A procurement decision method and system based on multi-time scale coal demand prediction

Abstract

This invention belongs to the field of coal procurement technology, and relates to a procurement decision-making method and system based on multi-timescale coal demand forecasting. It includes: acquiring power generation influencing factor data at at least two time scales, and generating power generation forecast values ​​corresponding to each time scale based on the power generation influencing factor data; fusing the power generation forecast values ​​at each time scale to obtain a fused power generation forecast value; decomposing the fused power generation forecast value into initial demand quantities for different coal types according to the consumption rate of each generating unit for different coal types, and adjusting the initial demand quantities according to a preset blending strategy to obtain the target demand quantity for each coal type; acquiring coal price time series data for each coal type, and generating future coal price forecast values ​​for each coal type based on the coal price time series data; and determining a procurement decision scheme based on the target demand quantity for each coal type and the coal price forecast value. This invention achieves accurate decision-making and cost control in fuel procurement.

Description

Technical Field

[0001] This invention belongs to the field of coal procurement technology, and relates to a procurement decision-making method and system based on multi-timescale coal demand forecasting. Background Technology

[0002] Fuel procurement is a significant component of operating costs for thermal power plants, typically accounting for 60% to 80% of total operating costs. Therefore, developing a reasonable fuel procurement strategy is crucial for cost control and ensuring supply security. Currently, fuel procurement for thermal power plants mainly relies on a planned procurement model based on annual contracts supplemented by monthly replenishments. In practice, procurement plans are usually formulated based on historical consumption data and empirical electricity forecasts. Demand forecasting often employs time series analysis or simple linear regression methods, rarely quantifying and integrating actual production factors such as unit start-up and shutdown schedules, equipment maintenance arrangements, and constraints on blending different coal types. At the procurement execution level, procurement contracts are often signed separately for each single coal type, with relatively fixed contract periods and a lack of flexible mechanisms for dynamic adjustments based on market price fluctuations. Furthermore, the control of coal quality indicators and boiler compatibility typically relies on post-procurement inspections and manual adjustments based on experience, failing to achieve systematic coordination between the procurement front-end and the blending scheme.

[0003] In summary, the aforementioned procurement technologies still have the following shortcomings in practical applications: Regarding demand forecasting, forecast accuracy is generally low. Existing methods are mostly based on simple linear extrapolation or empirical judgment, failing to fully consider the combined impact of various factors such as load changes, unit maintenance schedules, and coal quality fluctuations. Furthermore, forecasts are typically made only on a single time scale, such as monthly forecasts, failing to fully utilize the complementarity of information across different time scales, further limiting the improvement of forecast accuracy. In terms of procurement decision-making, existing strategies mostly focus on independent decisions for single coal types, lacking overall optimization from the perspective of multi-coal combinations. Moreover, the decision-making models are mostly static optimization models, assuming that all parameters are known and fixed, making it difficult to effectively cope with the uncertainties of actual demand and market prices. In addition, procurement plans are mostly formulated on a fixed cycle, making it difficult to adjust in a timely manner according to market price fluctuations, resulting in a lag in response. Regarding risk and multi-objective management, existing methods lack a systematic risk control mechanism, rarely considering the impact of factors such as supply disruption risks and drastic price fluctuations on procurement decisions. They also lack the ability to balance multiple objectives such as cost control, supply security, and coal quality matching, and lack a systematic multi-objective optimization framework. Summary of the Invention

[0004] To address the problems in the existing technology, this invention provides a procurement decision-making method and system based on multi-timescale coal demand forecasting, which enables accurate decision-making and cost control in fuel procurement.

[0005] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a procurement decision-making method based on multi-timescale coal demand forecasting, comprising the following steps: Obtain power generation influencing factor data at at least two time scales, and generate power generation prediction values ​​for each time scale based on the power generation influencing factor data; The power generation forecasts at different time scales are merged to obtain the merged power generation forecast. Based on the consumption rate of each generator unit for different types of coal, the predicted value of the combined power generation is decomposed into the initial demand for different types of coal, and the initial demand is adjusted according to the preset blending strategy to obtain the target demand for each type of coal. Obtain time series data of coal prices for each type of coal, and generate predicted coal prices for each type of coal at future points in time based on the coal price time series data; Based on the target demand for each type of coal and the predicted coal price, a procurement decision plan is determined.

[0006] Preferably, the step of acquiring power generation influencing factor data at at least two time scales, and generating power generation forecast values ​​corresponding to each time scale based on the power generation influencing factor data, includes: Obtain data on factors influencing power generation at an annual time scale, and generate an annual power generation forecast based on the data on factors influencing power generation at the annual time scale. Obtain data on factors influencing power generation at the quarterly time scale, and generate quarterly power generation forecasts based on the annual power generation forecast and the historical proportion coefficients and seasonal adjustment factors for each quarter in the data on factors influencing power generation at the quarterly time scale. Obtain data on factors influencing power generation at a monthly time scale, and generate monthly power generation forecasts based on the data on factors influencing power generation at the monthly time scale. Data on factors influencing power generation on a weekly time scale is obtained. Based on the rated capacity, available hours, and capacity factor of each generator unit in the data on factors influencing power generation on a weekly time scale, a predicted value for weekly power generation is generated.

[0007] Preferably, the step of fusing the power generation forecasts from different time scales to obtain the fused power generation forecast includes: Obtain the predicted values ​​and corresponding actual power generation values ​​for each time scale within the historical period, and calculate the prediction deviation index for each time scale based on the predicted values ​​and corresponding actual power generation values. The fusion weights of the predicted power generation values ​​at each time scale are determined based on the prediction deviation index. The power generation prediction values ​​for each time scale are weighted and summed according to the fusion weights to obtain the fused power generation prediction value.

[0008] Preferably, the step of decomposing the predicted fusion power generation value into initial demand for different coal types based on the consumption rate of each generator unit for different coal types includes: The planned power generation of each generator unit within the forecast period is determined based on the predicted value of the combined power generation and the power generation plan allocation ratio of each generator unit. Based on the planned power generation of each generator set and the consumption rate of each generator set for different types of coal, the unit demand for each type of coal on different generator sets is calculated, and the unit demand for the same type of coal on all generator sets is summed to obtain the initial demand for different types of coal.

[0009] Preferably, adjusting the initial demand based on a preset blending strategy to obtain the target demand for each coal type includes: Based on the preset blending strategy, the required adjustment amount for each of the various coal types participating in blending is determined; For the target coal type among the various coal types involved in blending, the initial demand and the demand adjustment corresponding to the target coal type are summed to obtain the target demand of the target coal type. The preset blending strategy includes multiple coal types participating in the blending and their preset blending ratios.

[0010] Preferably, the step of acquiring time-series coal price data for each coal type and generating predicted coal price values ​​for each coal type at future points in time based on the coal price time-series data includes: Acquire historical coal price time series data and external influencing factor data for each type of coal, including macroeconomic indicators, policy change indicators, and supply and demand relationship indicators. Based on the historical coal price time series data and the external influencing factor data, extract the time-series dependency features in the historical coal price time series data and the nonlinear mapping relationship between the external influencing factor data and coal prices; Based on the time-dependent features and the nonlinear mapping relationship, the predicted coal prices for each coal type at future points in time are generated.

[0011] Preferably, determining the procurement decision plan based on the target demand for each type of coal and the predicted coal price includes: Obtain the demand uncertainty parameters and price uncertainty parameters for each type of coal, and construct multiple scenarios based on the demand uncertainty parameters and price uncertainty parameters. Each scenario corresponds to a set of fluctuations in demand and price, and each scenario has a corresponding probability of occurrence. Based on the target demand for each type of coal, the predicted coal price, and the multiple scenarios, determine the procurement cost, inventory holding cost, stockout cost, and coal quality deviation cost under each scenario. With the objective of minimizing the weighted sum of procurement costs, inventory holding costs, stockout costs, and coal quality deviation costs under each scenario, and constrained by inventory capacity, procurement funds, and the supply capacity of each type of coal, a procurement decision scheme for the current decision-making cycle is determined.

[0012] Preferably, after determining the procurement decision plan for the current decision-making cycle, the method further includes: Obtain actual power generation data, actual coal price data, and actual inventory data for each type of coal at the end of the current decision-making cycle; Based on the actual power generation data, the actual coal price data, and the actual inventory data of each coal type, update the power generation influencing factor data, the coal price time series data of each coal type, and the inventory capacity constraints; The updated data on factors affecting power generation are used to regenerate the power generation forecast values ​​for each time scale. The regenerated power generation forecast values ​​for each time scale are then merged and decomposed into the target demand for each type of coal. The updated coal price time series data for each coal type are used to regenerate the predicted coal price values ​​for future points in time. Based on the regenerated target demand for each type of coal and the regenerated coal price forecast for each type of coal, the procurement decision plan for the next decision cycle is redefined.

[0013] Secondly, the present invention provides a procurement decision-making system based on multi-timescale coal demand forecasting, comprising: Multi-scale power generation prediction module: used to acquire power generation influencing factor data at at least two time scales, and generate power generation prediction values ​​corresponding to each time scale based on the power generation influencing factor data; Prediction fusion module: used to fuse the power generation predictions at various time scales to obtain the fused power generation prediction. Coal type demand decomposition module: It is used to decompose the predicted value of the integrated power generation into the initial demand of different coal types according to the consumption rate of each generator unit for different coal types, and adjust the initial demand according to the preset blending strategy to obtain the target demand of each coal type. Coal price time series forecast module: used to acquire coal price time series data for each type of coal, and generate coal price forecast values ​​for each type of coal at future points in time based on the coal price time series data; Procurement decision optimization module: used to determine procurement decision schemes based on the target demand of each type of coal and the predicted coal price.

[0014] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a procurement decision-making method based on multi-timescale coal demand forecasting.

[0015] Compared with the prior art, the present invention has the following beneficial effects: By acquiring power generation influencing factor data at at least two time scales and generating power generation forecasts at each scale, multi-dimensional information extraction from macro trends to micro fluctuations is achieved, effectively overcoming the information gap problem of single-scale forecasts. The fusion of power generation forecasts at each time scale yields a fused power generation forecast, comprehensively utilizing the complementary advantages of different scale forecasts and significantly improving the accuracy and stability of power generation forecasts. Based on the consumption rate of different coal types by each generating unit, the fused power generation forecast is decomposed into initial demand for different coal types, and adjusted according to a preset blending strategy to obtain the target demand for each coal type. This achieves a precise mapping from total power generation to specific coal type demand, ensuring the adaptability of unit operation to coal quality requirements while expanding the substitution space for low-cost coal types through blending optimization. Acquiring time series data of coal prices for each coal type and generating future coal price forecasts provides crucial forward-looking price information for procurement decisions. Determining procurement decision schemes based on the target demand and coal price forecasts for each coal type enables a systematic balance between procurement costs, supply security, and coal quality adaptability under the dual constraints of demand and price. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0019] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0020] The first objective of this invention is to provide a procurement decision-making method based on multi-timescale coal demand forecasting, such as... Figure 1 As shown, it includes the following steps: S1. Obtain power generation influencing factor data at at least two time scales, and generate power generation prediction values ​​corresponding to each time scale based on the power generation influencing factor data.

[0021] Obtain data on factors influencing power generation at the annual time scale, and generate an annual power generation forecast based on this data. The annual time scale factors influencing power generation include GDP growth rate, industrial electricity consumption growth rate, population growth rate, changes in power supply structure, and the impact of energy policies. The expression for the annual power generation forecast is as follows:

[0022] in, This is the projected annual power generation value; This refers to the growth rate of domestic GDP or the growth rate of regional GDP. This refers to the growth rate of industrial electricity consumption. Population growth rate; This is due to changes in the power supply structure; Due to the influence of energy policy.

[0023] For example, using a combination of multiple regression and machine learning to predict annual power generation, the formula for calculating the predicted value is as follows:

[0024] in, For explanatory variables, i.e. , , , , And so on. Among them, a relatively high GDP growth rate By driving the expansion of output in the secondary and tertiary industries, it directly increases the demand for electricity in industry and services, which is positively correlated with the annual power generation of thermal power plants; the growth rate of industrial electricity consumption... As a coincident indicator of the secondary industry's prosperity, it reflects the capacity utilization rate of energy-intensive industries and has a positive correlation with the annual power generation of thermal power plants. Population growth rate This reflects the growth rate of residential electricity consumption. Assuming per capita electricity consumption remains constant, there is a positive correlation between population growth rate and annual power generation from thermal power plants. Changes in power supply structure... The impact manifests as a structural power loss due to a decrease in weighted average utilization hours. When high-utilization-hour power sources (thermal power plants) are replaced by low-utilization-hour power sources (wind farms, photovoltaic power plants), even if the total installed capacity increases, actual power generation may stagnate or even decline. Energy Policy The impact is transmitted through two channels. On the one hand, mandatory supply-side policies (such as renewable energy quotas and coal-fired power capacity elimination) directly change the power structure and affect the power generation of thermal power plants. On the other hand, price mechanisms (capacity pricing, time-of-use pricing, etc.) indirectly affect the utilization hours of thermal power units by changing the priority of unit dispatch and investment decisions. These are the regression coefficients corresponding to the explanatory variables; This is the random error term; For correction terms in machine learning models; The intercept term of the multiple linear regression model represents the power generation level of the thermal power plant base.

[0025] Based on the annual power generation forecast, quarterly power generation influencing factor data are obtained. Based on the annual power generation forecast and the historical proportion coefficients and seasonal adjustment factors for each quarter in the quarterly power generation influencing factor data, a quarterly power generation forecast is generated. The formula for calculating the quarterly power generation forecast is as follows:

[0026] in, For the first Quarterly power generation forecast; For the first Historical percentage coefficient for each quarter; This is a seasonal adjustment factor.

[0027] Data on factors influencing power generation at a monthly time scale is obtained, and monthly power generation forecasts are generated based on this data. Specifically, monthly power generation can be decomposed into trend, seasonal, cyclical, and irregular components. A Seasonal Autoregressive Integrated Moving Average (SARIMA) model is used to forecast monthly power generation. The formula for calculating the monthly power generation forecast is as follows:

[0028] in, For the first Monthly power generation forecast; It is a lag operator; This is the non-seasonal difference order, usually taken as 0, 1, or 2 to eliminate trends; This is the seasonality difference order, usually taken as 0 or 1; and These are the non-seasonal moving average coefficient and the seasonal moving average coefficient, respectively. It is white noise.

[0029] Data on power generation influencing factors at a weekly time scale is obtained. Based on the rated capacity, available hours, and capacity factor of each generator unit in the weekly time scale power generation influencing factor data, a weekly power generation forecast value is generated. The calculation formula for the weekly power generation forecast value is as follows:

[0030] in, For the first Weekly power generation forecast; For the first Rated capacity of the unit; For the first Available hours of the unit; For the first Capacity factor of the unit.

[0031] This step achieves multi-dimensional collaborative forecasting from macro trends to micro operations by forecasting power generation at four time scales: annual, quarterly, monthly, and weekly. At the annual scale, it integrates macroeconomic indicators and machine learning correction terms to grasp the long-term trend of electricity demand. At the quarterly scale, it uses historical proportional coefficients and seasonal adjustment factors to perform seasonal decomposition of annual forecasts and capture cyclical fluctuation patterns. At the monthly scale, it uses the SARIMA model to refine the decomposition of trend, seasonal, cyclical, and irregular terms to improve the stability of short- and medium-term forecasts. At the weekly scale, it forecasts based on unit rated capacity, available hours, and capacity factors to accurately reflect real-time fluctuations caused by unit maintenance and load changes.

[0032] S2. Merge the power generation forecasts at each time scale to obtain the merged power generation forecast.

[0033] Obtain the predicted values ​​and corresponding actual power generation values ​​for each time scale within the historical period, and calculate the prediction deviation index for each time scale based on the predicted values ​​and corresponding actual power generation values. The fusion weights of the predicted power generation values ​​at each time scale are determined based on the prediction deviation index. The power generation prediction values ​​for each time scale are weighted and summed according to the fusion weights to obtain the fused power generation prediction value.

[0034] Specifically, the predicted value sequence and corresponding actual power generation value for each time scale within the historical rolling window are obtained, and the root mean square error of the annual, quarterly, monthly, and weekly predicted values ​​are calculated respectively. As a prediction bias index at various scales, this index quantifies the prediction accuracy of each predicted value over a historical period. The smaller the value, the more accurate the prediction at that scale; The fusion weights for each time scale are determined based on the prediction bias index, and the weight coefficients are... Adaptive adjustment is achieved using the inverse error weighting method, i.e. ,in For the first The root mean square error of the predicted values ​​at each time scale results in a larger fusion weight for scales with higher historical prediction accuracy. Subsequently, the power generation prediction values ​​for each time scale are weighted and summed according to the fusion weights to obtain the fused power generation prediction value, which is expressed as follows:

[0035]

[0036] in, This represents the predicted value of the combined power generation. The fusion weights for the annual power generation forecast; The fusion weight of quarterly power generation forecasts; This is a quarterly power generation forecast. The fusion weights for monthly power generation forecasts; This is the monthly power generation forecast. The fusion weights for weekly power generation forecasts; This is the weekly power generation forecast.

[0037] This fusion mechanism fully leverages the complementarity of forecasting methods across different time scales: annual forecasting methods capture macroeconomic trends, quarterly forecasting methods capture cyclical seasonal fluctuations, monthly forecasting methods reflect short- to medium-term inertial changes, and weekly forecasting methods respond to unit maintenance and real-time load adjustments. Simultaneously, through dynamic weighting based on historical forecast accuracy, the fusion results achieve an optimal balance between guiding long-term trends and correcting short-term deviations. Compared to single-scale forecasting or fixed-weight fusion, this method automatically adjusts weight allocation according to the time-varying characteristics of forecast performance at each scale, effectively reducing overall forecast bias and improving the accuracy and stability of power generation forecasts. This provides more reliable input data for subsequent coal demand decomposition, thereby ensuring the scientific and accurate nature of procurement decisions from the outset.

[0038] S3. Based on the consumption rate of each generator unit for different types of coal, the predicted value of the combined power generation is decomposed into the initial demand for different types of coal, and the initial demand is adjusted according to the preset blending strategy to obtain the target demand for each type of coal.

[0039] The planned power generation of each generator unit within the forecast period is determined based on the predicted value of the integrated power generation and the power generation plan allocation ratio of each generator unit. The power generation plan allocation ratio can be set according to the economic ranking of the units, grid dispatch instructions, or the principle of balancing the utilization hours of the units to ensure the reasonable allocation of power generation of each unit. Based on the planned power generation of each generator unit and the consumption rate of each generator unit for different coal types (this consumption rate is pre-calibrated according to the boiler design parameters, historical operating data, and coal quality compatibility characteristics, reflecting the unit power generation coal consumption level of different units when burning specific coal types), the unit demand for each coal type on different generator units is calculated, and the unit demand for the same coal type on all generator units is summed to obtain the initial demand for each coal type; the formula for calculating the initial demand is:

[0040] in, For the first The initial demand for coal; For the first Power generation of the unit; For the first Taiwan unit to the first The consumption rate of different types of coal. This decomposition method fully considers the different adaptability requirements of different units for indicators such as calorific value, volatile matter, sulfur content, and ash content of different coal types, and realizes a structured mapping from total power generation to the demand for multiple coal types.

[0041] The preset blending strategy includes multiple coal types participating in blending and their preset blending ratios. It is typically determined based on multi-objective optimization, such as minimizing fuel costs, environmental emission constraints, and boiler combustion stability. Specifically, the impact of each blending scheme on the demand for each coal type is calculated based on the fuel blending optimization results, thereby obtaining the adjusted target demand for each coal type. The formula for calculating the target demand is:

[0042] in, For the adjusted number The target demand for coal; For the first The first type of co-firing scheme is for the first The impact of coal demand.

[0043] This step organically combines the characteristics of the power generation equipment with the blending optimization on the fuel side. It avoids the problems of limited unit output or reduced combustion efficiency caused by coal quality mismatch, and expands the range of procurable coal types through the blending strategy. This enhances the flexibility and bargaining power of fuel procurement, thereby significantly reducing fuel procurement costs while ensuring power generation safety, and providing accurate coal type demand input that fits the actual operating needs for subsequent procurement decisions.

[0044] S4. Obtain the coal price time series data for each type of coal, and generate the predicted coal price values ​​for each type of coal at future points in time based on the coal price time series data.

[0045] Acquire historical coal price time series data and external influencing factor data for each type of coal. The external influencing factor data includes macroeconomic indicators such as GDP growth rate, industrial added value growth rate, and purchasing managers index; policy change indicators such as coal capacity reduction policies, import tariff adjustments, and environmental protection production restriction policies; and supply and demand relationship indicators such as coal port inventory, key power plant inventory, coal freight rate index, and downstream industry operating rate. Based on the historical coal price time series data and the external influencing factor data, extract the time-series dependency features in the historical coal price time series data and the nonlinear mapping relationship between the external influencing factor data and coal prices; Based on the time-dependent features and the nonlinear mapping relationship, the predicted coal prices for each coal type at future points in time are generated.

[0046] For example, a long short-term memory neural network is used to construct a coal price time series prediction model. This network effectively captures the long-term time-series dependence features in historical coal price series through gating mechanisms of input gate, forget gate, and output gate. At the same time, external influencing factors are embedded as parallel inputs into the network structure to extract the nonlinear mapping relationship between external factors and coal prices. The expression of this mapping relationship is as follows:

[0047] in, for Coal prices at any given time; The time window length of the historical coal price series is used to capture the time-series dependence and volatility inertia of coal prices; This represents a vector of external influencing factors (such as macroeconomic factors, policy changes, supply and demand). The time window length for external influencing factors is used to characterize the lag effect of external factors on coal prices; by setting... and Different values ​​can flexibly adapt to the differences between the fluctuation cycle of coal prices themselves and the transmission time of external shocks. The input layer will Historical coal price data and A multidimensional feature matrix is ​​formed by concatenating historical external factor data. The LSTM network selectively discards non-critical historical information through the forget gate, extracts current effective features through the input gate, and generates prediction results through the output gate. This allows the network to simultaneously learn the long-term dependence of coal price series and the complex nonlinear mapping relationship between coal prices and multiple sources of factors such as macroeconomics, policies, supply and demand.

[0048] This method leverages the long-term memory capability of LSTM neural networks for time series data and their strong fitting ability for nonlinear relationships. It overcomes the limitations of traditional time series models in handling multivariate nonlinear interactions, effectively capturing the dynamic response characteristics of coal prices during changes in supply and demand patterns and the transmission of policy shocks. This enables accurate prediction of coal prices for various coal types at future points in time, providing crucial price inputs for optimizing subsequent procurement decisions. It allows for forward-looking prediction of market trends in procurement timing and quantity decisions, effectively mitigating price volatility risks and seizing low-price procurement opportunities.

[0049] Furthermore, the prediction model used in this invention has several alternatives. Specifically: the Prophet model can be used instead of the SARIMA model for monthly power generation prediction. This model is based on a decomposable time series structure, effectively handles seasonality and holiday effects, and has good robustness and automated parameter tuning capabilities. The N-BEATS (Neural Basis Expansion Analysis for Time Series) model can be used instead of LSTM for coal price prediction. This model is based on a pure neural network architecture and achieves hierarchical decomposition of the time series by stacking multiple basic blocks, achieving excellent prediction accuracy without introducing external features. Alternatively, the Transformer architecture can be used instead of LSTM, utilizing its multi-head attention mechanism to capture long-range dependencies in the time series, making it particularly suitable for coal price prediction in multivariate input scenarios. The above alternative models can be flexibly selected according to the data characteristics, computing resources, and prediction accuracy requirements of the actual application scenario, and all can achieve similar technical effects to the method disclosed in this invention.

[0050] S5. Based on the target demand for each type of coal and the predicted coal price, determine the procurement decision plan.

[0051] Obtain the demand uncertainty parameters and price uncertainty parameters for each type of coal, and construct multiple scenarios based on the demand uncertainty parameters and price uncertainty parameters. Each scenario corresponds to a set of fluctuations in demand and price, and each scenario has a corresponding probability of occurrence. Specifically, based on the target demand and predicted coal prices for each type of coal, the uncertainty parameters of demand and price for each type of coal are first obtained. To facilitate the solution, scenario analysis is used to transform the uncertainty parameters into finite discrete scenarios. This set is discretized into several typical scenarios, each corresponding to a combination of demand and price fluctuations. Simultaneously, a corresponding probability of occurrence is assigned to each scenario based on its likelihood of occurrence. The problem is then transformed into a deterministic optimization problem with the objective of minimizing the expected cost of each scenario, resulting in the objective function:

[0052] in, For the first The probability of each scenario; For the first Variables for purchasing decisions under various scenarios; For the first Costs under different scenarios.

[0053] Based on the target demand for each type of coal, the predicted coal price, and the multiple scenarios, determine the procurement cost, inventory holding cost, stockout cost, and coal quality deviation cost under each scenario. The objective function is to minimize the weighted sum of procurement costs, inventory holding costs, stockout costs, and coal quality deviation costs under each of the aforementioned scenarios.

[0054] in, For the first Coal planting in the first Price at a specific point in time of purchase; For the first The amount of coal to be purchased; Inventory holding costs; Costs related to stockouts; Cost of coal quality deviation.

[0055] Based on constraints such as inventory capacity, procurement funds, and the supply capacity of each type of coal, determine the procurement decision-making scheme for the current decision-making cycle.

[0056] Among them, the demand constraints are: ; Inventory constraints: , For the first Coal inventory; Funding constraints: , This is the upper limit of the procurement budget; Supply capacity constraints: , For the first Coal planting in the first Maximum supply capacity at each procurement point in time.

[0057] Furthermore, based on the target demand and coal price forecasts for each type of coal, this invention can first obtain the demand uncertainty parameters and price uncertainty parameters for each type of coal, and then construct an uncertainty set based on historical forecast deviation distribution, market fluctuation characteristics, or expert experience. A robust procurement optimization model considering prediction bias is established, and its expression is:

[0058] in, For procurement decision variables (such as the procurement quantity of each type of coal); The parameters are uncertain (i.e., the actual demand and market price of each type of coal). The model achieves robustness against prediction bias by solving for the maximum cost under the worst-case scenario in the inner layer and finding the decision scheme that minimizes the maximum cost in the outer layer.

[0059] This method extends traditional single-prediction-driven procurement decisions into a robust optimization mechanism that covers multiple scenarios and weighs costs across multiple dimensions. It effectively avoids supply disruption risks through inventory constraints and stockout cost penalties, ensures the compatibility of procured coal quality with boiler operating requirements through coal quality deviation cost constraints, and guarantees the financial feasibility and supply chain executability of the decision-making scheme through budget constraints and supply capacity constraints. Thus, it achieves comprehensive optimization of procurement costs, inventory turnover efficiency, and risk control level while ensuring the safe and stable operation of generator units, significantly improving the scientific, robust, and economical nature of fuel procurement decisions in complex and volatile market environments.

[0060] Furthermore, the optimization method employed in this invention has several alternatives. Specifically, stochastic programming can be used to replace scenario analysis to handle the uncertainty of demand and price. This method is an uncertainty optimization method based on probability theory. By constructing a decision tree or a multi-stage stochastic programming model, the probability distribution of uncertain parameters is discretized into multiple stochastic scenarios, and the optimal procurement strategy is solved under the expected value objective. Alternatively, fuzzy programming can be used to replace scenario analysis. This method is based on fuzzy set theory, which represents the uncertainty of demand and price as fuzzy numbers or fuzzy intervals. By setting a membership function to describe the fluctuation range of uncertain parameters, fuzzy chance-constrained programming or fuzzy multi-objective programming is used for solving the problem. This method is suitable for scenarios where historical data is insufficient or it is difficult to accurately characterize the probability distribution.

[0061] S6. Achieve dynamic closed-loop adjustment and continuous iterative optimization of procurement decisions by constructing a rolling optimization framework.

[0062] Obtain actual power generation data, actual coal price data, and actual inventory data for each type of coal at the end of the current decision-making cycle; Based on the actual power generation data, the actual coal price data, and the actual inventory data of each coal type, update the power generation influencing factor data, the coal price time series data of each coal type, and the inventory capacity constraints; The updated data on factors affecting power generation are used to regenerate the power generation forecast values ​​for each time scale. The regenerated power generation forecast values ​​for each time scale are then merged and decomposed into the target demand for each type of coal. The updated coal price time series data for each coal type are used to regenerate the predicted coal price values ​​for future points in time. Based on the regenerated target demand for each type of coal and the regenerated coal price forecast for each type of coal, the procurement decision plan for the next decision cycle is redefined.

[0063] Specifically, after the current decision-making cycle ends, the system acquires actual power generation data, actual coal price data, and actual inventory data for each type of coal within that cycle. Based on this feedback information, the system updates the model parameters and constraints: It uses actual power generation data to correct prediction biases in the forecasting models at each time scale, recalculates the root mean square error of the annual, quarterly, monthly, and weekly models, and adaptively adjusts the fusion weights accordingly. Simultaneously, it adds actual data as new samples to the training sets of the SARIMA model and LSTM neural network, updating model parameters through incremental learning or periodic full retraining to ensure the forecasting model continuously adapts to current load changes and market fluctuations. It also uses actual coal price data to update the input sequence of the LSTM model and adjusts the statistics of the external influencing factor vector. This system enhances the timeliness of coal price forecasts. It adjusts the initial inventory levels in the inventory capacity constraint based on actual inventory data and dynamically adjusts the upper limit of the procurement budget for the next decision-making cycle based on current actual procurement expenditures and fund usage. On this basis, it regenerates the predicted power generation values ​​for each time scale using updated data on factors influencing power generation. After adaptive fusion, it obtains the target demand for each coal type using the fusion and decomposition methods described in S2 and S3. Simultaneously, it regenerates the predicted coal prices for each coal type at future points in time using updated coal price time series data and the LSTM prediction method described in S4. Then, based on the regenerated target demand and predicted coal prices, it again uses scenario analysis or robust optimization models described in S5 to solve for the procurement decision scheme for the next decision-making cycle. This method transforms static, open-loop procurement planning into a dynamic, adaptive optimization process. Through continuous feedback of actual data, it constantly corrects the deviations of the prediction model and updates the constraints, enabling procurement decisions to respond promptly to changes in load, market fluctuations, and inventory status. It effectively overcomes the shortcomings of traditional fixed-cycle procurement models in responding to changes in a lagging manner. In long-term operation, it has achieved continuous improvement in prediction accuracy, gradual optimization of procurement costs, and effective convergence of supply risks, significantly enhancing the adaptability and overall benefits of fuel procurement management in complex dynamic environments.

[0064] Furthermore, precise triggering and delayed purchasing decisions can be achieved by jointly judging price changes and inventory status. The judgment rules are as follows: if and This will trigger a purchase; if and Then the procurement will be delayed; in, The price change threshold; This is the inventory threshold; For safety stock.

[0065] The second objective of this invention is to provide a procurement decision-making system based on multi-timescale coal demand forecasting, comprising: Multi-scale power generation prediction module: used to acquire power generation influencing factor data at at least two time scales, and generate power generation prediction values ​​corresponding to each time scale based on the power generation influencing factor data; Prediction fusion module: used to fuse the power generation predictions at various time scales to obtain the fused power generation prediction. Coal type demand decomposition module: It is used to decompose the predicted value of the integrated power generation into the initial demand of different coal types according to the consumption rate of each generator unit for different coal types, and adjust the initial demand according to the preset blending strategy to obtain the target demand of each coal type. Coal price time series forecast module: used to acquire coal price time series data for each type of coal, and generate coal price forecast values ​​for each type of coal at future points in time based on the coal price time series data; Procurement decision optimization module: used to determine procurement decision schemes based on the target demand of each type of coal and the predicted coal price.

[0066] This system integrates multi-timescale collaborative forecasting and multi-coal type demand decomposition, realizing full-process automation and intelligence from forecasting to decision-making. It significantly improves the scientific nature, accuracy, and risk resistance of fuel procurement decisions, providing systematic technical support for thermal power companies to effectively reduce fuel procurement costs, optimize inventory structure, and ensure supply security.

[0067] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a procurement decision-making method based on multi-timescale coal demand forecasting.

[0068] This invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the procurement decision-making method based on multi-timescale coal demand forecasting in the above embodiments.

[0069] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0070] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0071] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0072] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A procurement decision-making method based on multi-timescale coal demand forecasting, characterized in that, Includes the following steps: Obtain power generation influencing factor data at at least two time scales, and generate power generation prediction values ​​for each time scale based on the power generation influencing factor data; The power generation forecasts at different time scales are merged to obtain the merged power generation forecast. Based on the consumption rate of each generator unit for different types of coal, the predicted value of the combined power generation is decomposed into the initial demand for different types of coal, and the initial demand is adjusted according to the preset blending strategy to obtain the target demand for each type of coal. Obtain time series data of coal prices for each type of coal, and generate predicted coal prices for each type of coal at future points in time based on the coal price time series data; Based on the target demand for each type of coal and the predicted coal price, a procurement decision plan is determined.

2. The procurement decision-making method based on multi-timescale coal demand forecasting according to claim 1, characterized in that, The step of acquiring power generation influencing factor data at at least two time scales, and generating power generation forecast values ​​corresponding to each time scale based on the power generation influencing factor data, includes: Obtain data on factors influencing power generation at an annual time scale, and generate an annual power generation forecast based on the data on factors influencing power generation at the annual time scale. Obtain data on factors influencing power generation at the quarterly time scale, and generate quarterly power generation forecasts based on the annual power generation forecast and the historical proportion coefficients and seasonal adjustment factors for each quarter in the data on factors influencing power generation at the quarterly time scale. Obtain data on factors influencing power generation at a monthly time scale, and generate monthly power generation forecasts based on the data on factors influencing power generation at the monthly time scale. Data on factors influencing power generation on a weekly time scale is obtained. Based on the rated capacity, available hours, and capacity factor of each generator unit in the data on factors influencing power generation on a weekly time scale, a predicted value for weekly power generation is generated.

3. The procurement decision-making method based on multi-timescale coal demand forecasting according to claim 1, characterized in that, The process of fusing the power generation forecasts from various time scales to obtain the fused power generation forecast includes: Obtain the predicted values ​​and corresponding actual power generation values ​​for each time scale within the historical period, and calculate the prediction deviation index for each time scale based on the predicted values ​​and corresponding actual power generation values. The fusion weights of the predicted power generation values ​​at each time scale are determined based on the prediction deviation index. The power generation prediction values ​​for each time scale are weighted and summed according to the fusion weights to obtain the fused power generation prediction value.

4. The procurement decision-making method based on multi-timescale coal demand forecasting according to claim 1, characterized in that, The step of decomposing the predicted fusion power generation into initial demand for different coal types based on the consumption rate of each generator unit for different coal types includes: The planned power generation of each generator unit within the forecast period is determined based on the predicted value of the combined power generation and the power generation plan allocation ratio of each generator unit. Based on the planned power generation of each generator set and the consumption rate of each generator set for different types of coal, the unit demand for each type of coal on different generator sets is calculated, and the unit demand for the same type of coal on all generator sets is summed to obtain the initial demand for different types of coal.

5. The procurement decision-making method based on multi-timescale coal demand forecasting according to claim 1, characterized in that, The step of adjusting the initial demand based on a preset blending strategy to obtain the target demand for each type of coal includes: Based on the preset blending strategy, the required adjustment amount for each of the various coal types participating in blending is determined; For the target coal type among the various coal types involved in blending, the initial demand and the demand adjustment corresponding to the target coal type are summed to obtain the target demand of the target coal type. The preset blending strategy includes multiple coal types participating in the blending and their preset blending ratios.

6. The procurement decision-making method based on multi-timescale coal demand forecasting according to claim 1, characterized in that, The step of acquiring time-series coal price data for each type of coal and generating future price forecasts for each type of coal based on the time-series coal price data includes: Acquire historical coal price time series data and external influencing factor data for each type of coal, including macroeconomic indicators, policy change indicators, and supply and demand relationship indicators. Based on the historical coal price time series data and the external influencing factor data, extract the time-series dependency features in the historical coal price time series data and the nonlinear mapping relationship between the external influencing factor data and coal prices; Based on the time-dependent features and the nonlinear mapping relationship, the predicted coal prices for each coal type at future points in time are generated.

7. The procurement decision-making method based on multi-timescale coal demand forecasting according to claim 1, characterized in that, The process of determining a procurement decision based on the target demand for each type of coal and the predicted coal price includes: Obtain the demand uncertainty parameters and price uncertainty parameters for each type of coal, and construct multiple scenarios based on the demand uncertainty parameters and price uncertainty parameters. Each scenario corresponds to a set of fluctuations in demand and price, and each scenario has a corresponding probability of occurrence. Based on the target demand for each type of coal, the predicted coal price, and the multiple scenarios, determine the procurement cost, inventory holding cost, stockout cost, and coal quality deviation cost under each scenario. With the objective of minimizing the weighted sum of procurement costs, inventory holding costs, stockout costs, and coal quality deviation costs under each scenario, and constrained by inventory capacity, procurement funds, and the supply capacity of each type of coal, a procurement decision scheme for the current decision-making cycle is determined.

8. A procurement decision-making method based on multi-timescale coal demand forecasting according to claim 7, characterized in that, After determining the procurement decision plan for the current decision-making cycle, the following is also included: Obtain actual power generation data, actual coal price data, and actual inventory data for each type of coal at the end of the current decision-making cycle; Based on the actual power generation data, the actual coal price data, and the actual inventory data of each coal type, update the power generation influencing factor data, the coal price time series data of each coal type, and the inventory capacity constraints; The updated data on factors affecting power generation are used to regenerate the power generation forecast values ​​for each time scale. The regenerated power generation forecast values ​​for each time scale are then merged and decomposed into the target demand for each type of coal. The updated coal price time series data for each coal type are used to regenerate the predicted coal price values ​​for future points in time. Based on the regenerated target demand for each type of coal and the regenerated coal price forecast for each type of coal, the procurement decision plan for the next decision cycle is redefined.

9. A procurement decision-making system based on multi-timescale coal demand forecasting, characterized in that, include: Multi-scale power generation prediction module: used to acquire power generation influencing factor data at at least two time scales, and generate power generation prediction values ​​corresponding to each time scale based on the power generation influencing factor data; Prediction fusion module: used to fuse the power generation predictions at various time scales to obtain the fused power generation prediction. Coal type demand decomposition module: It is used to decompose the predicted value of the integrated power generation into the initial demand of different coal types according to the consumption rate of each generator unit for different coal types, and adjust the initial demand according to the preset blending strategy to obtain the target demand of each coal type. Coal price time series forecast module: used to acquire coal price time series data for each type of coal, and generate coal price forecast values ​​for each type of coal at future points in time based on the coal price time series data; Procurement decision optimization module: used to determine procurement decision schemes based on the target demand of each type of coal and the predicted coal price.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-8.

Images