary production plan, the bill of materials expansion method is used to calculate the demand for various raw materials, the time matching analysis method is used to assess the raw material supply capacity, and a raw material procurement plan is generated. Based on the preliminary production plan and raw material procurement plan, a hierarchical optimization strategy is adopted to make collaborative decisions and obtain a collaboratively optimized production plan. Based on the collaboratively optimized production plan, identify disturbance events, use impact propagation analysis to assess the impact of disturbances, generate response plans, and update the production plan and order delivery plan.

2. The pesticide production order and inventory collaborative management method according to claim 1, characterized in that, The feature extraction includes: We acquired agricultural production cycle datasets and pest and disease monitoring datasets, and used a growth stage mapping method to divide the agricultural production cycle datasets into stages, namely, sowing period, tillering period, jointing period, heading period, grain filling period, and maturity period. A demand correlation matrix between growth stages and pesticide varieties was established based on the stage segmentation results. Risk assessment of the pest and disease monitoring dataset was conducted using a risk quantification scoring method. The pest and disease risk index was calculated by weighted summation based on the degree of temperature deviation, the degree of humidity satisfaction, and the number of consecutive rainfall days, resulting in a standardized feature vector set.

3. The method for collaborative management of pesticide production orders and inventory according to claim 1, characterized in that, The demand forecast includes: Based on the standardized feature vector set, a time series analysis is performed on the standardized feature vector set using a differential autoregressive moving average model. The autoregressive term captures the short-term autocorrelation of demand, the moving average term captures the impact of random disturbances, and the differential operation eliminates non-stationary trends to obtain the detrended prediction component. A periodic superposition model is used to perform periodic analysis on the standardized feature vector set. Based on the identified principal period parameters, a trigonometric function fitting method is used to construct a periodic fluctuation function to obtain the periodic prediction component. The basic demand forecast value is obtained by adding the detrended forecast component and the cyclical forecast component.

4. The pesticide production order and inventory collaborative management method according to claim 3, characterized in that, The demand forecast also includes: Based on the basic demand forecast, a regional differentiation adjustment model is used to adjust the basic demand forecast regionally. The baseline demand distribution matrix is ​​obtained by performing matrix multiplication between the periodic correlation matrix and the regional demand time window distribution. The risk-driven demand adjustment vector is used as the adjustment coefficient to adjust the baseline demand distribution matrix. For high-risk spatiotemporal regions, the baseline demand value of this region for this time period is multiplied by the risk adjustment coefficient. The cross-regional aggregation method is used to sum the demand values ​​of each region according to the pesticide variety dimension to obtain the demand adjustment value after considering regional differences and meteorological risks.

5. The pesticide production order and inventory collaborative management method according to claim 1, characterized in that, The calculation of the demand gap includes: Based on a fine-grained demand forecasting matrix and forecast confidence, a dynamic safety stock calculation method is used to measure the uncertainty of forecast confidence. The safety stock level is set as the service level coefficient multiplied by the demand standard deviation multiplied by the square root of the replenishment cycle and then multiplied by the confidence adjustment factor. The confidence adjustment factor is equal to 1 plus the confidence adjustment coefficient multiplied by 1 minus the forecast confidence. When the forecast confidence is high, the safety stock is calculated according to the traditional formula; when the forecast confidence is low, the safety stock is increased accordingly. The demand gap is set as the cumulative demand forecast for the coming weeks plus safety stock minus effective inventory.

6. The method for collaborative management of pesticide production orders and inventory according to claim 1, characterized in that, The multidimensional assessment of inventory status includes: Based on current finished goods inventory data, an inventory validity assessment method is used to classify the finished goods inventory data. Based on the remaining shelf life, inventory with a longer remaining shelf life is classified as normal inventory, inventory with a moderate remaining shelf life is classified as near-expiry inventory, and inventory with a short remaining shelf life is classified as slow-moving inventory. The effective inventory level is set as the sum of normal inventory and near-expiry inventory. A turnover risk identification method is used to assess the risk of inventory data. If the weighted average inventory age exceeds the historical average inventory age and the inventory turnover rate is lower than the historical average turnover rate, then this product is considered to have turnover risk.

7. The method for collaborative management of pesticide production orders and inventory according to claim 1, characterized in that, The formulation of the preliminary production plan includes: Based on the demand gap value and inventory allocation priority, a multi-factor comprehensive scoring method is used to prioritize the demand gap value. The production priority score is calculated by weighted summation of demand gap normalization, time urgency, switching cost, and profit margin. The production priority score is optimized by using an economic production batch model. The optimal production batch is determined based on the demand gap and production cost parameters. If the demand gap is small, the actual batch is equal to the demand gap. If the demand gap is moderate, the actual batch is equal to the economic batch. If the demand gap is large, the demand gap is split into multiple batches.

8. The method for collaborative management of pesticide production orders and inventory according to claim 1, characterized in that, The generation of the raw material procurement plan includes: Based on the preliminary production plan and product formula data, the bill of materials expansion method is used to calculate the raw material requirements of the preliminary production plan. The planned production quantity of each product is multiplied by the proportion coefficient of each raw material in the formula and summed according to the type of raw material to obtain the total demand of each type of raw material. The time-matching analysis method is used to assess the supply capacity of the total demand for raw materials. The latest order time for raw materials is calculated as the planned start time of the earliest production task using this raw material minus the supply cycle and the raw material warehousing and inspection time. If the regular supply cycle exceeds the time difference but the shortest supply cycle can meet the demand, there is a supply risk and expedited supply is required. If the shortest supply cycle still exceeds the time difference, the raw materials cannot be supplied on time and the preliminary production plan needs to be adjusted.

9. The method for collaborative management of pesticide production orders and inventory according to claim 1, characterized in that, The updates to the production plan and order delivery plan include: Based on the collaboratively optimized production plan and real-time production data, a disturbance impact assessment method is used to identify disturbances in the real-time production data. The scope of tasks that need to be adjusted is determined by calculating the degree of impact of disturbance events on each production task. An adaptive scheduling method is used to optimize the scheduling based on the disturbance impact assessment results. The corresponding scheduling strategy is selected according to the disturbance type and impact range. For equipment failure disturbances, a task reassignment strategy is adopted; for raw material shortage disturbances, a task postponement strategy is adopted; and for urgent order disturbances, a task insertion strategy is adopted. The production plan and order delivery plan are updated accordingly.

10. A computer-readable storage medium, characterized in that, It is used to store computer-readable instructions, which, when read by a computer, enable the execution of a pesticide production order and inventory collaborative management method as described in any one of claims 1-9.

Images