Productivity prediction method and device, storage medium and equipment

Abstract

The invention discloses a productivity prediction method and device, a storage medium and equipment. The method comprises the following steps: obtaining product demand data of a current station, WPH data of each product at the current station, and historical WIP data of each product at the current station; based on the product demand data, the WPH data of each product at the current station and historical WIP data of each product at the current station, performing first prediction on future WIP data of the current station to obtain a first prediction result of the future WIP data of the current station; and based on the product demand data and the first prediction result, utilizing a preset prediction model to obtain a second prediction result of future WIP data of the current station, so as to predict the future productivity of the current station based on the second prediction result. By adopting the scheme, the accuracy of productivity prediction can be improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, and more specifically to a capacity forecasting method, apparatus, storage medium, and device. Background Technology

[0002] The semiconductor industry is characterized by long factory construction cycles and complex processes. In order to enhance its ability to withstand risks, capacity planning is carried out based on market forecasts at that time to formulate planned capacity for each platform.

[0003] When market demand is high, orders can be selected based on factors such as order volume, order continuity, and customer stability, balancing the capacity of various product platforms to maximize efficiency. When the market is sluggish, it is necessary to maximize the production of products in demand and ensure on-time or early delivery to achieve higher customer satisfaction and enhance competitiveness.

[0004] In both of these scenarios, the prerequisite for accepting a sales order is that the factory has the ability to produce on schedule. How to assess whether the factory has the ability to meet the order demand, i.e., the prediction of capacity parameters, is extremely important in the semiconductor industry and even the entire manufacturing industry.

[0005] However, current capacity forecasting methods are not very accurate in predicting capacity. Summary of the Invention

[0006] The problem this invention aims to solve is: how to improve the accuracy of production capacity forecasting.

[0007] To address the above problems, embodiments of the present invention provide a capacity forecasting method, the method comprising:

[0008] Obtain product demand data for the current site, WPH data for each product on the current site, and historical WIP data for each product on the current site;

[0009] Based on the product demand data, the WPH data of each product at the current site, and the historical WIP data of each product at the current site, a first prediction is made on the future WIP data of the current site to obtain the first prediction result of the future WIP data of the current site.

[0010] Based on the product demand data and the first prediction result, a second prediction result for the future WIP data of the current site is obtained using a preset prediction model, and the future production capacity of the current site is predicted based on the second prediction result.

[0011] In one possible embodiment, the product demand data includes: the identifiers of each product to be produced at the current site and the demand quantity of each product to be produced.

[0012] In one possible embodiment, the step of making a first prediction of the future WIP data of the current site based on the product demand data, the WPH data of each product at the current site, and the historical WIP data of each product at the current site, to obtain a first prediction result of the future WIP data of the current site, includes:

[0013] Based on the product demand data and the WPH data of each product at the current site, the historical WIP data of each product at the current site is reconstructed using a sliding window to obtain the first prediction result of the future WIP data of the current site.

[0014] In one possible embodiment, the step of reconstructing the historical WIP data of each product at the current site using a sliding window based on the product demand data and the WPH data of each product at the current site, to obtain a first prediction result of the future WIP data of the current site, includes:

[0015] Based on the WPH data of each product at the current site, the product demand data, and the average WPH data of each product at the current site, the predicted time step of each product at the current site is obtained.

[0016] Using the predicted time step and a preset sliding window size, the historical WIP data of each product at the current site is reconstructed to obtain the first prediction result of the future WIP data of the current site.

[0017] In one possible embodiment, before reconstructing the historical WIP data of each product at the current site using a sliding window based on the product demand data and the WPH data of each product at the current site, the method further includes:

[0018] The historical WIP data of each product at the current site is preprocessed.

[0019] In one possible embodiment, the preprocessing of the historical WIP data of each product at the current site includes:

[0020] The historical WIP data of each product at the current site is supplemented using spline interpolation.

[0021] The Savitzky-Golay method is used to smooth the supplemented data, and then a sliding window is used to reconstruct the smoothed data.

[0022] In one possible embodiment, the step of obtaining a second prediction result of the future WIP data of the current site based on the product demand data and the first prediction result using a preset prediction model includes:

[0023] The future WIP data of each product in the first prediction result is combined with the product demand data, and each data combination is input into the preset XGBoost model to obtain the second prediction result of the future WIP data of each product.

[0024] The final capacity parameter values ​​are obtained based on the second forecast results of the future WIP data for each product.

[0025] This invention also provides a capacity forecasting device, the device comprising:

[0026] The acquisition unit is suitable for acquiring product demand data for the current site, WPH data for each product at the current site, and historical WIP data for each product at the current site.

[0027] The first prediction unit is adapted to make a first prediction of the future WIP data of the current site based on the product demand data, the WPH data of each product at the current site and the historical WIP data of each product at the current site, and to obtain the first prediction result of the future WIP data of the current site.

[0028] The second prediction unit is adapted to obtain a second prediction result of the future WIP data of the current site based on the product demand data and the first prediction result, using a preset prediction model, so as to predict the future production capacity of the current site based on the second prediction result.

[0029] This invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of any of the methods described above.

[0030] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the steps of any of the methods described above when running the computer program.

[0031] Compared with the prior art, the technical solution of the embodiments of the present invention has the following advantages:

[0032] The solution of this invention, based on the product demand data of the current site, the WPH data of each product at the current site, and the historical WIP data of each product at the current site, firstly predicts the future WIP data of the current site, and then, based on the first prediction result of the future WIP data of the current site, obtains a second prediction result of the future WIP data of the current site. Compared with obtaining the prediction result of the future WIP data of the current site solely based on the WPH data of each product at the current site, the solution of this invention also combines product demand data and the historical WIP data of each product at the current site for prediction. Since the product mix to be produced and changes in product demand at the site will affect the site's capacity, and the historical WIP data of each product at the current site also has a certain reference value for capacity parameter prediction, combining product demand data and the historical WIP data of each product at the current site for prediction can better reflect the characteristics of the products, thereby obtaining a more accurate prediction result of the future WIP data of the current site, and thus enabling a more precise prediction of capacity. Attached Figure Description

[0033] Figure 1 This is a flowchart of a capacity forecasting method according to an embodiment of the present invention;

[0034] Figure 2 This is a summary diagram of the capacity parameter values ​​of all stations on a factory production line;

[0035] Figure 3 This is a schematic diagram of the structure of a capacity prediction device in an embodiment of the present invention. Detailed Implementation

[0036] Currently, capacity forecasting is usually based solely on the output per hour (WPH) of the machines to predict the capacity parameters of representative products. The forecast results of the capacity parameters of representative products are then used to calculate the factory capacity and finally draw a conclusion.

[0037] In the aforementioned capacity forecasting methods, some factors affecting capacity are not reflected, leading to discrepancies between the predicted and actual capacity parameters. For example, a predicted capacity parameter of 100% indicates that capacity is fully utilized and no new orders can be accepted. However, in reality, different platforms have different process requirements. Such a forecast could result in unmet order demands or idle capacity, leading to decreased company profits, reduced customer satisfaction, or failure to maximize company profits.

[0038] To address this problem, this invention provides a capacity forecasting method. This method, when forecasting capacity parameters, combines not only the current site's WPH data but also product demand data and historical Work-In-Process (WIP) data for each product at the current site. Since changes in the product mix to be produced and product demand at a site affect the site's capacity, and the historical WIP data for each product at the current site also provides a certain reference for capacity parameter forecasting, the method can more accurately obtain the future WIP data forecast results for the current site, thereby enabling more precise forecasting of capacity.

[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0040] Reference Figure 1 This invention provides a capacity forecasting method, which may include the following steps:

[0041] Step 11: Obtain product demand data for the current site, WPH data for each product on the current site, and historical WIP data for each product on the current site.

[0042] In practice, the product demand data may include: the identifiers of each product to be produced at the current site and the required quantity of each product. For example, the product demand data may include: product 1 and its required quantity, product 2 and its required quantity, product 3 and its required quantity, etc. Product 1, product 2, and product 3 are the identifiers of the products to be produced. The required quantity of product 1 can be 10 pieces, the required quantity of product 2 can be 20 pieces, and the required quantity of product 3 can be 30 pieces.

[0043] In practice, the WPH data for each product at the current site represents the output data of each product per hour at the current site. The WPH data for each product at the current site is usually a fixed value. For example, if the WPH data for product 1 at the current site is 2, it means that product 1 can produce 2 pieces per hour at the current site.

[0044] In practice, the historical WIP data for each product at the current site refers to the historical work-in-process inventory data for each product at the current site, also known as the stockpiling quantity. Specifically, historical WIP data can be obtained based on the predicted duration of each WIP data point. The predicted duration of each WIP data point is equal to the historical WIP data acquisition period.

[0045] For example, when each predicted WIP data point spans one month, historical WIP data is also retrieved monthly. When predicting WIP data for the first month of the future, only the WIP data for the previous month for each product on the current site needs to be retrieved. When predicting WIP data for the first, second, third, and fourth months of the future, then the WIP data for the previous first, second, third, and fourth months for each product on the current site needs to be retrieved.

[0046] Step 12: Based on the product demand data, the WPH data of each product on each machine at the current site, and the historical WIP data of each product at the current site, make a first prediction of the future WIP data of the current site to obtain the first prediction result of the future WIP data of the current site.

[0047] In one embodiment of the present invention, based on the product demand data and the WPH data of each product at the current site, a sliding window can be used to reconstruct the historical WIP data of each product at the current site to obtain a first prediction result of the future WIP data of the current site.

[0048] Specifically, based on the WPH data of each product at the current site and the product demand data, the predicted time step of each product at the current site can be obtained first. Then, using the predicted time step and a preset sliding window size, the historical WIP data of each product at the current site can be reconstructed to obtain the first prediction result of the future WIP data of the current site.

[0049] In practical implementation, the WPH data of each product at the current site can be used as the input time step, combined with the product demand, to obtain the predicted time step. The predicted time step H is defined as: (Sum of demand for each product) / (Total processing time for each product). The total processing time for each product is calculated as: (∑Product demand) / WPH data.

[0050] For example, the products to be manufactured include product 1, product 2 and product 3. The demand for product 1 is 10 pieces with a WPH of 2, the demand for product 2 is 20 pieces with a WPH of 5, and the demand for product 3 is 10 pieces with a WPH of 2. Then the total processing time = 10 / 2 + 20 / 3 + 10 / 2, and the predicted time step H = (10 + 20 + 10) / (10 / 2 + 20 / 5 + 10 / 2) ≈ 3.57.

[0051] In practice, the sliding window size *n* is equal to the number of products to be produced. For example, when the number of products to be produced is 3, then *n* = 3. When the number of products to be produced is 4, then *n* = 4. The historical WIP data of each product is sequentially concatenated to form a one-dimensional array *G*. The sliding window *n* slides within this one-dimensional array, acquiring *n* data points in array *G* with each slide. These *n* data points form a row. After sliding *H* times, a two-dimensional array *X* with *H* rows and *n* columns is obtained. Each column of this two-dimensional array *X* serves as the first prediction result for future WIP data of the same product.

[0052] For example, assuming there are n products and N historical WIP data points, each historical WIP data point can be represented as gi, where gi = G{1, 2, 3, 4, ..., N}. Then, after sliding window reconstruction, the resulting two-dimensional array X can be represented as:

[0053]

[0054] In one embodiment of the present invention, in order to further improve the accuracy of capacity forecasting, the historical WIP data of each product at the current site can be preprocessed before using a sliding window to reconstruct the historical WIP data of each product at the current site.

[0055] In practice, various methods can be used to preprocess historical WIP data.

[0056] In one embodiment of the present invention, reference is made to... Figure 1 First, the historical WIP data of each product at the current site can be supplemented using spline interpolation. Then, the Savitzky-Golay method can be used to smooth the supplemented data, and a sliding window reconstruction can be performed on the smoothed data.

[0057] Specifically, when forecasting production capacity parameters, historical work-in-process (WIP) data for a site will change due to variations in the product mix. Based on historical production information, there may be situations where the production capacity of a particular product mix can be met, but historical WIP data is not available. For these missing historical WIP data, spline interpolation is used to supplement them. A spline function is a special function whose interpolation algorithm can effectively solve the problem of missing values, thus ensuring that each product has corresponding historical WIP data, thereby improving the accuracy of the forecast.

[0058] Furthermore, in actual production, historical WIP data acquired due to machine maintenance, gas and water treatment alarms, etc., may contain outliers, significantly affecting prediction accuracy. To remove outliers and smooth the data, the Savitzky-Golay method can be used. Savitzky-Golay is a filtering method, also known as the SG filter, which is a filtering method based on local polynomial least squares fitting in the time domain. Using the Savitzky-Golay method to filter data can remove noise without reducing signal resolution, resulting in better filtering performance and thus improving prediction accuracy.

[0059] Step 13: Based on the product demand data and the first prediction result, a second prediction result of the future WIP data of the current site is obtained using a preset prediction model, so as to predict the future production capacity of the current site based on the second prediction result.

[0060] In practice, the prediction model can be implemented using a variety of machine learning models.

[0061] In one embodiment of the present invention, the prediction model is an XGBoost model. XGBoost is a boosting tree model, an improvement on Gradient Boosting Decision Tree (GBDT), which can integrate many tree models together to form a powerful classifier. XGBoost can achieve parallel construction of regression trees through multi-threading, maximizing computational speed and efficiency. While GBDT only utilizes information from the first derivative, XGBoost performs a second-order Taylor expansion on the loss function, resulting in higher computational accuracy.

[0062] The XGBoost algorithm selects features by calculating the importance of an attribute within a single decision tree. This importance is determined by the magnitude of the performance improvement metric at each attribute split point, with each node weighting and recording the number of improvements. The greater the performance improvement metric of an attribute at a split point, the higher its weight and the more important the attribute. Finally, the results for each attribute across all boosting trees are weighted, summed, and averaged to obtain the importance score.

[0063] The prediction accuracy of the XGBoost model is determined by both model bias and variance. The loss function represents the model bias, while the regularization term reduces variance and prevents overfitting. Therefore, the objective function consists of the model's loss function L and the regularization term Ω, which suppresses model complexity. Specifically, it can be expressed as: Objective function Obj = L + Ω. When constructing an XGBoost model, it is necessary to find the optimal parameters and establish the optimal model based on the principle of minimizing the objective function.

[0064] When using the XGBoost model to predict capacity parameters, the future WIP data of each product in the first prediction result can be combined with the product demand data, and each data combination can be input into the preset XGBoost model to obtain the second prediction result of the future WIP data of each product. Then, based on the second prediction result of the future WIP data of each product, the final capacity parameter value can be obtained.

[0065] Specifically, taking a two-dimensional array X as an example, each column in the two-dimensional array X represents the future WIP data of the same product, and there are n columns for n products.

[0066] The first column of the two-dimensional array X and the demand data of the corresponding product in the first column are used as input to the XGBoost model. The XGBoost model is then used to make a second prediction of the future WIP data of the product, and the second prediction result of the future WIP data of the product is obtained.

[0067] The demand data of the second column of the two-dimensional array X and the corresponding product in the second column are used as another input to the XGBoost model. The XGBoost model is then used to make a second prediction of the future WIP data of the product corresponding to the second column, and the second prediction result of the future WIP data of the product is obtained.

[0068] ...

[0069] The demand data of the nth column and the corresponding product in the two-dimensional array X are used as another input to the XGBoost model. The XGBoost model is then used to make a second prediction of the future WIP data of the product corresponding to the nth column, and the second prediction result of the future WIP data of the product is obtained.

[0070] Therefore, a second prediction of the future WIP data for each product to be produced can be obtained, which serves as the predicted capacity parameter value. Based on the second prediction of the future WIP data for the products to be produced, the final capacity parameter value can be obtained.

[0071] In practical implementation, when obtaining the final capacity parameter value based on the second prediction result of the future WIP data of the products to be produced, the ratio of the sum of the demand for each product to be produced to the total processing time of each product can be used as the final capacity parameter. At this time, the total processing time of each product = the sum of the ratios of the second prediction result of the future WIP data of each product to be produced to the corresponding WPH data, which is used as the predicted capacity parameter value.

[0072] For example, assuming the second forecast result for product 1 is 15, the second forecast result for product 2 is 10, and the second forecast result for product 3 is 10, and the WPH data for products 1 to 3 is 5, then the capacity parameter value = (15+10+10) / (15 / 5+10 / 5+10 / 5) = 35 / 7 = 5.

[0073] In practice, the range of the capacity parameter is (0, 1). If the final predicted capacity parameter value is greater than 1, it means that the factory cannot meet the current product demand, that is, the factory cannot meet the customer's demand, and it is not recommended to accept the order or to negotiate a delay in delivery.

[0074] Using the solution of this invention, the capacity parameters of all stations on a factory production line are summarized, and the capacity parameter values ​​of each station are as follows: Figure 2 As shown. From Figure 2 As can be seen, within this factory, the production capacity of some stations is greater than 1 (as shown in the dashed box). For stations with a production capacity parameter greater than 1, it indicates that the station cannot meet product demand. The product demand for the station can be adjusted, for example, by reducing the demand for a certain product to be produced, so that the production capacity parameter value of the station is between (0, 1).

[0075] As can be seen from the above, the solutions of the embodiments of the present invention consider all factors affecting production capacity and then use a prediction model for prediction, which can effectively improve the accuracy of product prediction. Furthermore, in some embodiments, interpolation and smoothing methods are combined for the first time to process production capacity data and embedded into the XGboost model to predict production capacity parameters, which can further improve the accuracy of product prediction.

[0076] The solution adopted in this invention can be used to locate bottlenecks in the production line, so as to proactively address overload situations and thereby improve production capacity. The solution of this invention can also be embedded in a market order-taking model as a standard for determining whether an order can be accepted, thus improving the order-taking system.

[0077] To enable those skilled in the art to better understand and implement the present invention, the apparatus, testing system, electronic device, and computer-readable storage medium corresponding to the above method are described in detail below.

[0078] Reference Figure 3 This invention also provides a capacity prediction device 30, which may include: an acquisition unit 31, a first prediction unit 32, and a second prediction unit 33. Wherein:

[0079] The acquisition unit 31 is adapted to acquire product demand data of the current site, WPH data of each product at the current site, and historical WIP data of each product at the current site.

[0080] The prediction unit 32 is adapted to make a first prediction of the future WIP data of the current site based on the product demand data, the WPH data of each product at the current site and the historical WIP data of each product at the current site, and to obtain a first prediction result of the future WIP data of the current site.

[0081] The output unit 113 is adapted to obtain a second prediction result of the future WIP data of the current site based on the product demand data and the first prediction result, using a preset prediction model, so as to predict the future production capacity of the current site based on the second prediction result.

[0082] The acquisition unit 31, the first prediction unit 32, and the second prediction unit 33 can be implemented with reference to the above description of steps 11 to 13, and will not be repeated here.

[0083] This invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of any of the above methods.

[0084] In specific implementations, the computer-readable storage medium may include ROM, RAM, disk, or optical disk, etc.

[0085] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor runs the computer program, it performs the steps of any of the methods described above.

[0086] Regarding the modules / units included in the various devices and products described in the above embodiments, they can be software modules / units, hardware modules / units, or a combination of both. For example, for various devices and products applied to or integrated into a chip, all of their modules / units can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs that run on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits; for various devices and products applied to or integrated into a chip module, all of their modules / units can be implemented using hardware methods such as circuits, and different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The components can be implemented using software programs that run on the processor integrated within the chip module. The remaining (if any) modules / units can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into the terminal, each of its components / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or in different components within the terminal. Alternatively, at least some modules / units can be implemented using software programs that run on the processor integrated within the terminal, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits.

[0087] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A capacity forecasting method, characterized in that, include: Obtain product demand data for the current site, WPH data for each product on the current site, and historical WIP data for each product on the current site; Based on the product demand data, the WPH data of each product at the current site, and the historical WIP data of each product at the current site, a first prediction is made on the future WIP data of the current site to obtain the first prediction result of the future WIP data of the current site. Based on the product demand data and the first prediction result, a second prediction result for the future WIP data of the current site is obtained using a preset prediction model, and the future production capacity of the current site is predicted based on the second prediction result.

2. The capacity forecasting method as described in claim 1, characterized in that, The product demand data includes: the identifiers of each product to be produced at the current site and the demand quantity of each product to be produced.

3. The capacity forecasting method as described in claim 1, characterized in that, The first prediction of future WIP data for the current site is made based on the product demand data, the WPH data of each product at the current site, and the historical WIP data of each product at the current site, resulting in a first prediction result of future WIP data for the current site, including: Based on the product demand data and the WPH data of each product at the current site, the historical WIP data of each product at the current site is reconstructed using a sliding window to obtain the first prediction result of the future WIP data of the current site.

4. The capacity forecasting method as described in claim 3, characterized in that, Based on the product demand data and the WPH data of each product at the current site, the historical WIP data of each product at the current site is reconstructed using a sliding window to obtain the first prediction result of the future WIP data of the current site, including: Based on the WPH data of each product at the current site, the product demand data, and the average WPH data of each product at the current site, the prediction time step of each product at the current site is obtained; using the prediction time step and a preset sliding window size, the historical WIP data of each product at the current site is reconstructed to obtain the first prediction result of the future WIP data of the current site.

5. The capacity forecasting method as described in claim 3, characterized in that, Before reconstructing the historical WIP data of each product at the current site using a sliding window based on the product demand data and the WPH data of each product at the current site, the method further includes: The historical WIP data of each product at the current site is preprocessed.

6. The capacity forecasting method as described in claim 5, characterized in that, The preprocessing of the historical WIP data of each product at the current site includes: The historical WIP data of each product at the current site is supplemented using spline interpolation. The Savitzky-Golay method is used to smooth the supplemented data, and then a sliding window is used to reconstruct the smoothed data.

7. The capacity forecasting method as described in claim 6, characterized in that, The second prediction result for the future WIP data of the current site, obtained by using a preset prediction model based on the product demand data and the first prediction result, includes: The future WIP data of each product in the first prediction result is combined with the product demand data, and each data combination is input into the preset XGBoost model to obtain the second prediction result of the future WIP data of each product. The final capacity parameter values ​​are obtained based on the second forecast results of the future WIP data for each product.

8. A capacity prediction device, characterized in that, include: The acquisition unit is suitable for acquiring product demand data for the current site, WPH data for each product at the current site, and historical WIP data for each product at the current site. The first prediction unit is adapted to make a first prediction of the future WIP data of the current site based on the product demand data, the WPH data of each product at the current site and the historical WIP data of each product at the current site, and to obtain the first prediction result of the future WIP data of the current site. The second prediction unit is adapted to obtain a second prediction result of the future WIP data of the current site based on the product demand data and the first prediction result, using a preset prediction model, so as to predict the future production capacity of the current site based on the second prediction result.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor to implement the steps of the method according to any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program capable of running on the processor, characterized in that, When the processor runs the computer program, it performs the steps of the method according to any one of claims 1 to 7.

Images