Visualized warehouse tracking management system based on steel structure production

Abstract

The present invention discloses a visualized warehousing tracking management system based on steel structure production, comprising: an original data acquisition module, used for acquiring a number of original material demand data at the current time point; a feature processing module, used for performing feature processing on a number of original material demand data at the current time point to obtain a material demand vector at the current time point; a vector input module, used for inputting the material demand vector at the current time point into a pre-trained ARIMA parent model; a demand output module, used for the ARIMA parent model to remove long-term trend fluctuations and short-term random disturbances in a time series respectively through an internal sub-model, and output a forecast value of original material demand within a time window defined by a user at the current time point; the present invention can accurately detect the lag structure and residual influence of the time series in the raw material demand data for steel structure production, and significantly improves the accuracy of the prediction of the raw material demand for steel structure production in combination with the user-defined time window and sliding calculation method.

Description

Technical Field

[0001] The invention relates to warehouse management, in particular to a visual warehouse tracking management system based on steel structure production. Background Art

[0002] In the process of steel structure production, the types, demand and distribution requirements of materials are highly dynamic and complex, especially during the peak production period, and accurate management of material inventory and distribution is particularly important. The patent document with patent publication number CN116228092A discloses a method for intelligent management of raw materials based on MES of prefabricated steel structure factory, which effectively helps enterprises to achieve refined management of raw materials through the coordination of software and hardware; however, the above patent technology and the existing technology in the field of steel structure warehousing, its warehousing management method still relies heavily on static data or simple rules to estimate material demand, which makes it difficult to respond to changes in production rhythm in real time, which may lead to material shortages or surpluses, thereby affecting production efficiency. More importantly, for the demand forecasting of raw materials for steel structure production, its method does not adequately analyze the characteristics of long-term fluctuations and short-term disturbances of time series data, fails to make full use of trend and periodicity information, and results in insufficient prediction accuracy. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a visual warehouse tracking management system based on steel structure production, which solves the technical problems raised in the background technology by constructing an ARIMA parent model that can capture long-term trends and short-term fluctuations.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0005] Visual warehouse tracking management system based on steel structure production, including:

[0006] The original data acquisition module is used to obtain some original material demand data at the current time point;

[0007] A feature processing module is used to perform feature processing on a number of original material demand data at the current time point to obtain a material demand vector at the current time point;

[0008] The vector input module is used to input the material demand vector at the current time point into the pre-trained ARIMA parent model;

[0009] The demand output module is used for the ARIMA parent model to remove the long-term trend fluctuations and short-term random disturbances in the time series through the internal sub-model, and output the original material demand forecast value within the user-defined time window based on the current time point.

[0010] In some of the embodiments, the modeling steps of the ARIMA parent model include:

[0011] S1. On the time axis, obtain a number of original material demand data at N historical moments according to a continuous time series, and adjust the data length of the original material demand data based on the time window defined by the user;

[0012] S2. For each historical moment, feature processing is performed on a number of original material demand data to obtain material demand vectors of N historical moments; wherein the material demand vectors include numerical features and non-numerical features.

[0013] S3, performing d times of difference processing on the numerical characteristics of the material demand vector at N historical moments within the user-defined time window until Nd data-stable difference vectors are generated;

[0014] S4, performing autocorrelation analysis and partial autocorrelation analysis on the stable differential vector of the data, and generating an autocorrelation function graph and a partial autocorrelation function graph respectively;

[0015] S5. According to the generated autocorrelation function graph and partial autocorrelation function graph, the lag periods of the truncated positions are defined as the autoregressive order p and the sliding average order q respectively;

[0016] S6. Configure the parameters of the ARIMA model according to the number of differences d, the autoregressive order p and the sliding average order q, and construct an initialization model;

[0017] Wherein, the initialization model consists of an autoregressive sub-model and a sliding average sub-model;

[0018] S7, divide the Nd data stationary difference vectors into multiple time windows according to the time window length w defined by the user, and use each time window as input, slide it into the initialization model in sequence, and predict the joint prediction vector of each historical moment in the first round point by point;

[0019] S8, based on the historical moments in each time window in the first round, the prediction vector and the true difference vector are combined to calculate the residual in the time window point by point; after the calculation is completed, the time window is slid to update to the next batch of historical moments, and the residual is continued to be calculated;

[0020] S9, substituting the residual of the first round into the parameter optimization function to iteratively update the model parameters of the second round;

[0021] The parameter optimization function is:

[0022] Among them, L(φ,θ,μ) represents the parameter optimization function;

[0023] φ=[φ1,φ2,...,φ p] is the autoregressive coefficient of the autoregressive sub-model, which represents the influence of the lagged difference vector in the time series on the autoregressive prediction vector at the current moment;

[0024] θ=[θ1,θ2,…,θ q ] is the sliding average coefficient of the sliding average sub-model, which represents the influence of the lagged residual on the sliding average prediction vector at the current moment;

[0025] μ is a constant term, which represents the predicted benchmark value of the material demand vector of the moving average sub-model;

[0026] σ 2 is the global variance of the residual, indicating that the residual follows a normal distribution;

[0027] ∈ t is the residual, which represents the prediction error between the true difference vector and the joint prediction vector;

[0028] S10, using the model parameters of the second round, rolling the material demand vector in the time window defined by the user, and recalculating the joint prediction vector at each historical moment until the Nd-th historical moment;

[0029] S11. Iteratively optimize the model parameters of subsequent rounds until the update amplitude of the model parameters is less than the set change threshold.

[0030] In some embodiments, performing autocorrelation analysis and partial autocorrelation analysis on the stationary difference vector of the data to generate an autocorrelation function graph and a partial autocorrelation function graph respectively includes:

[0031] S4-1, receive the stable difference vector of the data, define the current period and its lag period;

[0032] S4-2, for each lag period k, using the autocorrelation function to calculate the autocorrelation coefficient of the difference vector between the current time point and the k-lag time point;

[0033] S4-3, for each lag period k, calculate the partial autocorrelation coefficient using the partial autocorrelation function;

[0034] S4-4. Draw bar graphs according to the calculated autocorrelation coefficient and partial autocorrelation coefficient corresponding to the lag period k to generate the autocorrelation function graph and the partial autocorrelation function graph; wherein the horizontal axis represents the lag period k, and the vertical axis represents the autocorrelation coefficient and the partial autocorrelation coefficient, respectively.

[0035] In some of the embodiments, based on the generated autocorrelation function graph and partial autocorrelation function graph, the lag periods of the truncated positions are defined as the autoregressive order p and the sliding average order q, respectively, including:

[0036] S5-1. In the autocorrelation function diagram, find the lag period at which the autocorrelation coefficient drops rapidly from the significance value to zero, and determine it as the sliding average order q;

[0037] S5-2. In the partial autocorrelation function diagram, find the lag period at which the partial autocorrelation coefficient drops rapidly from the significance value to zero, and determine it as the autoregressive order p.

[0038] In some embodiments, the calculation expression of the autocorrelation function is:

[0039]

[0040] Among them, r k represents the autocorrelation coefficient when the lag period is k, y t represents the difference vector at the tth historical moment, represents the mean of the difference vector.

[0041] In some embodiments, the partial autocorrelation function is calculated as:

[0042]

[0043] Among them, φ k，k represents the partial autocorrelation coefficient when the lag period is k, φ j,k-1 It represents the autocorrelation coefficient when the lag period is j.

[0044] In some embodiments, predicting the joint prediction vector of each historical moment in the first round point by point includes:

[0045] S7-1, selecting a time window from the t-w+1th historical moment to the tth historical moment, and calculating a joint prediction vector of the difference vector in the time window;

[0046] S7-2, after each prediction is completed, the time window slides forward by one unit of time, removes the earliest difference vector, and adds the current difference vector to form a new input window, and outputs several joint prediction vectors;

[0047] S7-3. Repeat the above steps until the joint prediction vector from the first historical moment to the Nd-th historical moment is predicted.

[0048] In some embodiments, calculating a joint prediction vector of the difference vectors in the time window includes:

[0049] S7-1-1, define the initial lagged difference vector and initial lagged residual at the first historical moment;

[0050] S7-1-2, using the initial lagged difference vector at the first historical moment as the input of the autoregressive sub-model to obtain the autoregressive prediction vector at the first historical moment;

[0051] The expression of the autoregressive prediction vector at the first historical moment is:

[0052]

[0053] in, represents the autoregressive prediction vector at the first historical moment, y lm represents the historical mean of the lagged difference vector;

[0054] φ1, φ2, …φ p is the coefficient of the autoregressive sub-model, indicating the weights of different lag moments;

[0055] S7-1-3, using the initial lagged residual at the first historical moment as the input of the sliding average sub-model to obtain the sliding average prediction vector at the first historical moment;

[0056] Among them, the expression of the sliding average prediction vector at the first historical moment is:

[0057]

[0058] in, represents the moving average forecast vector at the first historical moment;

[0059] μ is a constant term, which represents the predicted benchmark value of the material demand vector of the moving average sub-model;

[0060] θ1,θ2,…θ q is the coefficient of the moving average sub-model, indicating the weight of the lagged residual;

[0061] ∈ li represents the lagged residual at the first historical moment, and its value is zero;

[0062] S7-1-4, adding the autoregressive prediction vector at the first historical moment and the sliding average prediction vector at the first historical moment to obtain a joint prediction vector at the first historical moment;

[0063] The expression of the joint prediction vector at the first historical moment is:

[0064] represents the joint prediction vector at the first historical moment;

[0065] S7-1-5, calculating the current residual at the first historical moment according to the joint prediction vector at the first historical moment and the difference vector at the first historical moment, and defining it as the lagged residual at the second historical moment;

[0066] The expression of the current residual at the first historical moment is:

[0067] Among them, ∈1 represents the current residual at the first historical moment, and y1 represents the difference vector at the first historical moment;

[0068] S7-1-6, according to the joint prediction vector of the first historical moment and the initial lagged difference vector of the first historical moment, update the current lagged difference vector of the first historical moment, and define it as the lagged difference vector of the second historical moment;

[0069] S7-1-7, calculating a joint prediction vector at the second historical moment according to the lagged difference vector at the second historical moment and the lagged residual at the second historical moment;

[0070] S7-1-8, iteratively update the lagged difference vector and the lagged residual, repeat the steps S7-1-5 to S7-1-7, and slide the time window until all historical moments are traversed to complete the calculation of the joint prediction vector from the first historical moment to the Ndth;

[0071] The expression for updating the lagged difference vector is:

[0072] The expression for updating the lagged residual is:

[0073] in, represents the lagged difference vector at the current historical moment t, represents the lagged residual at the current historical moment t, y t-1 Represents the material demand vector at the previous historical moment, ∈ t-1 Represents the residual at the previous historical moment.

[0074] In some of the embodiments, the residual in the time window is calculated point by point based on the historical moments in each time window in the first round, the prediction vector and the true difference vector are combined; after the calculation is completed, the time window is slid to update to the next batch of historical moments, and the residual is continued to be calculated, including:

[0075] S8-1, select a time window from the t-w+1th historical moment to the tth historical moment, and calculate the residuals of w difference vectors and the corresponding joint prediction vector in the time window;

[0076] S8-2, slide the time window forward by one unit time, remove the t-w+1th historical moment from the time window, and add the difference vector and the joint prediction vector at the t+1th historical moment;

[0077] S8-3. Continue to calculate the residuals in the sliding time window point by point until the residuals of Nd historical moments are calculated.

[0078] The expression of the residual of the first round is:

[0079]

[0080] ∈ t represents the residual at the tth historical moment in the first round, y t represents the difference vector at the tth historical moment, represents the joint prediction vector at the t-th historical moment; represents the autoregressive prediction vector output by the autoregressive sub-model, A moving average forecast vector representing the output of the moving average model.

[0081] In some embodiments, the output of the original material demand forecast value includes:

[0082] A-1. Initialize the reverse differential parameters;

[0083] The reverse difference parameters include: the number of differences d and Nd joint prediction vectors;

[0084] A-2. For each joint forecast vector, perform reverse difference to restore it to the material demand forecast value of the original scale;

[0085] The expression of the material requirement vector in the original scale is:

[0086] in, Represents the material demand forecast value at the original scale after restoration. represents the joint prediction vector at the t-th historical moment, represents the joint prediction vector of the previous historical moment;

[0087] A-3. If the number of differential operations d is greater than 1, continue to perform reverse differential operations until d reverse differential operations are completed, and output the material demand forecast value within the user-defined time window.

[0088] The present invention provides a visual warehousing tracking management system based on steel structure production, which has the following beneficial effects:

[0089] The present invention uses the ARIMA model to deeply explore the long-term trend and short-term volatility characteristics in the time series, and accurately extracts the lag structure and residual effect of the time series through differential processing, autocorrelation and partial autocorrelation analysis. At the same time, combined with the user-defined time window and sliding calculation method, it dynamically adapts to the changing characteristics of material demand, significantly improving the accuracy of the forecast of raw material demand for steel structure production.

[0090] Furthermore, through the reverse difference operation, the forecast results are restored to the original data scale to generate forecast values ​​that are consistent with the actual material demand. Combined with the visual display of the forecast values, managers can grasp the material demand in the future time window in real time, thereby optimizing the warehouse layout, rationally dispatching transportation resources, reducing inventory costs and avoiding production interruptions caused by insufficient material supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 This is a structural block diagram of the visual warehousing tracking management system based on steel structure production of the present invention;

[0092] Figure 2 This is a management flow chart of the visual warehousing tracking management system based on steel structure production of the present invention;

[0093] Figure 3 A flow chart for generating an autocorrelation function graph and a partial autocorrelation function graph in the present invention;

[0094] Figure 4 This is a logic flow chart for outputting material demand forecast values ​​in the present invention. DETAILED DESCRIPTION

[0095] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0096] Example 1: Please refer to Figure 1 to Figure 2 The present invention provides a visual storage tracking management system based on steel structure production, including:

[0097] The original data acquisition module is used to obtain some original material demand data at the current time point;

[0098] A feature processing module is used to perform feature processing on a number of original material demand data at the current time point to obtain a material demand vector at the current time point;

[0099] Specifically, the material demand vector at each time point in the time series contains features of multiple dimensions, and the features are represented as demand attributes, such as material type (steel, welding materials), material demand quantity, demand location (warehouse A, production line B), transportation method (sea transportation, land transportation), etc.

[0100] For numerical features (such as material demand quantity), the Z-score standardization method is used to eliminate the dimension effect; for non-numerical features (such as material type, demand location, and transportation method), feature encoding (such as one-hot encoding or label encoding) is performed and converted into numerical form.

[0101] The vector input module is used to input the material demand vector at the current time point into the pre-trained ARIMA parent model;

[0102] The demand output module is used for the ARIMA parent model to remove the long-term trend fluctuations and short-term random disturbances in the time series through the internal sub-model, and output the original material demand forecast value within the user-defined time window based on the current time point.

[0103] The output of the original material demand forecast value includes the following steps:

[0104] A-1. Initialize the reverse differential parameters;

[0105] The reverse difference parameters include: the number of differences d and Nd joint prediction vectors;

[0106] A-2. For each joint forecast vector, perform reverse difference to restore it to the material demand forecast value of the original scale;

[0107] The expression of the material requirement vector in the original scale is:

[0108] in, Represents the material demand forecast value at the original scale after restoration. represents the joint prediction vector at the t-th historical moment, represents the joint prediction vector of the previous historical moment;

[0109] A-3. If the number of differential operations d is greater than 1, continue to perform reverse differential operations until d reverse differential operations are completed, and output the material demand forecast value within the user-defined time window.

[0110] In this embodiment, a visual warehouse tracking and management system based on steel structure production gradually restores the joint prediction vector output by the model to the material demand prediction value of the original scale through the reverse difference operation, and finally realizes the accurate prediction of material demand within the user-defined time window.

[0111] In this embodiment, the material demand vector at the current time point is input into the pre-trained ARIMA model to accurately predict the material demand within the user-defined time window, providing data support for the visual warehouse tracking management system in steel structure production. The system analyzes the time dependence and fluctuation law of historical material demand data, removes long-term trend fluctuations and short-term random disturbances in the data, and extracts features that can reflect changes in material demand.

[0112] In steel structure production, there are many types of materials, large fluctuations in demand, and limited storage space. Traditional storage management methods are difficult to cope with complex production needs. This system realizes dynamic prediction of future material demand by modeling historical material demand data and combining it with the user-defined time window. Based on the predicted original material demand forecast value, it is digitally displayed on the user's management terminal interface through visual data means, such as the control center display screen or mobile phone APP and other management terminals with visual display functions. In this way, a direct connection between the allocation and utilization of storage resources and users can be established, ensuring the timely supply of key production materials while avoiding excessive backlog or shortage of resources.

[0113] Example 2: See Figures 1 to 4 The technical solution of this embodiment 2 is different from that of embodiment 1 in that the modeling steps of the ARIMA parent model in embodiment 1 are disclosed, and the modeling steps include:

[0114] S1. On the time axis, obtain a number of original material demand data at N historical moments according to a continuous time series, and adjust the data length of the original material demand data based on the time window defined by the user;

[0115] S2. For each historical moment, feature processing is performed on a number of original material demand data to obtain material demand vectors of N historical moments; wherein the material demand vectors include numerical features and non-numerical features.

[0116] S3, performing d times of difference processing on the numerical characteristics of the material demand vector at N historical moments within the user-defined time window, gradually removing the trend and periodic fluctuations, until Nd data-stable difference vectors are generated;

[0117] The main purpose of differential processing is to remove trend or cyclical fluctuations in time series, thereby generating stable time series data. This processing includes first-order and second-order differential operations. It should be noted that non-numerical attributes (such as material type, demand location, and transportation method) retain their original information in differential processing and do not participate in numerical calculations.

[0118] Specifically, data stationarity means that the statistical properties of the time series (such as mean, variance, autocorrelation structure, etc.) remain unchanged over time. This means that stationary data does not have a significant trend or seasonal component.

[0119] In the material demand forecasting of steel structure production, "data stability" means that the long-term trend (for example, a trend of continuous growth or decrease) is removed from the material demand data. Taking the annual expansion of production scale as an example, this may lead to a long-term upward trend in material demand. This trend can be removed through differential processing, so that the data can be more focused on the short-term fluctuation part. At the same time, differential processing can also remove cyclical components (for example, seasonal changes in material demand that increase every winter), so that the data reaches a stable state, which is more suitable for subsequent analysis and forecasting.

[0120] S4, performing autocorrelation analysis and partial autocorrelation analysis on the stable differential vector of the data, and generating an autocorrelation function graph and a partial autocorrelation function graph respectively;

[0121] Among them, autocorrelation analysis is used to reveal the overall correlation between the lagged value and the current value in the sequence, and partial autocorrelation analysis is used to reveal the direct correlation between the lagged value and the current value.

[0122] S5. According to the generated autocorrelation function graph and partial autocorrelation function graph, the lag periods of the truncated positions are defined as the autoregressive order p and the sliding average order q respectively;

[0123] Among them, the number of differences indicates the number of differences required to make the data stable, the autoregressive order indicates the number of lagged values ​​in the time series related to the current value, and the moving average order indicates the number of lagged values ​​of the prediction residual.

[0124] S6. Configure the parameters of the ARIMA model according to the number of differences d, the autoregressive order p and the sliding average order q, and construct an initialization model;

[0125] The initialization model consists of an autoregressive sub-model (AR part) and a moving average sub-model (MA part); the autoregressive model is used to capture the influence of lagged values ​​in the time series, and the moving average sub-model is used to capture the lagged influence of the residual.

[0126] S7, divide the Nd data stationary difference vectors into multiple time windows according to the time window length w defined by the user, and use each time window as input, slide it into the initialization model in sequence, and predict the joint prediction vector of each historical moment in the first round point by point;

[0127] S8, based on the historical moments in each time window in the first round, the prediction vector and the true difference vector are combined to calculate the residual in the time window point by point; after the calculation is completed, the time window is slid to update to the next batch of historical moments, and the residual is continued to be calculated;

[0128] S9, substituting the residual of the first round into the parameter optimization function to iteratively update the model parameters of the second round;

[0129] The parameter optimization function is:

[0130] Where L(φ, θ, μ) represents the parameter optimization function;

[0131] φ=[φ1,φ2,…φ p ] is the autoregressive coefficient of the autoregressive sub-model, which represents the influence of the lagged difference vector in the time series on the autoregressive prediction vector at the current moment;

[0132] θ=[θ1,θ2,...,θ q ] is the sliding average coefficient of the sliding average sub-model, which represents the influence of the lagged residual on the sliding average prediction vector at the current moment;

[0133] μ is a constant term, which represents the predicted benchmark value of the material demand vector of the moving average sub-model;

[0134] σ 2 is the global variance of the residual, indicating that the residual follows a normal distribution;

[0135] ∈ t is the residual, which represents the prediction error between the true difference vector and the joint prediction vector;

[0136] Specifically, The normalization term that indicates that the residual follows the normal distribution probability density is a constant term; and It is a weighted term of the residual squares. The parameter optimization function minimizes the prediction error of the model by minimizing the sum of the residual squares.

[0137] S10, using the model parameters of the second round, rolling the material demand vector in the time window defined by the user, and recalculating the joint prediction vector at each historical moment until the Nd-th historical moment;

[0138] S11. Iteratively optimize the model parameters of subsequent rounds until the update amplitude of the model parameters is less than the set change threshold.

[0139] For the material demand data of steel structure production, the ARIMA parent model is used to provide a modeling method that dynamically adapts to time series fluctuations. The modeling process of the ARIMA parent model includes multiple steps.

[0140] First, the model extracts raw data within a user-defined time window from the material demand data at historical moments. The model gradually eliminates long-term trend and seasonal fluctuations through differential processing, making the data reach a stable state, thus focusing on short-term fluctuation characteristics.

[0141] Next, based on the generated autocorrelation function (ACF) graph and partial autocorrelation function (PACF) graph, the autoregressive order and moving average order are determined to accurately describe the impact of lagged values ​​and residuals in the material demand series.

[0142] In the initialization model, the autoregressive part captures the lagged impact of historical material demand through the lagged difference vector, and the moving average part eliminates the impact of random disturbances through the lagged residuals.

[0143] Subsequently, the stationary data is divided into multiple sliding time windows and gradually input into the initialization model to predict the first round of joint prediction vectors. In the first round, the joint prediction results of each time window provide feedback for subsequent parameter adjustments through residual calculation and optimization. Through continuous iteration of residual optimization and sliding update of time windows, the model realizes dynamic adjustment of key parameters, making the prediction results gradually more accurate.

[0144] In this embodiment, step S4 specifically includes:

[0145] S4-1, receive the stable difference vector of the data, define the current period and its lag period;

[0146] S4-2, for each lag period k, using the autocorrelation function to calculate the autocorrelation coefficient of the difference vector between the current time point and the k-lag time point;

[0147] S4-3, for each lag period k, calculate the partial autocorrelation coefficient using the partial autocorrelation function;

[0148] S4-4. Draw bar graphs according to the calculated autocorrelation coefficient and partial autocorrelation coefficient corresponding to the lag period k to generate the autocorrelation function graph and the partial autocorrelation function graph; wherein the horizontal axis represents the lag period k, and the vertical axis represents the autocorrelation coefficient and the partial autocorrelation coefficient, respectively.

[0149] In this embodiment, step S5 specifically includes:

[0150] S5-1. In the autocorrelation function diagram, find the lag period at which the autocorrelation coefficient drops rapidly from the significance value to zero, and determine it as the sliding average order q;

[0151] S5-2. In the partial autocorrelation function diagram, find the lag period at which the partial autocorrelation coefficient drops rapidly from the significance value to zero, and determine it as the autoregressive order p.

[0152] The calculation expression of the autocorrelation function in step S4-2 is:

[0153]

[0154] Among them, r k represents the autocorrelation coefficient when the lag period is k, y t represents the difference vector at the tth historical moment, represents the mean of the difference vector.

[0155] The autocorrelation function is used to measure the overall correlation between the difference vector at the current moment in the time series and the historical moments lagged by k, reflecting the lag structure of the entire series; it reveals the overall correlation between each lagged difference vector in the time series and the difference vector at the current moment, reflecting the periodicity or trend of the series.

[0156] The calculation expression of the partial autocorrelation function in step S4-3 is:

[0157]

[0158] Among them, φ k，k represents the partial autocorrelation coefficient when the lag period is k, φ j，k-1 It represents the autocorrelation coefficient when the lag period is j.

[0159] The partial autocorrelation function is used to measure the direct correlation between the differential vector at the current moment and the historical moments lagged by k, eliminating the influence of other lagged moments.

[0160] In this embodiment, step S7 specifically includes:

[0161] S7-1, selecting a time window from the t-w+1th historical moment to the tth historical moment, and calculating a joint prediction vector of the difference vector in the time window;

[0162] S7-2, after each prediction is completed, the time window slides forward by one unit of time, removes the earliest difference vector, and adds the current difference vector to form a new input window, and outputs several joint prediction vectors;

[0163] S7-3. Repeat the above steps until the joint prediction vector from the first historical moment to the Nd-th historical moment is predicted.

[0164] The modeling stage of the model uses the autocorrelation and partial autocorrelation analysis of historical data to set the model parameters. It includes: after obtaining a stable difference vector, the autocorrelation function (ACF) and partial autocorrelation function (PACF) are used to analyze the lag structure characteristics of the time series. Specifically, the autocorrelation function is used to measure the overall correlation between the difference vector at the current moment and the lag moment, thereby revealing the periodicity and trend of the time series; the partial autocorrelation function eliminates the influence of other lag moments and only measures the direct correlation between the difference vector at the current moment and the specific lag moment. By calculating the autocorrelation coefficient and partial autocorrelation coefficient of the lag period k, and drawing the ACF graph and PACF graph, the correlation characteristics of the time series are intuitively displayed.

[0165] The key parameters of the ARIMA model are further determined through the significant features in the ACF and PACF diagrams. Specifically, in the ACF diagram, the number of lags at which the autocorrelation coefficient drops rapidly from the significant value to zero is found, which is determined as the sliding average order q; in the PACF diagram, the number of lags at which the partial autocorrelation coefficient drops rapidly from the significant value to zero is found, which is determined as the autoregressive order p.

[0166] Furthermore, step S7-1 further includes:

[0167] S7-1-1, define the initial lagged difference vector and initial lagged residual at the first historical moment;

[0168] Specifically, the historical mean of the difference vector is defined as the lagged difference vector at the first historical moment; it is assumed that the lagged residual at the first historical moment is zero.

[0169] Since there is no true lagged difference vector before the first historical moment, it is a reasonable assumption to use the historical mean as the initial lagged difference vector; the residual represents the difference between the true difference vector and the joint prediction vector. At the first historical moment, since no prediction has been made, the actual residual cannot be obtained. Therefore, the lagged residual between the first historical moments can be assumed to be zero, providing a starting point for the prediction of the first historical moment, while ensuring that the model proceeds in an orderly manner from the second historical moment. Although this setting is somewhat hypothetical, it has a very limited impact on subsequent predictions because the model will continue to optimize parameters in subsequent rounds.

[0170] S7-1-2, using the initial lagged difference vector at the first historical moment as the input of the autoregressive sub-model to obtain the autoregressive prediction vector at the first historical moment;

[0171] The expression of the autoregressive prediction vector at the first historical moment is:

[0172]

[0173] in, represents the autoregressive prediction vector at the first historical moment, y lm Represents the historical mean of the lagged difference vector, which is used to replace the missing lagged difference vector;

[0174] φ1, φ2, …φ p is the coefficient of the autoregressive sub-model, indicating the weights of different lag moments;

[0175] S7-1-3, using the initial lagged residual at the first historical moment as the input of the sliding average sub-model to obtain the sliding average prediction vector at the first historical moment;

[0176] Among them, the expression of the sliding average prediction vector at the first historical moment is:

[0177]

[0178] in, represents the moving average forecast vector at the first historical moment;

[0179] μ is a constant term, which represents the predicted benchmark value of the material demand vector of the moving average sub-model;

[0180] θ1,θ2,…θ q is the coefficient of the moving average sub-model, indicating the weight of the lagged residual;

[0181] ∈ li It represents the lagged residual at the first historical moment, and its value is zero; it can be understood that there is no lagged residual before the first historical moment, because the lagged residual of the model needs to be calculated by the real difference vector and the joint prediction vector of the previous moment, and the model has not yet generated a joint prediction vector before the first historical moment, so the lagged residual is undefined before this moment. In order to ensure the logical coherence of the model calculation, it is a reasonable and necessary initialization strategy to assume that the lagged residual at the first historical moment is zero.

[0182] S7-1-4, adding the autoregressive prediction vector at the first historical moment and the sliding average prediction vector at the first historical moment to obtain a joint prediction vector at the first historical moment;

[0183] The expression of the joint prediction vector at the first historical moment is:

[0184] represents the joint prediction vector at the first historical moment;

[0185] In this embodiment, the autoregressive prediction vector and the moving average prediction vector are both the results of predictions for the same time point, so their dimensions are the same and can be directly added; the autoregressive sub-model and the moving average sub-model are two components of ARIMA, which respectively capture the influence of different information lag values ​​and residuals, and finally the overall joint prediction vector is obtained by the vector sum of the sub-models.

[0186] S7-1-5, calculating the current residual at the first historical moment according to the joint prediction vector at the first historical moment and the difference vector at the first historical moment, and defining it as the lagged residual at the second historical moment;

[0187] The expression of the current residual at the first historical moment is:

[0188] Among them, ∈1 represents the current residual at the first historical moment, and y1 represents the difference vector at the first historical moment;

[0189] S7-1-6, according to the joint prediction vector of the first historical moment and the initial lagged difference vector of the first historical moment, update the current lagged difference vector of the first historical moment, and define it as the lagged difference vector of the second historical moment;

[0190] S7-1-7, calculating a joint prediction vector at the second historical moment according to the lagged difference vector at the second historical moment and the lagged residual at the second historical moment;

[0191] S7-1-8, iteratively update the lagged difference vector and the lagged residual, repeat the steps S7-1-5 to S7-1-7, and slide the time window until all historical moments are traversed to complete the calculation of the joint prediction vector from the first historical moment to the Ndth;

[0192] The expression for updating the lagged difference vector is:

[0193] The expression for updating the lagged residual is:

[0194] in, represents the lagged difference vector at the current historical moment t, represents the lagged residual at the current historical moment t, y t-1 Represents the material demand vector at the previous historical moment, ∈ t-1 Represents the residual at the previous historical moment.

[0195] In this embodiment, by defining initial conditions and iterative calculation logic, a joint prediction vector for each historical moment is gradually generated, thereby realizing dynamic prediction of material demand in steel structure production.

[0196] The model uses the historical mean of the difference vector as the initial lagged difference vector at the first historical moment, assuming that the lagged residual is zero, to ensure the consistency of the forecast logic. At subsequent moments, the forecast vectors are calculated separately through the autoregressive and sliding average models, and they are synthesized into a joint forecast vector. By updating the lagged difference vector and lagged residual, combined with the sliding time window mechanism, the model predicts the material demand at all historical moments in turn.

[0197] In this embodiment, step S8 specifically includes:

[0198] S8-1, select a time window from the t-w+1th historical moment to the tth historical moment, and calculate the residuals of w difference vectors and the corresponding joint prediction vector in the time window;

[0199] S8-2, slide the time window forward by one unit time, remove the t-w+1th historical moment from the time window, and add the difference vector and the joint prediction vector at the t+1th historical moment;

[0200] S8-3. Continue to calculate the residuals in the sliding time window point by point until the residuals of Nd historical moments are calculated.

[0201] The expression of the residual of the first round is:

[0202]

[0203] ∈ t represents the residual at the tth historical moment in the first round, y t represents the difference vector at the tth historical moment, represents the joint prediction vector at the t-th historical moment; represents the autoregressive prediction vector output by the autoregressive sub-model, A moving average forecast vector representing the output of the moving average model.

[0204] In this embodiment, the residuals of the first round are dynamically calculated through a sliding time window mechanism. Specifically, the model calculates the residuals of the differential vector and the joint prediction vector contained in each time window point by point, and slides the window forward after the calculation is completed, removing the data of the earliest moment and adding the new historical moment data. Through this point-by-point sliding and calculation, the residual calculations from the first historical moment to the Ndth historical moment are finally covered. It effectively captures the local characteristics within the time window while ensuring global calculation coverage, providing complete residual data support for subsequent model parameter optimization, thereby improving the accuracy of material demand forecasting.

[0205] The above embodiments can be implemented in whole or in part by software, hardware, firmware or other arbitrary combinations. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0206] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0207] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division of a waterway underwater terrain change analysis system and method. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0208] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. Visual warehouse tracking management system based on steel structure production, characterized by: include: The original data acquisition module is used to obtain some original material demand data at the current time point; A feature processing module is used to perform feature processing on a number of original material demand data at the current time point to obtain a material demand vector at the current time point; The vector input module is used to input the material demand vector at the current time point into the pre-trained ARIMA parent model; The demand output module is used for the ARIMA parent model to remove the long-term trend fluctuations and short-term random disturbances in the time series through the internal sub-models, and output the original material demand forecast value within the time window defined by the user at the current time point; The modeling step of the ARIMA parent model includes constructing an initialization model; wherein the initialization model is composed of an autoregressive sub-model and a sliding average sub-model; After building the initial model, perform the following steps: The Nd data stationary difference vectors are divided into multiple time windows according to the time window length w defined by the user, and each time window is used as input and slid into the initialization model in sequence to predict the joint prediction vector of each historical moment in the first round point by point; According to the historical moments in each time window in the first round, the prediction vector and the true difference vector are combined to calculate the residual in the time window point by point. After the calculation is completed, the time window is slid to update to the next batch of historical moments and continue to calculate the residual. Substitute the residuals of the first round into the parameter optimization function to iteratively update the model parameters of the second round; The parameter optimization function is: ; in, represents the parameter optimization function; is the autoregressive coefficient of the autoregressive sub-model, which indicates the influence of the lagged difference vector in the time series on the autoregressive prediction vector at the current moment; is the sliding average coefficient of the sliding average sub-model, which represents the influence of the lagged residual on the sliding average prediction vector at the current moment; is a constant term, which represents the predicted benchmark value of the material demand vector of the moving average sub-model; is the global variance of the residual, indicating that the residual follows a normal distribution; is the residual, which represents the prediction error between the true difference vector and the joint prediction vector; Using the model parameters of the second round, the material demand vector within the user-defined time window is rolled over, and the joint forecast vector is recalculated for each historical moment up to the Ndth historical moment; Iteratively optimize the model parameters of subsequent rounds until the update amplitude of the model parameters is less than the set change threshold; The output of the original material demand forecast value includes: A-1. Initialize the reverse differential parameters; The reverse difference parameters include: the number of differences d and Nd joint prediction vectors; A-2. For each joint forecast vector, perform reverse difference to restore it to the material demand forecast value of the original scale; The expression of the material requirement vector in the original scale is: ; in, Represents the material demand forecast value at the original scale after restoration. represents the joint prediction vector at the t-th historical moment, represents the joint prediction vector of the previous historical moment; A-3. If the number of differential operations d is greater than 1, continue to perform reverse differential operations until d reverse differential operations are completed, and output the material demand forecast value within the user-defined time window.

2. According to the visual warehousing tracking management system based on steel structure production according to claim 1, the modeling steps of the ARIMA parent model include: S1. On the time axis, obtain a number of original material demand data at N historical moments according to a continuous time series, and adjust the data length of the original material demand data based on the time window defined by the user; S2. For each historical moment, perform feature processing on a number of original material demand data to obtain material demand vectors at N historical moments; wherein the material demand vectors contain numerical features and non-numerical features; S3, performing d times of difference processing on the numerical characteristics of the material demand vector at N historical moments within the user-defined time window until Nd data-stable difference vectors are generated; S4, performing autocorrelation analysis and partial autocorrelation analysis on the stable differential vector of the data, and generating an autocorrelation function graph and a partial autocorrelation function graph respectively; S5. According to the generated autocorrelation function graph and partial autocorrelation function graph, the lag periods of the truncated positions are defined as the autoregressive order p and the sliding average order q respectively; S6. Configure the parameters of the ARIMA model according to the number of differences d, the autoregressive order p and the sliding average order q, and build an initialization model.

3. The visual warehousing tracking and management system based on steel structure production according to claim 2 is characterized in that: Performing autocorrelation analysis and partial autocorrelation analysis on the stable difference vector of the data to generate an autocorrelation function graph and a partial autocorrelation function graph respectively includes: S4-1, receive the stable difference vector of the data, define the current period and its lag period; S4-2, for each lag period k, using the autocorrelation function to calculate the autocorrelation coefficient of the difference vector between the current time point and the k-lag time point; S4-3, for each lag period k, calculate the partial autocorrelation coefficient using the partial autocorrelation function; S4-4. Draw bar graphs according to the calculated autocorrelation coefficient and partial autocorrelation coefficient corresponding to the lag period k to generate the autocorrelation function graph and the partial autocorrelation function graph; wherein the horizontal axis represents the lag period k, and the vertical axis represents the autocorrelation coefficient and the partial autocorrelation coefficient, respectively.

4. The visual warehousing tracking and management system based on steel structure production according to claim 2 is characterized in that: According to the generated autocorrelation function graph and partial autocorrelation function graph, the lag period of the truncated position is defined as the autoregressive order p and the sliding average order q, including: S5-1. In the autocorrelation function diagram, find the lag period at which the autocorrelation coefficient drops rapidly from the significance value to zero, and determine it as the sliding average order q; S5-2. In the partial autocorrelation function diagram, find the lag period at which the partial autocorrelation coefficient drops rapidly from the significance value to zero, and determine it as the autoregressive order p.

5. The visual warehousing tracking management system based on steel structure production according to claim 4 is characterized in that: The calculation expression of the autocorrelation function is: ; in, represents the autocorrelation coefficient when the lag period is k, represents the difference vector at the tth historical moment, represents the mean of the difference vector.

6. The visual warehousing tracking and management system based on steel structure production according to claim 4 is characterized in that: The calculation expression of the partial autocorrelation function is: ; in, represents the partial autocorrelation coefficient when the lag period is k, It represents the autocorrelation coefficient when the lag period is j.

7. The visual warehousing tracking and management system based on steel structure production according to claim 4 is characterized in that: Predict the joint prediction vector for each historical moment in the first round point by point, including: S7-1, selecting a time window from the t-w+1th historical moment to the tth historical moment, and calculating a joint prediction vector of the difference vector in the time window; S7-2, after each prediction is completed, the time window slides forward by one unit of time, removes the earliest difference vector, and adds the current difference vector to form a new input window, and outputs several joint prediction vectors; S7-3. Repeat the above steps until the joint prediction vector from the first historical moment to the Nd-th historical moment is predicted.

8. The visual warehousing tracking and management system based on steel structure production according to claim 7 is characterized in that: Calculate the joint prediction vector of the difference vector in the time window, including: S7-1-1, define the initial lagged difference vector and initial lagged residual at the first historical moment; S7-1-2, using the initial lagged difference vector at the first historical moment as the input of the autoregressive sub-model to obtain the autoregressive prediction vector at the first historical moment; The expression of the autoregressive prediction vector at the first historical moment is: ; in, represents the autoregressive prediction vector at the first historical moment, represents the historical mean of the lagged difference vector; , , is the coefficient of the autoregressive sub-model, indicating the weights of different lag moments; S7-1-3, using the initial lagged residual at the first historical moment as the input of the sliding average sub-model to obtain the sliding average prediction vector at the first historical moment; Among them, the expression of the sliding average prediction vector at the first historical moment is: ; in, represents the moving average forecast vector at the first historical moment; is a constant term, which represents the predicted benchmark value of the material demand vector of the moving average sub-model; , , is the coefficient of the moving average sub-model, indicating the weight of the lagged residual; represents the lagged residual at the first historical moment, and its value is zero; S7-1-4, adding the autoregressive prediction vector at the first historical moment and the sliding average prediction vector at the first historical moment to obtain a joint prediction vector at the first historical moment; The expression of the joint prediction vector at the first historical moment is: ; represents the joint prediction vector at the first historical moment; S7-1-5, calculating the current residual at the first historical moment according to the joint prediction vector at the first historical moment and the difference vector at the first historical moment, and defining it as the lagged residual at the second historical moment; The expression of the current residual at the first historical moment is: ; in, represents the current residual at the first historical moment, represents the difference vector at the first historical moment; S7-1-6, according to the joint prediction vector of the first historical moment and the initial lagged difference vector of the first historical moment, update the current lagged difference vector of the first historical moment, and define it as the lagged difference vector of the second historical moment; S7-1-7, calculating a joint prediction vector at the second historical moment according to the lagged difference vector at the second historical moment and the lagged residual at the second historical moment; S7-1-8, iteratively update the lagged difference vector and the lagged residual, repeat the steps S7-1-5 to S7-1-7, and slide the time window until all historical moments are traversed to complete the calculation of the joint prediction vector from the first historical moment to the Ndth; The expression for updating the lagged difference vector is: ; The expression for updating the lagged residual is: ; in, represents the lagged difference vector at the current historical moment t, represents the lagged residual at the current historical moment t, represents the material demand vector at the previous historical moment, Represents the residual at the previous historical moment.

9. The visual warehousing tracking management system based on steel structure production according to claim 8 is characterized in that: Based on the historical moments in each time window in the first round, the prediction vector and the true difference vector are combined to calculate the residual in the time window point by point; After the calculation is completed, the time window is slid to update to the next batch of historical moments and continue to calculate the residuals, including: S8-1, selecting a time window from the t-w+1th historical moment to the tth historical moment, and calculating the residual of the differential vector and the corresponding joint prediction vector in the time window; S8-2, slide the time window forward by one unit time, remove the t-w+1th historical moment from the time window, and add the difference vector and the joint prediction vector at the t+1th historical moment; S8-3, continue to calculate the residuals in the sliding time window point by point until the residuals of Nd historical moments are calculated; The expression of the residual of the first round is: ； ； represents the residual at the tth historical moment in the first round, represents the difference vector at the tth historical moment, represents the joint prediction vector at the t-th historical moment; represents the autoregressive prediction vector output by the autoregressive sub-model, A moving average forecast vector representing the output of the moving average model.

Images