Intelligent optimization method and system for monthly production plan adjustment oriented to market demand influence

Abstract

The invention provides a market demand influence-oriented monthly production plan adjustment intelligent optimization method and system, and relates to the technical field of intelligent manufacturing and production plan optimization. The method specifically comprises the steps of obtaining commodity future data of a target bulk commodity, and constructing a two-dimensional observation vector sequence; constructing an HMM demand prediction model and performing training; inputting the two-dimensional observation vector sequence into a trained HMM demand prediction model, and performing state decoding on the output of the model to obtain a demand state sequence and a state transition matrix; a multi-scene plan generation model is used for monthly production planning, and an effective alternative plan scheme set of the target bulk commodity is obtained; and screening the effective alternative plan scheme set by using a production utility evaluation model to obtain an optimal plan scheme. According to the invention, three links of market demand prediction, multi-scene plan generation and intelligent evaluation optimization are organically connected in series to form a complete and data-driven decision closed loop.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and production planning optimization technology, and in particular to an intelligent optimization method and system for adjusting monthly production plans in response to market demand. Background Technology

[0002] In the field of commodity production management, companies typically formulate monthly production plans based on historical data and static models. Currently, existing methods for adjusting and optimizing monthly production plans mainly rely on periodic (e.g., monthly or quarterly) plan revisions. When market demand changes, common practices include:

[0003] Manual adjustments based on rules and experience: Based on recent sales data and market feedback, planners manually adjust parameters such as output and variety ratio in the original monthly plan according to their experience.

[0004] Rolling planning mechanism: Some systems adopt a rolling planning approach, for example, updating the plan for the following months monthly based on the latest actual sales and inventory data. Its core is mostly based on time series forecasting models (such as ARIMA, exponential smoothing) to predict demand at specific points and generate a single optimal plan accordingly.

[0005] Combining traditional operations research methods: Some optimization methods introduce linear programming or integer programming models to solve for the optimal production quantity given demand forecasts and objective functions (such as minimum cost and maximum profit).

[0006] Furthermore, in the planning and evaluation phase, tools such as Data Envelopment Analysis (DEA) and Analytic Hierarchy Process (AHP) are used to assess the efficiency of production plans or to make multi-indicator decisions. These existing methods can play a role when market demand is relatively stable or changes slowly, and also provide a basic framework for enterprise production planning management.

[0007] However, while existing methods have been applied in the formulation of monthly production plans for commodities, they still exhibit the following gradually emerging technical limitations when faced with dynamic changes in market demand:

[0008] 1. Existing methods lack the ability to dynamically identify and predict market demand. They rely heavily on static historical data or single-scenario forecasts, failing to quantify and capture the patterns of market shifts. This results in planning adjustments often lagging behind actual demand changes, leading to a disconnect between plans and actual demand.

[0009] 2. Due to the lack of modeling for the multi-state evolution of demand, existing methods struggle to systematically generate alternative plans covering different market scenarios and risk appetites. When facing uncertainty, enterprises lack a diverse and flexible set of plans, making it difficult to optimize production targets while maintaining controllable risks.

[0010] 3. In the planning and evaluation stage, existing evaluation methods have failed to deeply integrate efficiency quantification and multi-objective decision-making. The evaluation process often relies on human experience or a single indicator, lacking an objective evaluation system that comprehensively considers demand matching, resource utilization and production stability, resulting in one-sided and highly subjective evaluation results.

[0011] In summary, existing methods suffer from insufficient overall adaptability in the formulation of monthly production plans for bulk commodities. They are unable to achieve rapid, closed-loop dynamic optimization and adjustment of monthly production plans when external markets fluctuate frequently, thus limiting enterprises' agility in responding to market changes and the efficiency of production resource utilization. Summary of the Invention

[0012] To address the shortcomings of the existing technologies, this invention proposes a systematic technical solution that integrates a Hidden Markov Model (HMM) demand forecasting model, a multi-scenario plan generation model, and a production plan utility evaluation model. This solution aims to achieve dynamic optimization of production plans by constructing a systematic technical solution that integrates a Hidden Markov Model (HMM) demand forecasting model, a multi-scenario plan generation model, and a production plan utility evaluation model.

[0013] On the one hand, this invention proposes an intelligent optimization method for adjusting monthly production plans in response to market demand, which includes the following process:

[0014] Acquire market observation data, historical operational data of enterprises, and production resource and constraint data for the target commodities;

[0015] Based on the market observation data, a two-dimensional observation vector sequence for the target commodity is constructed;

[0016] Construct and train an HMM demand forecasting model to obtain a trained HMM demand forecasting model.

[0017] The two-dimensional observation vector sequence is input into the trained HMM demand prediction model, and the output of the model is decoded to obtain the demand state sequence and state transition matrix of the target commodity. ;

[0018] Based on the enterprise's historical operational data, production resource and constraint data, demand state sequence, and state transition matrix By using a multi-scenario planning generation model, monthly production planning is carried out to obtain a set of effective alternative planning schemes for the target bulk commodities;

[0019] Based on the production resources and constraints data, the optimal plan is obtained by screening the set of effective alternative plans using a production utility evaluation model.

[0020] Furthermore, the market observation data includes: historical futures price series and historical trading volume series; the enterprise historical operating data includes: historical commodity demand; and the production resource and constraint data includes: maximum equipment capacity and process limitations.

[0021] Furthermore, the specific content of constructing the two-dimensional observation vector sequence of the target commodity based on the market observation data is as follows:

[0022] Using the obtained historical futures price series, calculate the logarithmic return of the target commodity for each period; using the obtained historical trading volume series, calculate the logarithmic return of the target commodity for each period.

[0023] By taking the logarithmic return of price and the logarithmic return of trading volume in the same period as a set of observation vectors, a two-dimensional observation vector sequence of the commodity can be obtained.

[0024] Furthermore, the specific content of constructing and training the HMM demand prediction model to obtain the trained HMM demand prediction model is as follows:

[0025] Define the HMM demand forecasting model as: quintuple ;in The set of hidden states includes three market demand states: high demand, medium demand, and low demand. It is a two-dimensional observation vector sequence. , For the first The observation vector for the period; Let be the initial state probability distribution vector. , For state The state probability distribution, ; The state transition probability matrix is... , From state Transition to state The probability, ; Let be the observation probability density function. , In the state Generate observation vector The probability density;

[0026] Get the length is Commodity observation sequence and construct a length of observation sequence ;in Indicates the first Price data for the period; Indicates the first Trading volume data for the period; and For the first and second phases The observation vector for the period;

[0027] According to the observation sequence An adaptive window mechanism based on volatility quantiles is used to determine the observed sequence. The initial hidden state sequence;

[0028] Based on observation sequence and in the observation sequence The corresponding initial hidden state sequence is used to initialize the initial state probability distribution vector in the HMM demand forecasting model. State transition probability matrix and observation probability density function ;

[0029] The Baum-Welch algorithm is used to iteratively train the HMM demand forecasting model. The following operations are performed during each iteration:

[0030] Based on observation sequence Calculate the forward probability and backward probability for the HMM demand forecasting model under the current iteration round, respectively;

[0031] Calculate the state probability using the forward and backward probabilities. and transition probability ;

[0032] Based on state probability and transition probability Update the state transition probability matrix and observation probability density function ;

[0033] Determine whether the log-likelihood function has converged in the current iteration round. If it has converged, end the iterative training and obtain the trained HMM demand prediction model. If it has not converged, continue to the next iteration round until the log-likelihood function converges.

[0034] Furthermore, the observation sequence An adaptive window mechanism based on volatility quantiles is used to determine the observed sequence. The specific contents of the initial hidden state sequence are as follows:

[0035] Calculate the observation sequence The rolling annualized volatility series;

[0036] Obtain the median and third quartile of historical volatility;

[0037] For the Rolling annualized volatility over the period If the rolling annualized volatility If the value is greater than the third or fourth quartile, then select the first window length; if the rolling annualized volatility... If the volatility is greater than the median of historical volatility but not greater than the third quartile, then select the second window length; if the rolling annualized volatility... If the value is not greater than the median of historical volatility, then the third window length is selected.

[0038] According to the Rolling annualized volatility over the period Calculate the first using the selected window length. Rolling average rate of return over the period ;

[0039] The first Rolling average rate of return over the period Compare with the preset demand state division threshold to determine the first Hidden state during the period To determine if a demand is high, medium, or low, and thus obtain the observed sequence... The corresponding hidden state sequence.

[0040] Furthermore, the multi-scenario plan generation model is as follows:

[0041] The planning period for setting the monthly production plan is Month, for the first month of the planning period Expect, Based on the enterprise's historical operational data, demand state sequence, and state transition matrix Calculate the future number of... Monthly baseline forecasts under different demand conditions;

[0042] Three risk preference types are defined, and Gaussian noise is used to generate perturbations for each type of risk preference.

[0043] For the demand state and risk appetite Any combination of scenario solutions For the future Demand status Monthly baseline forecast Repeated application Secondary disturbance ,get The future Expected in scenario solution combination The following demand forecasts; of which ;

[0044] Using a preset mapping function, the future Monthly scenario solution combination The demand forecasts are mapped to production plans to obtain future... Monthly scenario solution combination The following are alternative plans and schemes;

[0045] Based on the aforementioned production resources and constraint data, a filtering function is defined, and this filtering function is used to filter all future... Monthly scenario solution combination Constraint filtering is performed on the alternative plans to obtain the future Monthly scenario solution combination A set of effective alternative plans and schemes.

[0046] Furthermore, the data is based on the enterprise's historical operational data, demand state sequences, and state transition matrix. Calculate the future number of... The method for calculating the monthly baseline forecast under different demand conditions is as follows:

[0047] Based on the aforementioned demand state sequence, the enterprise's historical operational data is divided into a set of days with high market demand. Daily collection of medium market demand status and daily collection of low market demand status Three categories of demand states are defined, and the historical daily average demand value is calculated for each category of demand state subset. ;

[0048] Define the state transition probability matrix of Step transfer power ;

[0049] For the Demand status during the period Using the aforementioned transfer power Calculate the future The state probability distribution of the period;

[0050] Based on the historical daily average demand value of each subset of demand states and future chapters The probability distribution of the state during the period is used to calculate the monthly baseline forecast value for each type of demand state.

[0051] Furthermore, the specific details of using a production utility evaluation model to screen the set of effective alternative plans based on the production resources and constraint data to obtain the optimal plan are as follows:

[0052] Based on the aforementioned production resources and constraints data, for the future... Monthly scenario solution combination The following is a set of effective alternative plans. Using the DEA efficiency evaluation system based on the BCC model, the set of effective alternative plans was calculated respectively. The planned benefit value of each valid alternative plan; among which Representing a combination of scenario solutions The number of valid alternative plans; Representing a combination of scenario solutions The next One effective alternative plan;

[0053] All calculated planned benefit values ​​are normalized to obtain normalized planned benefit values;

[0054] Build a combination of scenario solutions The AHP hierarchy is used to calculate the combination of scenario solutions using the normalized plan benefit values. A comprehensive score for each valid alternative plan;

[0055] From the set of valid alternative plans In the selection process, the effective alternative plan with the highest overall score will be chosen as the future plan. Monthly scenario solution combination The optimal plan is as follows.

[0056] Furthermore, the combination of construction scenario schemes The AHP hierarchy is used to calculate the combination of scenario solutions using the normalized plan benefit values. The specific content of the comprehensive score for each valid alternative plan is as follows:

[0057] Build a combination of scenario solutions The AHP hierarchy includes: target layer, criteria layer, and scheme layer;

[0058] The target layer is: selecting a combination of scenario solutions. The optimal production plan under the given conditions;

[0059] The criteria layer includes three dimensions: DEA overall efficiency. Resource utilization rate Matching degree with needs ;

[0060] The solution layer is: a combination of scenario solutions. The following is a set of effective alternative plans;

[0061] For scenario solution combinations Next One effective alternative plan Effective alternative plans Normalized planned benefit value As DEA overall efficiency index ;

[0062] Based on the aforementioned production resources and constraints data, calculate effective alternative plans. Resource utilization rate indicators ;

[0063] According to the effective alternative plans Demand forecasts and calculation of effective alternative plans. Demand matching index ;

[0064] Construction of scene scheme combination The criterion layer judgment matrix is ​​calculated, and the global weight vector of the criterion layer is calculated. At the same time, the consistency check of the criterion layer judgment matrix is ​​performed. If the check fails, the criterion layer judgment matrix is ​​modified until the check succeeds.

[0065] According to the effective alternative plans DEA overall efficiency index Resource utilization rate indicators Matching index with demand Using the global weight vector corresponding to the successfully validated criterion layer judgment matrix, effective alternative plans are calculated by weighted summation. The overall score.

[0066] On the other hand, this invention proposes an intelligent optimization system for adjusting monthly production plans in response to market demand, the system comprising:

[0067] The data acquisition module is used to acquire market observation data, historical operational data of enterprises, and production resource and constraint data of the target bulk commodities;

[0068] An observation sequence construction module is used to construct a two-dimensional observation vector sequence of the target commodity based on the market observation data.

[0069] The HMM model training module is used to build and train the HMM demand prediction model to obtain a trained HMM demand prediction model.

[0070] The demand state decoding module is used to input the two-dimensional observation vector sequence into the trained HMM demand prediction model and perform state decoding on the output of the model to obtain the demand state sequence and state transition matrix of the target commodity.

[0071] The multi-scenario planning generation module is used to perform monthly production planning based on the enterprise's historical operational data, production resource and constraint data, demand state sequence and state transition matrix, and to obtain a set of effective alternative planning schemes for the target bulk commodity.

[0072] The production planning decision module is used to filter the set of effective alternative plans based on the production resources and constraint data using a production utility evaluation model to obtain the optimal plan.

[0073] The beneficial effects of adopting the above technical solution are as follows:

[0074] The method and system of this invention construct an integrated dynamic optimization closed loop for production planning, encompassing "prediction-generation-evaluation." Its core concept lies in organically linking market demand forecasting, multi-scenario plan generation, and intelligent evaluation and optimization into a complete, data-driven decision-making closed loop, resulting in the following multi-layered beneficial effects:

[0075] Addressing the core shortcomings of existing technologies that rely on static data and single forecasts, leading to planning delays, this invention's method and system construct a Hidden Markov Model (HMM) demand forecasting model and innovatively introduce an adaptive window mechanism. This enables real-time decoding of market demand status and its transition probabilities from commodity futures data. This transforms production planning from being based on outdated information to being based on quantitative predictions of future market trends. Consequently, production planning adjustments are elevated from "post-event response" to "pre-event prediction," significantly enhancing the plan's foresight and alignment with actual demand, and achieving dynamic perception and proactive response to market demand fluctuations.

[0076] To address the shortcomings of existing methods in rapidly generating diverse alternative solutions, this invention's method and system combine the demand state output by the Hidden Markov Model (HMM) with different risk preferences (aggressive / neutral / conservative) through a multi-scenario plan generation model. Utilizing a Gaussian noise perturbation mechanism, it systematically generates nine sets of alternative plans covering different market scenarios. This fundamentally solves the vulnerability of traditional single-plan systems to uncertainty, providing enterprise decision-makers with a decision-making space covering multiple scenarios and risk preferences, greatly enhancing the adaptability and decision-making flexibility of the production planning system.

[0077] To address the shortcomings of traditional evaluation methods, such as strong subjectivity and single indicators, this invention creatively integrates the objective efficiency quantification capability of DEA with the multi-criteria weight allocation capability of AHP, constructing a production plan utility evaluation model. This evaluation model not only comprehensively considers key indicators such as demand fulfillment rate, production efficiency, and resource utilization rate, but also adjusts indicator weights according to different scenario characteristics. This ensures the scientific and comprehensive nature of solution selection and overcomes the arbitrariness of human evaluation. By establishing an objective, comprehensive, and realistic intelligent evaluation and selection mechanism, this invention ensures that the final selected optimal production plan achieves the best balance between efficiency, risk, and demand matching.

[0078] This invention's method and system, through the closed-loop collaboration of a Hidden Markov Model (HMM) demand forecasting model, a multi-scenario plan generation model, and a production plan utility evaluation model, forms a complete optimization chain from "market dynamic perception" to "intelligent generation of multiple solutions" and then to "comprehensive evaluation and selection," achieving a systematic improvement in production efficiency and resource utilization. Its core technologies, such as HMM-based state recognition, multi-scenario generation, and fusion evaluation, possess good universality and scalability. Therefore, this method can not only be effectively applied to the mining field but also provides a replicable technical path and systematic solution for optimizing and adjusting production plans for other commodities facing dynamic fluctuations in market demand (such as iron ore and rare earths), possessing strong industry promotion value.

[0079] In summary, the method and system of this invention abandon the traditional single-plan generation model. It innovatively combines the market demand state (high / medium / low) decoded by the Hidden Markov Model (HMM) with the decision-maker's risk preference (aggressive / neutral / conservative) in a matrix manner, and utilizes Gaussian noise perturbation to simulate market uncertainty, systematically generating a set of alternative plan schemes covering multiple future scenarios. In other words, the method and system of this invention provide a complete, data-driven methodology and system framework, establishing a multi-scenario plan generation mechanism driven by demand state and risk preference. This mechanism ensures that market uncertainty and decision-making flexibility are fully considered at the front end of production planning, providing a rich and adaptable decision-making basis for subsequent optimization, and is applicable to bulk commodity production. Attached Figure Description

[0080] Figure 1 This is a flowchart of the intelligent optimization method for adjusting monthly production plans in response to market demand in this embodiment;

[0081] Figure 2 This is a schematic diagram illustrating the intelligent optimization method for adjusting monthly production plans in response to market demand in this embodiment.

[0082] Figure 3 This is a flowchart of the multi-scenario plan generation model in this embodiment;

[0083] Figure 4 This is a flowchart of the production planning utility evaluation model in this embodiment;

[0084] Figure 5 This is a schematic diagram of the AHP hierarchy in this embodiment;

[0085] Figure 6 This is a structural diagram of the intelligent optimization system for adjusting monthly production plans in response to market demand in this embodiment. Detailed Implementation

[0086] To facilitate understanding of this application, specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The following embodiments are illustrative of the invention but are not intended to limit its scope. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0087] Example 1:

[0088] In this embodiment, firstly, Hidden Markov Models (HMMs) are used to decode market demand status (e.g., high / medium / low) from commodity futures price and trading volume fluctuations, providing a basis for the state transition probability distribution for generating dynamic plans. Secondly, based on demand status classification and risk preference grading, a Gaussian noise perturbation mechanism is used to apply three levels of intensity perturbation to the three types of benchmark demand forecasts, generating nine alternative planning schemes. A process constraint filtering mechanism is introduced to filter out planning schemes that do not meet production constraints, forming a multi-scenario alternative planning scheme that takes into account both market demand evolution and risk controllability. Finally, by integrating the efficiency quantification of DEA and the weight allocation of AHP, an intelligent evaluation system is constructed to comprehensively evaluate the generated alternative planning schemes, selecting the optimal production plan under various scenarios. This realizes the entire process of "prediction-generation-evaluation" for formulating production plans, providing a systematic solution for adaptive decision-making in production planning under dynamic market environments.

[0089] Specifically, this embodiment provides an intelligent optimization method for adjusting monthly production plans in response to market demand, such as... Figure 1 As shown, the method includes the following steps:

[0090] Obtain market observation data, enterprise historical operating data, and production resources and constraints data for the target commodity.

[0091] In this embodiment, as Figure 2 As shown, the commodity futures data that needs to be obtained includes: market observation data, historical operational data of enterprises used to establish demand-production benchmark mapping, and production resource and constraint data used for planning generation and filtering.

[0092] The market observation data includes: historical futures price series and historical trading volume series; the enterprise historical operating data includes: historical commodity demand; the production resources and constraints data includes: maximum equipment capacity and process limitations.

[0093] Based on the market observation data, a two-dimensional observation vector sequence for the target commodity is constructed.

[0094] The specific content of constructing the two-dimensional observation vector sequence of the target commodity based on the market observation data is as follows:

[0095] Using the obtained historical futures price series, calculate the logarithmic return of the target commodity for each period; using the obtained historical trading volume series, calculate the logarithmic return of the target commodity for each period.

[0096] By taking the logarithmic return of price and the logarithmic return of trading volume in the same period as a set of observation vectors, a two-dimensional observation vector sequence of the commodity can be obtained.

[0097] Construct and train an HMM demand forecasting model to obtain a trained HMM demand forecasting model.

[0098] In this embodiment, the HMM demand forecasting model dynamically identifies and predicts the state of market demand. Its core lies in establishing a probabilistic correlation between the market demand state (high, medium, low) that cannot be directly observed and the observable commodity futures data (price, trading volume), thereby decoding the state sequence of market demand and its evolution law.

[0099] The specific content of constructing and training the HMM demand prediction model to obtain the trained HMM demand prediction model is as follows:

[0100] Define the HMM demand forecasting model as: quintuple ;in The set of hidden states includes three market demand states: high demand, medium demand, and low demand. It is a two-dimensional observation vector sequence, and , For the first Periodic observation vector, , For the first Logarithmic return of prices over a period of time For the first Logarithmic return of trading volume over a period; Let be the initial state probability distribution vector. , For state The state probability distribution, ; This is the state transition probability matrix, used to describe the state transition patterns in the market. , From state Transition to state The probability, ; Let be the observation probability density function. , In the state Generate observation vector The probability density.

[0101] In this embodiment, for the state transition probability estimation part of the HMM demand forecasting model, the Baum-Welch algorithm is used for a given length of observation sequence Perform parameter estimation.

[0102] Get the length is Commodity observation sequence and construct a length of observation sequence ;in Indicates the first Price data for the period; Indicates the first Trading volume data for the period; and For the first observation vector and the second observation vector... Periodic observation vector. Periodic observation vector Represented as:

[0103] (1)

[0104] in, Indicates the first Price data for the period; Indicates the first Price data for the period; Indicates the first Trading volume data for the period; Indicates the first The transaction volume data for the period. In this embodiment, using... Calculate logarithmic returns and construct a series of observed variables to reflect the dynamic volatility of market prices.

[0105] According to the observation sequence An adaptive window mechanism based on volatility quantiles is used to determine the observed sequence. The initial hidden state sequence.

[0106] To adapt to the characteristics of market volatility, this embodiment proposes an adaptive window mechanism based on volatility quantiles.

[0107] According to the observation sequence An adaptive window mechanism based on volatility quantiles is used to determine the observed sequence. The specific contents of the initial hidden state sequence are as follows:

[0108] Calculate the observation sequence The rolling annualized volatility series.

[0109] In this embodiment, for financial and commodity markets, 20 trading days typically correspond to approximately one month (calculated based on an average of approximately 20-22 trading days per month). Furthermore, since this embodiment is geared towards monthly production plan adjustments, selecting 20 days directly aligns with the monthly planning cycle, facilitating the link between short-term fluctuations and monthly demand forecasts.

[0110] (2)

[0111] in, Indicates the first The rolling annualized volatility over the period; Indicates standard deviation; Indicates the first Periodic observation vector; This is the annualized factor.

[0112] Obtain the median and third quartile of historical volatility.

[0113] For the Rolling annualized volatility over the period If the rolling annualized volatility If the value is greater than the third or fourth quartile, then select the first window length; if the rolling annualized volatility... If the volatility is greater than the median of historical volatility but not greater than the third quartile, then select the second window length; if the rolling annualized volatility... If the value is not greater than the median of historical volatility, then the third window length is selected.

[0114] In this embodiment, the median (50th percentile) and third quartile (75th percentile) of historical volatility are dynamically selected with varying window lengths, as shown below:

[0115] (3)

[0116] in, Indicates the first The selected window length; Indicates the first Annualized rolling volatility of prices over the period; This represents the third and fourth quartile of historical price volatility. This represents the median of historical price volatility. Indicates the first The annualized volatility of trading volume over the period; This represents the third and fourth quartile of historical volatility in trading volume; This represents the median of historical volatility in trading volume.

[0117] Based on the selected window length, according to the... Rolling annualized volatility over the period Calculate the first Rolling average rate of return over the period .

[0118] The first Rolling average rate of return over the period Compare with the preset demand state division threshold to determine the first Initial hidden state of the period To determine if a demand is high, medium, or low, and thus obtain the observed sequence... The corresponding initial hidden state sequence.

[0119] In this embodiment, the rolling average rate of return is calculated based on a dynamic window. Divide the demand status.

[0120] (4)

[0121] in, Indicates the first The hidden state during the period; Indicates the first The rolling average return on prices over the period; Indicates the first The rolling average return on trading volume over a period of time.

[0122] It should be noted that the demand status classification threshold is an empirical parameter set based on historical data statistical characteristics. In this embodiment, the price demand status classification threshold is set as follows: The threshold for classifying the demand state of transaction volume is: (i.e., price yield) Transaction volume return In practical applications, adjustments can be made based on the volatility characteristics of the target product and the needs of the enterprise.

[0123] Based on observation sequence and in the observation sequence The corresponding initial hidden state sequence is used to initialize the initial state probability distribution vector in the HMM demand forecasting model. State transition probability matrix and observation probability density function .

[0124] In this embodiment, the initial state probability distribution vector Set as state State probability distribution The state transition probability matrix Initialize as a near-diagonal dominance matrix, and set the observation probability density function. Observational distribution parameters Initialize using the tertiary of the rate of return.

[0125] The Gaussian distribution is used to characterize the observed returns, i.e., the state. Generate observation vector probability density for:

[0126] (5)

[0127] in It is a mean vector used to represent the state. The corresponding average rate of return, ; Representing state The corresponding mean price return; Representing state The corresponding mean of the observations; Let represent the covariance matrix.

[0128] The Baum-Welch algorithm is used to iteratively train the HMM demand forecasting model. The following operations are performed during each iteration:

[0129] Based on observation sequence Calculate the forward probability and backward probability for the HMM demand forecasting model in the current iteration round.

[0130] In this embodiment, for a given HMM demand forecasting model , define the first Hidden state during the period For state forward variables for The recursive formula for the forward variable is:

[0131] (6)

[0132] (7)

[0133] in This is the observation vector for the first period; In the state Generate observation vector The probability density; Hidden status in Phase 1 For state The forward variable; For the first Hidden state during the period For state The forward variable; For the first Periodic observation vector; In the state Generate observation vector The probability density is obtained by recursively calculating all forward variables and summing the forward variables over all states at the final time step. The overall probability of the observed sequence is the forward probability.

[0134] In this embodiment, for a given HMM demand forecasting model , define the first Hidden state during the period For state backward variables for , For the first The recursive formula for the backward variable of the periodic observation vector is:

[0135] (8)

[0136] (9)

[0137] in, For the first Hidden state during the period For state The backward variable; For the first Hidden state during the period For state The backward variable; In the state Generate observation vector The probability density; For the first Hidden state during the period For state The backward variable is then used to calculate the backward probability.

[0138] Calculate the state probability using the forward and backward probabilities. and transition probability .

[0139] In this embodiment, all backward variables obtained through recursive calculation are combined with forward variables in the Baum-Welch algorithm to calculate the state probabilities. and transition probability This updates the model parameters.

[0140] (10)

[0141] in, Indicates the first The period is in a state And in Shifting to a new state The joint probability, i.e. the transition probability; This indicates that, given a Hidden Markov Model (HMM) demand forecasting model... Below, the observation vector is obtained. The conditional probability.

[0142] Based on state probability and transition probability Update the state transition probability matrix and observation probability density function .

[0143] In this embodiment, the desired update state transition probability matrix is ​​maximized. .

[0144] (11)

[0145] in, Indicates from state Transition to state Updated probabilities; express Always in a state of readiness And in Shifting to a new state The joint probability.

[0146] In this embodiment, updates are performed using maximum likelihood estimation. and This updates the observation probability density function. .

[0147] (12)

[0148] (13)

[0149] in, This represents the updated mean vector; state The posterior probability, and ; This represents the updated covariance matrix.

[0150] Determine whether the log-likelihood function has converged in the current iteration round. If it has converged, end the iterative training and obtain the trained HMM demand prediction model. If it has not converged, continue to the next iteration round until the log-likelihood function converges.

[0151] In this embodiment, the Baum-Welch algorithm is executed until the log-likelihood function converges:

[0152] (14)

[0153] in Represented as the first Hidden state during the period For state The forward variable; The convergence threshold is, and .

[0154] The two-dimensional observation vector sequence is input into the trained HMM demand prediction model, and the output of the model is decoded to obtain the demand state sequence and state transition matrix of the target commodity. .

[0155] In this embodiment, the Viterbi algorithm is used to decode the output state into the required state sequence. The Hidden Markov Model (HMM) demand forecasting model includes: demand state sequence. and state transition matrix The former enables the dynamic decoding of market demand trends, while the latter serves as a parameter expression for quantifying market evolution patterns. Both provide an expression of uncertain market demand for the dynamic optimization of production plans.

[0156] Multi-scenario plan generation models generate alternative plan solutions that match the evolution of market demand by leveraging the synergistic effect of demand status and risk appetite. For example... Figure 3 As shown, alternative plans are generated through a three-stage process: calculating the market demand baseline, superimposing Gaussian noise to generate disturbance prediction, and filtering process constraints.

[0157] Based on the enterprise's historical operational data, production resource and constraint data, demand state sequence, and state transition matrix By using a multi-scenario planning generation model, monthly production planning is carried out to obtain a set of effective alternative planning schemes for the target bulk commodities.

[0158] The multi-scenario plan generation model is as follows:

[0159] The planning period for setting the monthly production plan is Month, for the first month of the planning period Expect, Based on the enterprise's historical operational data, demand state sequence, and state transition matrix Calculate the future number of... Monthly baseline forecasts under different demand conditions.

[0160] The data is based on the enterprise's historical operating data, demand state sequence, and state transition matrix. Calculate the future number of... The method for calculating the monthly baseline forecast under different demand conditions is as follows:

[0161] Based on the demand state sequence, the enterprise's historical operating data is divided into a set of days with high demand states. Daily collection of medium demand status and low-demand daily collection Three categories of demand states are defined, and the historical daily average demand value is calculated for each category of demand state subset. .

[0162] In this embodiment, the market demand state sequence is based on HMM decoding. The historical demand for goods is divided into sets of days with high demand. Daily collection of medium demand status Low demand status daily collection Three subsets of demand states were used to calculate the historical daily average demand for each state. .

[0163] (15)

[0164] in For state The corresponding set of time indices; For the first Historical demand for the product during the period.

[0165] Define the state transition probability matrix of The step transition power is:

[0166] (16)

[0167] in Indicates from state go through Step transition to state The probability of.

[0168] For the Demand status during the period Using the aforementioned transfer power Calculate the future The probability distribution of the state during the period.

[0169] (17)

[0170] in All are initial state vectors, represented using the Kronecker function:

[0171] (18)

[0172] Based on the historical daily average demand value of each subset of demand states and future chapters The probability distribution of the state during the period is used to calculate the monthly baseline forecast value for each type of demand state.

[0173] In this embodiment, the monthly baseline forecast values ​​for three types of demand states are constructed by combining the state transition probability with the historical daily demand average.

[0174] (19)

[0175] in Representing state The next chapter Monthly baseline forecast for the period; Indicates the future number Planned production days for the period; The state adjustment factor is expressed as:

[0176] (20)

[0177] in and These represent the adjustment range under high demand and low demand conditions, respectively.

[0178] Three risk preference types are defined, and Gaussian noise is used to generate perturbations for each type of risk preference.

[0179] In this embodiment, Gaussian noise It is random noise that follows a normal distribution, and its probability density function is:

[0180] (twenty one)

[0181] in, It is the probability density function; As a variable; The standard deviation of the normal distribution; It is the expectation of a normal distribution; It is the natural base.

[0182] The perturbation mechanism of Gaussian noise can simulate the characteristics of random fluctuations. To simulate the uncertainty of market demand fluctuations, three levels of perturbation strength (aggressive, neutral, and conservative) are designed to correspond to different risk preferences. Gaussian noise is introduced to predict the baseline demand under the three demand states. Apply three types of fluctuation amplitude perturbations The disturbances of the three fluctuation amplitudes are represented as follows:

[0183] (twenty two)

[0184] in, Indicates the future The corresponding risk appetite imposed during the period The disturbance ; Risk preference The corresponding perturbation standard deviation, in this embodiment, is... Determined by human intervention, such as through expert experience.

[0185] For the demand state and risk appetite Any combination of scenario solutions For the future Demand status Monthly baseline forecast Repeated application Secondary disturbance ,get The future Expected in scenario solution combination The demand forecast below.

[0186] (twenty three)

[0187] in, Indicates the monthly baseline forecast Apply a disturbance The subsequent demand forecast.

[0188] Repeated perturbation After that, we get the future. The corresponding demand forecast after the disturbance for the month:

[0189] (twenty four)

[0190] in, Indicates the future number Expected in scenario solution combination The next One demand forecast value; Indicates the future Monthly scenario solution combination The next One demand forecast value.

[0191] Using a preset mapping function, the future Monthly scenario solution combination The demand forecasts are mapped to production plans to obtain future... Monthly scenario solution combination The following are alternative plans and schemes.

[0192] In this embodiment, the nine types of perturbation-adjusted demand forecasts generated for three types of demand states and three types of risk preferences are mapped to production plans, resulting in alternative planning schemes for each scenario. The mapping function is as follows:

[0193] (25)

[0194] (26)

[0195] in, For the future Expected in scenario solution combination The next One alternative plan; For the future Monthly scenario solution combination The next One alternative plan; Indications and risk preferences The relevant adjustment amount; Representing the state of demand Related adjustment amounts.

[0196] Based on the production resources and constraint data, a filtering function is defined, and the filtering function is used to perform constraint filtering on all the obtained alternative plans to obtain a set of effective alternative plans.

[0197] After generating multiple alternative planning schemes for various scenarios, a process constraint filtering mechanism is needed to eliminate schemes that do not meet specific industry production conditions. In this embodiment, each alternative planning scheme is defined as follows: The set of constraints that need to be satisfied is Define the filtering function In order to screen alternative plans that meet the constraints.

[0198] (27)

[0199] This leads to a combination of scenario solutions. The number of valid alternative plans after filtering by process constraints. .

[0200] (28)

[0201] Ultimately, the future Monthly scenario solution combination The set of valid alternative plans is represented as follows:

[0202] (29)

[0203] in, Indicates the future Monthly scenario solution combination A set of effective alternative plans and schemes.

[0204] Based on the production resources and constraints data, the optimal plan is obtained by screening the set of effective alternative plans using a production utility evaluation model.

[0205] The nine alternative planning schemes generated by the multi-scenario planning generation model need to undergo systematic evaluation to select the optimal scheme for each scenario. The production planning utility evaluation model used in this embodiment first uses DEA to objectively quantify the utility level of each alternative planning scheme, then uses AHP to allocate the comprehensive weights of the evaluation indicators, and finally generates a comprehensive score. This enables the selection of the optimal solution for nine different scenarios. The logical process is as follows: Figure 4 As shown.

[0206] The specific details of the optimal plan, obtained by screening the set of effective alternative plans using a production utility evaluation model based on the production resource and constraint data, are as follows:

[0207] Regarding the future Monthly scenario solution combination The following is a set of effective alternative plans. Using the DEA efficiency evaluation system based on the BCC model, the set of effective alternative plans was calculated respectively. The planned benefit value of each valid alternative plan.

[0208] In this embodiment, the scenario scheme combination is described. The set of alternative plans Perform DEA efficiency evaluation by scenario. Based on the set of valid alternative plans... The corresponding market demand forecast and maximum equipment capacity are used to calculate the Demand Fulfillment Rate (DFR) and Production Efficiency (PE) for each valid alternative plan as inputs to DEA. These represent the degree of matching between output and market demand, and the utilization efficiency of equipment capacity when meeting market demand, respectively.

[0209] (30)

[0210] (31)

[0211] in, Representing a combination of scenario solutions The demand satisfaction rate is below; Indicating a combination of scenario solutions The next One effective alternative plan; Indicates valid alternative plans Corresponding demand forecast; Representing a combination of scenario solutions Production efficiency; This indicates the maximum production capacity of the equipment.

[0212] In implementing DEA, the Banker-Charnes-Cooper (BCC) model, which can handle variable returns to scale, is used as the core framework for efficiency evaluation. For each alternative plan, the following linear programming problem is solved using the BCC model, forming a DEA efficiency evaluation system based on the BCC model, expressed as:

[0213] (32)

[0214] (33)

[0215] in, Representing a combination of scenario solutions Next The planned benefit value of one effective alternative plan; Indicates the first The first of the effective alternative plans The average of the output indicators. , Indicates the first The first of the effective alternative plans Output index value, These correspond to demand fulfillment rate and production efficiency, respectively. For the first The weight of each output indicator; This is the intercept term, used to reflect the returns to scale characteristics; Representing a combination of scenario solutions Any valid alternative plan or scheme is available.

[0216] All calculated planned benefit values ​​are normalized to obtain normalized planned benefit values.

[0217] In this embodiment, to eliminate the dimensional differences in efficiency scores among different alternative plans and enhance comparability, the original plan benefit values ​​are further normalized to obtain... .

[0218] (34)

[0219] in, Representing a combination of scenario solutions Next Normalized plan benefit values ​​of each valid alternative plan; Representing a combination of scenario solutions The minimum planned benefit value among all valid alternative plans; Representing a combination of scenario solutions The maximum planned benefit value among all valid alternative plans.

[0220] Build a combination of scenario solutions The AHP hierarchy is used to calculate the combination of scenario solutions using the normalized plan benefit values. The overall score of each valid alternative plan is calculated.

[0221] In this embodiment, in order to select alternative plans for different scenarios, a system is constructed as follows: Figure 5 The AHP hierarchy shown generates global weights for comprehensive scoring.

[0222] The combination of construction scenario solutions The AHP hierarchy is used to calculate the combination of scenario solutions using the normalized plan benefit values. The specific content of the comprehensive score for each valid alternative plan is as follows:

[0223] Build a combination of scenario solutions The AHP hierarchy includes: the objective layer, the criteria layer, and the scheme layer.

[0224] The target layer is: selecting a combination of scenario solutions. The optimal production plan under the given conditions.

[0225] In this embodiment, the target layer is a single node. The optimal production plan selection is represented as:

[0226] (35)

[0227] in, Representing a combination of scenario solutions Next A comprehensive score of the 100 effective alternative plans; This indicates taking the maximum value.

[0228] The criteria layer includes three dimensions: DEA overall efficiency index. Resource utilization rate indicators Matching index with demand .

[0229] The solution layer is: a combination of scenario solutions. A set of effective alternative plans and schemes.

[0230] In this embodiment, the solution layer is a multi-node cluster. The set of alternative plans is represented as:

[0231] (36)

[0232] For scenario solution combinations Next One effective alternative plan Effective alternative plans Normalized planned benefit value As DEA overall efficiency index .

[0233] In this embodiment, the DEA overall efficiency index To retain the DEA synthesis efficiency value, and to cover DFR and PE to reflect the overall efficiency, it is expressed as:

[0234] (37)

[0235] Based on the aforementioned production resources and constraints data, calculate effective alternative plans. Resource utilization rate.

[0236] In this embodiment, the resource utilization (RU) metric The calculation method is as follows:

[0237] (38)

[0238] in, Indicates valid alternative plans Resource utilization rate.

[0239] According to the effective alternative plans Demand forecasts and calculation of effective alternative plans. Degree of matching with needs.

[0240] Scenario-based solution combination The resource utilization rate and demand matching degree of all valid alternative plans are analyzed, and the results are compared with those of the valid alternative plans. The resource utilization rate and demand matching degree are normalized, and the normalized resource utilization rate is used as... Resource utilization rate indicators The normalized demand matching degree is used as Demand matching index .

[0241] In this embodiment, the Demand Match Rate (DMR) metric This is used to indicate the degree of alignment between the plan and market demand, and its calculation method is expressed as follows:

[0242] (39)

[0243] Construction of scene scheme combination The criterion layer judgment matrix is ​​determined, and the global weight vector of the criterion layer is calculated. Simultaneously, a consistency check is performed on the criterion-level judgment matrix. If the check fails, the criterion-level judgment matrix is ​​modified until the check succeeds. The first in the criterion layer Dimensional indicators .

[0244] In this embodiment, the criterion-layer judgment matrix for each scenario is constructed, the global weight of the criterion layer is calculated, and a consistency check is performed, which must satisfy... This yields the global weight vector. To calculate the overall score.

[0245] According to the effective alternative plans DEA overall efficiency index Resource utilization rate indicators Matching index with demand The global weight vector corresponding to the criterion layer judgment matrix that has been successfully verified is used. Calculate effective alternative plans The overall score.

[0246] (40)

[0247] in, Indicates valid alternative plans Overall score; Indicates valid alternative plans The Dimensional indicators.

[0248] From the set of valid alternative plans In the selection process, the effective alternative plan with the highest overall score will be chosen as the future plan. Monthly scenario solution combination The optimal plan is as follows.

[0249] In this embodiment, the scenario Optimal plan The selection rule is:

[0250] (41)

[0251] in, Representing a combination of scenario solutions The optimal plan is as follows.

[0252] In this embodiment, a complete optimization chain is formed through the closed-loop collaboration of the Hidden Markov Model (HMM) demand forecasting model, the multi-scenario plan generation model, and the production plan utility evaluation model, from "market dynamic perception" to "intelligent generation of multiple solutions" and then to "comprehensive evaluation and selection." Simulation experiments were conducted using mining production as a background for verification. The results show that, compared to the original plan, the optimal plans generated in this embodiment effectively improve production efficiency, resource utilization, and demand matching in each scenario. This proves that the present invention is not a localized improvement, but rather a substantial increase in overall operational efficiency through systemic innovation.

[0253] Example 2:

[0254] This embodiment presents an intelligent optimization system for adjusting monthly production plans in response to market demand, such as... Figure 6 As shown, the system includes:

[0255] The data acquisition module is used to acquire market observation data, historical operational data of enterprises, and production resource and constraint data for the target bulk commodities.

[0256] The observation sequence construction module is used to construct a two-dimensional observation vector sequence of the target commodity based on the market observation data.

[0257] The HMM model training module is used to build and train an HMM demand forecasting model to obtain a trained HMM demand forecasting model.

[0258] The demand state decoding module is used to input the two-dimensional observation vector sequence into the trained HMM demand prediction model and perform state decoding on the output of the model to obtain the demand state sequence and state transition matrix of the target commodity.

[0259] The multi-scenario planning generation module is used to perform monthly production planning based on the enterprise's historical operational data, production resources and constraint data, demand state sequence and state transition matrix, using a multi-scenario planning generation model to obtain a set of effective alternative planning schemes for the target bulk commodity.

[0260] The production planning decision module is used to filter the set of effective alternative plans based on the production resources and constraint data using a production utility evaluation model to obtain the optimal plan.

[0261] Example 3:

[0262] This embodiment proposes an electronic device, including: one or more processors, and a memory for storing instructions. When the instructions are executed by the one or more processors, the one or more processors execute the intelligent optimization method for adjusting monthly production plans based on market demand.

[0263] The electronic device may be a mobile phone, computer, or tablet computer, etc., and includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the intelligent optimization method for adjusting monthly production plans based on market demand, as described in the embodiments. It is understood that the electronic device may also include input / output (I / O) interfaces and communication components.

[0264] The processor is used to execute all or part of the steps in the intelligent optimization method for adjusting monthly production plans based on market demand, as described in the above embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in the electronic device, as well as application-related data.

[0265] The processor can be implemented as an Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic components, and is used to execute the intelligent optimization method for adjusting monthly production plans based on market demand as described in the above embodiments.

[0266] Example 4:

[0267] This embodiment proposes a computer-readable storage medium that stores executable instructions. When these instructions are executed, if they are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0268] The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the intelligent optimization method for adjusting monthly production plans in response to market demand as described in the various embodiments of this application.

[0269] The aforementioned storage media include: flash memory, hard disks, multimedia cards, card-type memory (e.g., SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR) memory), random access memory (RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, disks, optical discs, servers, APP (Application) app stores, and other media capable of storing program verification codes. These media store computer programs, which, when executed by a processor, can implement the various steps of the aforementioned intelligent optimization method for adjusting monthly production plans based on market demand.

[0270] Example 5:

[0271] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the aforementioned intelligent optimization method for adjusting monthly production plans in response to market demand.

[0272] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0273] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0274] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of this disclosure and its equivalents, then the intent of this disclosure also includes these modifications and variations.

Claims

1. A smart optimization method for adjusting monthly production plans in response to market demand, characterized in that: This method includes the following steps: Acquire market observation data, historical operational data of enterprises, and production resource and constraint data for the target commodities; Based on the market observation data, a two-dimensional observation vector sequence for the target commodity is constructed; Construct and train an HMM demand forecasting model to obtain a trained HMM demand forecasting model. The two-dimensional observation vector sequence is input into the trained HMM demand prediction model, and the output of the model is decoded to obtain the demand state sequence and state transition matrix of the target commodity. ; Based on the enterprise's historical operational data, production resource and constraint data, demand state sequence, and state transition matrix By using a multi-scenario planning generation model, monthly production planning is carried out to obtain a set of effective alternative planning schemes for the target bulk commodities; Based on the production resources and constraints data, the optimal plan is obtained by screening the set of effective alternative plans using a production utility evaluation model.

2. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 1, characterized in that, The market observation data includes: historical futures price series and historical trading volume series; the enterprise historical operating data includes: historical commodity demand; the production resources and constraints data includes: maximum equipment capacity and process limitations.

3. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 2, characterized in that, The specific content of constructing the two-dimensional observation vector sequence of the target commodity based on the market observation data is as follows: Using the obtained historical futures price series, calculate the logarithmic return of the target commodity for each period; using the obtained historical trading volume series, calculate the logarithmic return of the target commodity for each period. By taking the logarithmic return of price and the logarithmic return of trading volume in the same period as a set of observation vectors, a two-dimensional observation vector sequence of the commodity can be obtained.

4. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 2, characterized in that, The specific content of constructing and training the HMM demand prediction model to obtain the trained HMM demand prediction model is as follows: Define the HMM demand forecasting model as: quintuple ;in The set of hidden states includes three market demand states: high demand, medium demand, and low demand. It is a two-dimensional observation vector sequence. , For the first The observation vector for the period; Let be the initial state probability distribution vector. , For state The state probability distribution, ; The state transition probability matrix is... , From state Transition to state The probability, ; Let be the observation probability density function. , In the state Generate observation vector The probability density; Get the length is Commodity observation sequence and construct a length of observation sequence ;in Indicates the first Price data for the period; Indicates the first Trading volume data for the period; and For the first and second phases The observation vector for the period; According to the observation sequence An adaptive window mechanism based on volatility quantiles is used to determine the observed sequence. The initial hidden state sequence; Based on observation sequence and in the observation sequence The corresponding initial hidden state sequence is used to initialize the initial state probability distribution vector in the HMM demand forecasting model. State transition probability matrix and observation probability density function ; The Baum-Welch algorithm is used to iteratively train the HMM demand forecasting model. The following operations are performed during each iteration: Based on observation sequence Calculate the forward probability and backward probability for the HMM demand forecasting model under the current iteration round, respectively; Calculate the state probability using the forward and backward probabilities. and transition probability ; Based on state probability and transition probability Update the state transition probability matrix and observation probability density function ; Determine whether the log-likelihood function has converged in the current iteration round. If it has converged, end the iterative training and obtain the trained HMM demand prediction model. If it has not converged, continue to the next iteration round until the log-likelihood function converges.

5. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 4, characterized in that, According to the observation sequence An adaptive window mechanism based on volatility quantiles is used to determine the observed sequence. The specific contents of the initial hidden state sequence are as follows: Calculate the observation sequence The rolling annualized volatility series; Obtain the median and third quartile of historical volatility; For the Rolling annualized volatility over the period If the rolling annualized volatility If the value is greater than the third quartile, then select the first window length; If the rolling annualized volatility If the volatility is greater than the median of historical volatility but not greater than the third quartile, then the second window length is selected. If the rolling annualized volatility If the value is not greater than the median of historical volatility, then the third window length is selected. According to the Rolling annualized volatility over the period Calculate the first using the selected window length. Rolling average rate of return over the period ; The first Rolling average rate of return over the period Compare with the preset demand state division threshold to determine the first Hidden state during the period To determine if a demand is high, medium, or low, and thus obtain the observed sequence... The corresponding hidden state sequence.

6. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 5, characterized in that, The multi-scenario plan generation model is as follows: The planning period for setting the monthly production plan is Month, for the first month of the planning period Expect, Based on the enterprise's historical operational data, demand state sequence, and state transition matrix Calculate the future number of... Monthly baseline forecasts under different demand conditions; Three risk preference types are defined, and Gaussian noise is used to generate perturbations for each type of risk preference. For the demand state and risk appetite Any combination of scenario solutions For the future Demand status Monthly baseline forecast Repeated application Secondary disturbance ,get The future Expected in scenario solution combination The following demand forecasts; of which ; Using a preset mapping function, the future Monthly scenario solution combination The demand forecasts are mapped to production plans to obtain future... Monthly scenario solution combination The following are alternative plans and schemes; Based on the aforementioned production resources and constraint data, a filtering function is defined, and this filtering function is used to filter all future... Monthly scenario solution combination Constraint filtering is performed on the alternative plans to obtain the future Monthly scenario solution combination A set of effective alternative plans and schemes.

7. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 6, characterized in that, The data is based on the enterprise's historical operating data, demand state sequence, and state transition matrix. Calculate the future number of... The method for calculating the monthly baseline forecast under different demand conditions is as follows: Based on the aforementioned demand state sequence, the enterprise's historical operational data is divided into a set of days with high market demand. Daily collection of medium market demand status and daily collection of low market demand status Three categories of demand states are defined, and the historical daily average demand value is calculated for each category of demand state subset. ; Define the state transition probability matrix of Step transfer power ; For the Demand status during the period Using the aforementioned transfer power Calculate the future The state probability distribution of the period; Based on the historical daily average demand value of each subset of demand states and future chapters The probability distribution of the state during the period is used to calculate the monthly baseline forecast value for each type of demand state.

8. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 7, characterized in that, The specific details of the optimal plan, obtained by screening the set of effective alternative plans using a production utility evaluation model based on the production resource and constraint data, are as follows: Based on the aforementioned production resources and constraints data, for the future... Monthly scenario solution combination The following is a set of effective alternative plans. Using the DEA efficiency evaluation system based on the BCC model, the set of effective alternative plans was calculated respectively. The planned benefit value of each valid alternative plan; among which Representing a combination of scenario solutions The number of valid alternative plans; Representing a combination of scenario solutions The next One effective alternative plan; All calculated planned benefit values ​​are normalized to obtain normalized planned benefit values; Build a combination of scenario solutions The AHP hierarchy is used to calculate the combination of scenario solutions using the normalized plan benefit values. A comprehensive score for each valid alternative plan; From the set of valid alternative plans In the selection process, the effective alternative plan with the highest overall score will be chosen as the future plan. Monthly scenario solution combination The optimal plan is as follows.

9. The intelligent optimization method for adjusting monthly production plans based on market demand as described in claim 8, characterized in that, The combination of construction scenario solutions The AHP hierarchy is used to calculate the combination of scenario solutions using the normalized plan benefit values. The specific content of the comprehensive score for each valid alternative plan is as follows: Build a combination of scenario solutions The AHP hierarchy includes: target layer, criteria layer, and scheme layer; The target layer is: selecting a combination of scenario solutions. The optimal production plan under the given conditions; The criteria layer includes three dimensions: DEA overall efficiency. Resource utilization rate Matching degree with needs ; The solution layer is: a combination of scenario solutions. The following is a set of effective alternative plans; For scenario solution combinations Next One effective alternative plan Effective alternative plans Normalized planned benefit value As DEA overall efficiency index ; Based on the aforementioned production resources and constraints data, calculate effective alternative plans. Resource utilization rate indicators ; According to the effective alternative plans Demand forecasts and calculation of effective alternative plans. Demand matching index ; Construction of scene scheme combination The criterion layer judgment matrix is ​​calculated, and the global weight vector of the criterion layer is calculated. At the same time, the consistency check of the criterion layer judgment matrix is ​​performed. If the check fails, the criterion layer judgment matrix is ​​modified until the check succeeds. According to the effective alternative plans DEA overall efficiency index Resource utilization rate indicators Matching index with demand Using the global weight vector corresponding to the successfully validated criterion layer judgment matrix, effective alternative plans are calculated by weighted summation. The overall score.

10. A smart optimization system for adjusting monthly production plans based on market demand, used to implement the smart optimization method for adjusting monthly production plans based on market demand as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module is used to acquire market observation data, historical operational data of enterprises, and production resource and constraint data of the target bulk commodities; An observation sequence construction module is used to construct a two-dimensional observation vector sequence of the target commodity based on the market observation data. The HMM model training module is used to build and train the HMM demand prediction model to obtain a trained HMM demand prediction model. The demand state decoding module is used to input the two-dimensional observation vector sequence into the trained HMM demand prediction model and perform state decoding on the output of the model to obtain the demand state sequence and state transition matrix of the target commodity. The multi-scenario planning generation module is used to perform monthly production planning based on the enterprise's historical operational data, production resource and constraint data, demand state sequence and state transition matrix, and to obtain a set of effective alternative planning schemes for the target bulk commodity. The production planning decision module is used to filter the set of effective alternative plans based on the production resources and constraint data using a production utility evaluation model to obtain the optimal plan.

Images