A vented oxygen recovery method and system based on a prediction model and a storage medium

Abstract

The application provides a kind of vented oxygen recovery method, system and storage medium based on prediction model, by wavelet packet transform decomposition historical oxygen demand data, obtain low frequency trend component and high frequency disturbance component and calculate high frequency component energy value;Two kinds of components are respectively constructed B spline function model, according to the energy value adjustment model regular term weight coefficient, superimposed prediction result generates future oxygen demand prediction curve;Again in combination with preset oxygen production plan, calculate the surplus amount prediction curve of vented oxygen;According to the energy value, determine the integral judgment period and surplus amount cumulative threshold value, with surplus amount integral value in period as the moment of recovery opportunity when exceeding threshold value;Determine recovery execution period, calculate the kurtosis coefficient of surplus amount in period and set the target pressure of oxygen storage equipment, in combination with surplus amount integral value in period, the difference between current and target pressure of equipment, calculate total recovery amount.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of recycling, and particularly relates to a vented oxygen recovery method and system based on a prediction model and a storage medium. BACKGROUND

[0002] Air separation devices are key equipment in the fields of metallurgy, chemical industry and other modern industries, and the core function is to separate air to produce high-purity industrial gases such as oxygen and nitrogen. In the production process, in order to ensure the economy and safety of operation, the air separation device is usually maintained in a relatively stable working condition, that is, the oxygen output remains constant within a certain period of time. The production rhythm of the downstream oxygen unit is intermittent and volatile, resulting in non-stationary characteristics of the actual oxygen demand, so that the stable output of the air separation device often does not match the demand of the user. When the oxygen output exceeds the instantaneous demand, the excess oxygen will be directly discharged into the atmosphere, causing serious energy waste and increasing the production cost per unit product. Moreover, the time series model in the oxygen demand prediction is difficult to reliably detect the superimposed characteristics of the long-term trend and high-frequency random disturbance in the demand data, resulting in insufficient prediction accuracy and unable to provide reliable basis for recovery decision. In the determination of recovery time and recovery amount, a pressure or flow threshold is usually used as the trigger condition, for example, when the pipe network pressure exceeds a certain set value, the recovery is started. It is unable to adjust the recovery strategy according to the degree and form characteristics of the oxygen demand fluctuation. When the demand fluctuation is gentle, the recovery opportunity may be missed due to improper threshold setting; when the demand fluctuation is severe, the recovery action may be delayed or the recovery amount may be determined unreasonably, so that the peak shaving and valley filling effect of the oxygen storage device cannot be fully played, and the energy saving effect is limited. SUMMARY

[0003] The present application provides a vented oxygen recovery method based on a prediction model, which solves the problem that the prior art is difficult to reliably detect the superimposed characteristics of the long-term trend and high-frequency random disturbance in the demand data, and is unable to adjust the recovery strategy according to the degree and form characteristics of the oxygen demand fluctuation, comprising the following steps:

[0004] The historical oxygen demand data is decomposed into a low-frequency trend component and a high-frequency disturbance component by wavelet packet transform, and the energy value of the high-frequency component is calculated;

[0005] For the low-frequency and high-frequency components, B-spline function models are respectively constructed, and the weight coefficient of the model regularization term is adjusted according to the energy value; the prediction results of each model are superimposed to obtain a future oxygen demand prediction curve; and a future vented oxygen surplus prediction curve is calculated according to the oxygen demand prediction curve and a preset oxygen production plan;

[0006] determining an integral judgment period and a surplus amount accumulation threshold according to the energy value; determining a current time as a recovery opportunity when an integral value of the surplus amount prediction curve in the integral judgment period exceeds the threshold value;

[0007] determining a recovery execution period based on the recovery opportunity; calculating a kurtosis coefficient of the surplus amount prediction curve in the recovery execution period, and determining a target pressure of the oxygen storage device according to the kurtosis coefficient; and calculating a total recovery amount according to an integral value of the surplus amount in the recovery execution period, and a difference between a current pressure of the oxygen storage device and the target pressure.

[0008] Optionally, the wavelet packet transform is used to decompose the historical oxygen demand data into a low-frequency trend component and a high-frequency disturbance component, and the energy value of the high-frequency component is calculated, including:

[0009] a preset wavelet basis is used to perform wavelet packet decomposition on the historical oxygen demand data to a preset number of layers;

[0010] the decomposed preset low-frequency node coefficients are reconstructed into a low-frequency trend component, and the remaining node coefficients are added after reconstruction to obtain a high-frequency disturbance component;

[0011] the sum of squares of the sampling point values of the high-frequency disturbance component is calculated to obtain the energy value .

[0012] Optionally, the B-spline function model is constructed for the low-frequency and high-frequency components respectively, and the weight coefficient of the model regularization term is adjusted according to the energy value, including:

[0013] the weight coefficient of the model regularization term of the model constructed for the low-frequency component is set to be a fixed preset value;

[0014] the weight coefficient of the model regularization term of the model constructed for the high-frequency component is set to be , specifically: ;

[0015] wherein, is a preset proportion coefficient, is the energy value.

[0016] Optionally, the future vented oxygen surplus amount prediction curve is calculated according to the oxygen demand prediction curve and a preset oxygen production plan, including:

[0017] the predicted value of each time on the future oxygen demand prediction curve is subtracted from the preset oxygen production plan to obtain a future vented oxygen surplus amount prediction value.

[0018] Optionally, the integral judgment period and the surplus amount accumulation threshold are determined according to the energy value, including:​

[0019] According to the energy value Determine the integral judgment period , Specifically: ;

[0020] According to the energy value Determine the surplus amount accumulation threshold , Specifically: ;

[0021] Wherein, And Is a preset proportion coefficient.

[0022] Optionally, the recovery execution period is determined based on the recovery opportunity, comprising:

[0023] From the currently determined recovery opportunity, search the surplus amount prediction curve backward, determine the time when the curve falls from positive value to zero value or negative value as the recovery end time, and the recovery execution period is the time period from the recovery opportunity to the recovery end time.

[0024] Optionally, the kurtosis coefficient of the surplus amount prediction curve in the recovery execution period is calculated, and the target pressure of the oxygen storage device is determined according to the kurtosis coefficient, comprising:

[0025] Between the lowest safe pressure And the highest safe pressure Of the oxygen storage device, set the target pressure , So that the target pressure And the kurtosis coefficient Of the surplus amount prediction curve in the recovery execution period is a monotonically increasing function.

[0026] Optionally, the total recovery amount is calculated according to the surplus amount integral value in the recovery execution period, and the difference between the current pressure of the oxygen storage device and the target pressure, comprising:

[0027] Calculate the integral value of the surplus amount prediction curve in the recovery execution period , Get the predicted total recoverable amount ;

[0028] According to the difference between the target pressure And the current pressure of the oxygen storage device , And according to the preset pressure-volume conversion coefficient C, calculate the oxygen storage device absorbable amount : ;

[0029] Take the predicted total recoverable amount And the oxygen storage device absorbable amount the minimum value in the total recovery amount .

[0030] In addition, the present application also relates to a vent oxygen recovery system based on a prediction model, comprising the following modules:

[0031] A first calculation module is configured to decompose historical oxygen demand data into a low-frequency trend component and a high-frequency disturbance component using wavelet packet transform, and calculate an energy value of the high-frequency component;

[0032] A second calculation module is configured to construct a B-spline function model for the low-frequency and high-frequency components respectively, and adjust a weight coefficient of a model regularization term according to the energy value; superimpose the prediction results of each model to obtain a future oxygen demand prediction curve; and calculate a future vent oxygen surplus amount prediction curve according to the oxygen demand prediction curve and a preset oxygen yield plan;

[0033] A determination module is configured to determine an integral judgment period and a surplus amount accumulation threshold value according to the energy value; and determine a recovery time when an integral value of the surplus amount prediction curve within the integral judgment period exceeds the threshold value;

[0034] A determination module is configured to determine a recovery execution period based on the recovery time; calculate a kurtosis coefficient of the surplus amount prediction curve within the recovery execution period, and determine a target pressure of an oxygen storage device according to the kurtosis coefficient; and calculate a total recovery amount according to an integral value of the surplus amount within the recovery execution period, and a difference between a current pressure of the oxygen storage device and the target pressure.

[0035] Preferably, the wavelet packet transform is used to decompose historical oxygen demand data into a low-frequency trend component and a high-frequency disturbance component, and calculate an energy value of the high-frequency component, comprising:

[0036] A preset wavelet basis is used to perform wavelet packet decomposition on historical oxygen demand data at a preset number of layers;

[0037] The decomposed preset low-frequency node coefficients are reconstructed into a low-frequency trend component, and the remaining node coefficients are added after reconstruction to obtain a high-frequency disturbance component;

[0038] The sum of squares of each sampling point value of the high-frequency disturbance component is calculated to obtain the energy value .

[0039] Preferably, the B-spline function model is constructed for the low-frequency and high-frequency components respectively, and the weight coefficient of the model regularization term is adjusted according to the energy value, comprising:

[0040] The weight coefficient of the model regularization term of the model constructed for the low-frequency component is set to a fixed preset value; .

[0041] a model constructed for a high frequency component, setting a weight coefficient of a regular term of the model , specifically: ;

[0042] wherein, is a preset proportional coefficient, is the energy value.

[0043] Preferably, the calculation of the future purge oxygen surplus amount prediction curve according to the oxygen demand prediction curve and the preset oxygen production plan comprises:

[0044] subtracting the predicted value of each time point on the future oxygen demand prediction curve from the preset oxygen production plan to obtain a future purge oxygen surplus amount prediction value.

[0045] Preferably, the determination of the integral judgment period and the surplus amount accumulation threshold according to the energy value comprises:

[0046] determining an integral judgment period according to the energy value , specifically: ;

[0047] determining a surplus amount accumulation threshold according to the energy value , specifically: ;

[0048] wherein, and are preset proportional coefficients.

[0049] Preferably, the determination of the recovery execution period based on the recovery timing comprises:

[0050] starting from the currently determined recovery timing, searching the surplus amount prediction curve backward, determining the time point when the curve drops from a positive value to a zero value or a negative value as the recovery end timing, and the recovery execution period is the time period from the recovery timing to the recovery end timing.

[0051] Preferably, the calculation of the kurtosis coefficient of the surplus amount prediction curve in the recovery execution period and the determination of the target pressure of the oxygen storage device according to the kurtosis coefficient comprise:

[0052] setting the target pressure between the lowest safe pressure and the highest safe pressure allowed by the oxygen storage device, so that the target pressure and the kurtosis coefficient of the surplus amount prediction curve in the recovery execution period ​​The functional relationship is monotonically increasing.

[0053] Preferably, the step of calculating the total recovery amount based on the integral value of the surplus amount within the recovery execution cycle and the difference between the current pressure of the oxygen storage device and the target pressure includes:

[0054] Calculate the surplus prediction curve during the recovery execution cycle. The integral value within the range yields the predicted total recyclable amount. ;

[0055] According to the target pressure Current pressure of oxygen storage equipment The difference is used to calculate the oxygen storage device's absorbable capacity based on the preset pressure-volume conversion coefficient C. : ;

[0056] Take the predicted total amount of recyclables With the oxygen storage device's absorbable capacity The minimum value in the range is taken as the total recovery amount. .

[0057] This invention decomposes historical oxygen demand data into low-frequency trends and high-frequency disturbances, constructs B-spline function models for the characteristics of each component, and constrains the models according to the severity of demand fluctuations, thus enabling the prediction of future oxygen demand. Based on the predicted cumulative surplus oxygen within a time window, recovery is initiated only when the cumulative amount reaches a reasonable level determined by the characteristics of demand fluctuations, thereby identifying the time window with true recovery value. Further analysis of the peak shape characteristics of the surplus prediction curve allows for the setting of a more scientific target pressure for oxygen storage equipment. By utilizing the total surplus within the recovery cycle and the available space of the oxygen storage equipment, the total amount to be recovered is accurately calculated. This invention makes recovery decisions more comprehensive, improves the efficiency and effectiveness of vented oxygen recovery, maximizes the buffering capacity of oxygen storage equipment, reduces energy waste, and lowers production costs. Attached Figure Description

[0058] Figure 1 A flowchart of the first embodiment;

[0059] Figure 2 This is a schematic diagram of wavelet packet decomposition.

[0060] Figure 3 A schematic diagram for generating the oxygen demand forecast curve;

[0061] Figure 4 A schematic diagram for determining the timing of recycling;

[0062] Figure 5A determination diagram for recovery execution cycle;

[0063] Figure 6 A determination diagram for total recovery amount. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0065] In a first embodiment, the present application provides a blow-off oxygen recovery method based on a prediction model, as shown in Figure 1 , comprising the following steps:

[0066] S1, using wavelet packet transform to decompose historical oxygen demand data into a low-frequency trend component and a high-frequency disturbance component, and calculating an energy value of the high-frequency component;

[0067] Specifically, historical oxygen demand data of at least one past production cycle of the air separation device is obtained to form a time series; a wavelet basis such as db4 is selected, and the time series is decomposed by three layers of wavelet packet, as shown in Figure 2 , to obtain eight frequency band decomposition coefficients; the coefficients at the lowest frequency band, i.e., node three zero, are reconstructed to obtain a low-frequency trend component; the coefficients of the remaining seven frequency bands are reconstructed and superimposed to obtain a high-frequency disturbance component; the sum of squares of all data points in the high-frequency disturbance component time series is calculated to obtain an energy value.

[0068] In an optional embodiment, the wavelet packet transform is used to decompose the historical oxygen demand data into a low-frequency trend component and a high-frequency disturbance component, and the energy value of the high-frequency component is calculated, comprising:

[0069] A preset wavelet basis is used to perform wavelet packet decomposition on historical oxygen demand data with a preset number of layers;

[0070] The decomposed preset low-frequency node coefficients are reconstructed into a low-frequency trend component, and the remaining node coefficients are reconstructed and added to obtain a high-frequency disturbance component;

[0071] The sum of squares of each sample point value of the high-frequency disturbance component is calculated to obtain the energy value .

[0072] The historical oxygen demand data, such as oxygen consumption per minute in the past 24 hours, forms a time series containing 1440 data points. A specific wavelet basis function, such as Daubechies4 wavelet, is selected, and the decomposition layer is set to 3. The 1440 data points are decomposed into 8 different frequency bands by 3-layer wavelet packet decomposition using the wavelet basis, and each frequency band corresponds to a node at the end of the decomposition tree, from node (3, 0) to node (3, 7) in ascending order of frequency.

[0073] After decomposition, the reconstruction phase is entered. Node (3, 0) represents the lowest frequency signal, reflecting the smooth trend of oxygen demand with production shifts. Therefore, the coefficients of node (3, 0) are reconstructed separately to obtain the low-frequency trend component. The coefficients of the remaining seven nodes, i.e. (3, 1) to (3, 7), are reconstructed respectively, and the reconstructed signals are added point by point to combine into a single time series, which is the high-frequency disturbance component, representing short and rapid demand fluctuations in the production process. In order to represent the degree of intensity of the fluctuations, the square of each data point in the high-frequency disturbance component sequence is calculated, and all the square values are added together, and the sum is the energy value In an embodiment, the energy value is considered as a dimensionless value, and only the numerical value of the energy value is taken in subsequent calculations of the regularization term weight coefficient, the integral judgment period, and the surplus accumulation threshold. The regularization term weight coefficient, the integral judgment period, and the surplus accumulation threshold are only related to the size of the energy value, and the coefficient 、 、 The unit of the integral judgment period and the surplus accumulation threshold is set. This is common knowledge in the art and will not be repeated here.

[0074] S2, for the low-frequency and high-frequency components, respectively, construct B-spline function models, and adjust the weight coefficient of the model regularization term according to the energy value; superimpose the prediction results of each model to obtain a future oxygen demand prediction curve; according to the oxygen demand prediction curve and the preset oxygen production plan, calculate a future vent oxygen surplus prediction curve;

[0075] Specifically, for the low-frequency component and the high-frequency component, a target function is established respectively, which includes a data fitting error term and a model smoothness regularization term; the greater the energy value is, the more intense the demand fluctuation is, and then a smaller weight coefficient is given to the regularization term of the two models to reduce the penalty on the model complexity, so that the model can flexibly capture the details of the intense changes in the data; on the contrary, if the energy value is smaller, it indicates that the data tends to be stable or only contains slight noise, and then the weight coefficient is increased to enhance the smoothness of the model to prevent the model from overfitting to the slight noise; the coefficients of the B-spline basis functions are determined by solving an optimization problem to establish a prediction model; a time point in the future is input into the low-frequency and high-frequency models established above, and the prediction value sequences of the respective models are obtained by adding the two sequences point by point, that is, the future oxygen demand prediction curve is obtained, as shown in Figure 3 . The constant oxygen production planning value in the future period of the air separation device is obtained, for example, 10,000 cubic meters per hour; the production planning value is subtracted from the predicted demand at the corresponding time point on the future oxygen demand prediction curve obtained above to obtain a series of oxygen difference values at future time points; the time points at which the oxygen difference value is positive are connected to obtain a future vent oxygen surplus prediction curve.

[0076] In an optional embodiment, the B-spline function model is constructed for the low-frequency and high-frequency components respectively, and the weight coefficient of the model regularization term is adjusted according to the energy value, including:

[0077] The weight coefficient of the model regularization term is set to a fixed preset value for the model constructed for the low-frequency component;

[0078] The weight coefficient of the model regularization term is set to for the model constructed for the high-frequency component, specifically: ;

[0079] wherein is a preset proportion coefficient, and is the energy value.

[0080] In the B-spline function model, the regularization term (integral of the second derivative of the curve) is used to punish the roughness or geometric curvature of the curve to prevent the curve from appearing oscillation in order to deliberately pass each noise point; and the role of the weight coefficient of the regularization term is to adjust the trade-off between fitting accuracy and curve smoothness, the greater the coefficient is, the heavier the punishment for the curvature is, and the curve is forced to be smooth and rigid to filter out noise; the smaller the coefficient is, the higher the tolerance for the curvature is, and the curve is allowed to have more turns to closely fit the fluctuations of the data. The prediction model function is constructed by linear combination of the B-spline basis functions, and a target function is defined, which includes the sum of squares of the data fitting residuals and the integral of the second derivative of the curve, that is, the smoothness penalty term; the weight coefficient calculated according to the energy value in the above step is The objective function of the low-frequency and high-frequency component model is substituted respectively, and the linear equations about the B-spline basis function coefficients are solved by minimizing the objective function to determine the control coefficients of the best fitting curve; the B-spline function analytical expression with the determined coefficients is input to the future time point for extrapolation calculation, and the future prediction sequences of the low-frequency trend and high-frequency disturbance components are output respectively.

[0081] In an alternative embodiment, two independent long short-term memory network structures are used, one for predicting the low-frequency trend component and the other for predicting the high-frequency disturbance component. The regularization term (L2 norm of the weight) constrains the numerical value of the internal connection weight of the neural network, limits the complexity of the model, and prevents network overfitting; and the regularization term weight coefficient controls the suppression strength of the model complexity: the larger the coefficient, the stronger the generalization ability. In the model training process, in order to prevent overfitting, a regularization term is input, and the influence of the regularization term is controlled by the weight coefficient. For the low-frequency trend component, the change is gentle and regular, so a fixed small regularization term weight coefficient , for example 0.01, is set for this model. The long short-term memory network model is specifically composed of an input layer, two stacked long short-term memory layers containing 64 neurons, a dropout layer for preventing overfitting, and a fully connected output layer. The training set is composed of the wavelet packet decomposition results of the historical oxygen demand data. Specifically, the training set of the low-frequency model is the historical low-frequency trend component time series, and the training set of the high-frequency model is the historical high-frequency disturbance component time series. The data is constructed into input-output pairs by the sliding window method, for example, the past 60 minutes of data is used as the input sequence, and the future 15 minutes of data is used as the target output sequence. The input of the model is a time series data with a length of 60. In the training process, the mean square error is used as the loss function, and the Adam optimizer is used to minimize the error between the predicted value and the true value, and the network weight is updated through multiple iterations. After training is completed, the output of the model is a prediction sequence with a length of 15, representing the prediction value of the corresponding component in the next 15 minutes.

[0082] For the high-frequency disturbance component, the randomness and complexity are high, and the regularization term weight needs to be adjusted according to the volatility, so as to improve the response ability of the model to short-term changes when the demand fluctuates violently, thereby avoiding the waste of empty oxygen caused by prediction lag. The above calculated energy value is used for adjustment. A proportion coefficient is set, for example 50. When the historical data shows that the oxygen demand fluctuates violently, the calculated value will be larger, assuming 1000, then the regularization term weight of the high-frequency model is 0.05, a smaller weight allows the model to learn more complex fluctuation patterns. Conversely, when the demand is stable, The smaller the value, the smaller the value, assuming 100, 0.5, a larger weight will enhance the penalty of model complexity, prevent the model from learning the noise in the data, and improve the generalization ability of prediction.

[0083] In an optional embodiment, the future purge oxygen surplus prediction curve is calculated according to the oxygen demand prediction curve and the preset oxygen production plan, comprising:

[0084] Subtract the predicted value of each time point on the future oxygen demand prediction curve from the preset oxygen production plan to obtain the future purge oxygen surplus prediction value.

[0085] Add the prediction results of the low-frequency model and the high-frequency model for a future period of time at each time point to obtain the future oxygen demand prediction curve. For example, predict the oxygen demand for the next 60 minutes to obtain a sequence of prediction values for each minute, such as 1050, 1075, 1030,..., units of cubic meters per hour. Obtain the oxygen production plan for the same time period. The plan can be a constant value or a sequence that changes over time.

[0086] Assuming the oxygen production plan is constant at 1100 cubic meters per hour. At each time point, subtract the predicted demand from the planned production. For the above example, the first three minutes of purge oxygen surplus prediction values are: 50 for the first minute, 25 for the second minute, and 70 for the third minute. Arrange the 60 calculation results in chronological order to form a future purge oxygen surplus prediction curve. The values on the curve are positive, indicating that there is excess oxygen, which can be recovered; the values are negative, indicating that the oxygen supply is insufficient.

[0087] S3, according to the energy value, determine the integral judgment period and the surplus accumulation threshold value; when the integral value of the surplus prediction curve in the integral judgment period exceeds the threshold value, determine the current time as the recovery opportunity;

[0088] Specifically, an inverse proportional function relationship between the integral judgment period, the surplus accumulation threshold value and the energy value is established; when the calculated energy value is high, it indicates that the demand fluctuates frequently, then set a shorter integral judgment period and a lower accumulation threshold value to facilitate detection of short-term recovery opportunities; when the energy value is low, then set a longer integral judgment period and a higher accumulation threshold value to wait for a larger recovery window; from the current time, the integral value of the surplus prediction curve in the determined integral judgment period is calculated, if the integral result is greater than the determined accumulation threshold value, then determine the current time as the starting point of the recovery opportunity, such as Figure 4 .

[0089] To set criteria for triggering recovery operations based on the volatility of historical data, in one optional embodiment, determining the integration judgment period and surplus accumulation threshold based on the energy value includes:

[0090] According to the energy value Determine the integral judgment period Specifically: ;

[0091] According to the energy value Determine the cumulative threshold for surplus Specifically: ;

[0092] in, and This is a preset scaling factor.

[0093] Utilizing energy value Determine the integral judgment period This is used to determine if there is a sufficient time window for recovery value. A proportional coefficient is set. For example, 60,000. If historical demand fluctuates greatly, If the value is high, such as 1200, then A shorter window of 50 minutes is used for judgment because long-term forecasts have lower reliability. If demand is stable, If the value is low, such as 300, then The timeframe is 200 minutes, based on a longer-term stable trend.

[0094] Utilizing energy value Determine the cumulative threshold for surplus This refers to the minimum accumulated surplus required to trigger recycling. A proportional coefficient is set. For example, 0.4. In cases of large demand fluctuations, assuming... If it is 1200, then With a capacity of 480 m³, a higher threshold can prevent frequent recycling activation due to short-lived spurious peaks. Assuming stable demand... If it is 300, then With a capacity of 120m³, the relatively low threshold makes the system more sensitive to small, stable surpluses, enabling timely recovery.

[0095] S4. Determine the recovery execution cycle based on the recovery timing; calculate the kurtosis coefficient of the surplus prediction curve within the recovery execution cycle, and determine the target pressure of the oxygen storage device based on the kurtosis coefficient; calculate the total recovery amount based on the surplus integral value within the recovery execution cycle and the difference between the current pressure of the oxygen storage device and the target pressure.

[0096] Specifically, from the determined start point of the recovery opportunity, a time period between two time points where the surplus amount prediction curve changes from positive to zero or negative is determined as a recovery execution period by searching backward along the time axis; a kurtosis coefficient of the surplus amount data sequence in the period is calculated; if the kurtosis coefficient is much greater than three, it indicates that the surplus amount presents a sharp peak shape, and a target pressure close to the upper limit of the safety of the tank is set; if the kurtosis coefficient is close to three or less, it indicates that the surplus amount shape is gentle, and a relatively low target pressure is set; according to the ideal gas state equation, the difference between the current pressure of the oxygen storage device and the target pressure is converted into the volume of the storable oxygen; the total surplus amount that can be theoretically recovered is obtained by integrating the surplus amount prediction curve in the entire recovery execution period; and the smaller of the two calculated amounts is taken as the determined total recovery amount.

[0097] In an optional embodiment, the recovery execution period is determined based on the recovery opportunity, including:

[0098] From the currently determined recovery opportunity, the time point where the surplus amount prediction curve decreases from a positive value to a zero value or a negative value is determined as an end time of the recovery, and the recovery execution period is a time period from the recovery opportunity to the end time of the recovery.

[0099] The future blow-off oxygen surplus amount prediction curve is continuously monitored. When it is found that the predicted cumulative surplus amount exceeds the surplus amount accumulation threshold for the first time in an integral judgment period , the current time is determined as the recovery opportunity, that is, the start time of the recovery execution period, such as Figure 5 . For example, at 10:00 am, it is judged that the trigger condition is met. Starting from 10:00, the future blow-off oxygen surplus amount prediction curve is viewed backward. Assuming that the prediction curve is 50, 65, 70, 80, 55, 40, 20, 10, -5, -15,... in the next 15 minutes. The values are checked minute by minute, and it is found that the values are positive from 10:00 to 10:07. At 10:08, the predicted value becomes -5, and the curve enters the negative value region for the first time from the positive value region. Therefore, 10:08 is determined as the end time of the recovery. The recovery execution period of this time is the interval from 10:00 to 10:08.

[0100] In an optional embodiment, the kurtosis coefficient of the surplus amount prediction curve in the recovery execution period is calculated, and the target pressure of the oxygen storage device is determined according to the kurtosis coefficient, including:

[0101] The target pressure is set between the lowest safe pressure and the highest safe pressure of the oxygen storage device , so that the target pressure the function relationship between the kurtosis coefficient of the surplus amount prediction curve in the recovery execution period is monotonically increasing.

[0102] After determining the recovery execution period, for example, from 10:00 to 10:08, the morphological characteristics of the surplus amount prediction curve in this period are analyzed, that is, the kurtosis coefficient of the set of surplus amount data 50, 65, 70, 80, 55, 40, 20, 10 is calculated . The kurtosis coefficient can represent the sharpness of the data distribution. If the surplus amount quickly reaches a peak and then falls, such as 10, 20, 100, 20, 10, the kurtosis coefficient will be high; if the surplus amount is evenly distributed within the period, such as 40, 45, 50, 45, 40, the kurtosis coefficient will be lower.

[0103] Setting the target pressure of the oxygen storage device according to the kurtosis coefficient . Assuming that the minimum safe pressure of the storage tank is 6 bar, and the maximum safe pressure is 12 bar. When the calculated kurtosis coefficient is high, it indicates that the surplus oxygen amount is concentrated and large, which needs to be quickly absorbed, so a higher target pressure is set, such as 11.5 bar. When the kurtosis coefficient is low, it indicates that the surplus oxygen amount is gentle and continuous, and a more moderate recovery strategy can be adopted, setting a lower target pressure, such as 8 bar, to ensure that the recovery strategy can match the generation speed of the surplus amount, optimizing recovery and ensuring the smoothness of the device operation.

[0104] In an optional embodiment, the total recovery amount is calculated according to the surplus amount integral value in the recovery execution period, and the difference between the current pressure and the target pressure of the oxygen storage device, including:

[0105] Calculating the integral value of the surplus amount prediction curve in the recovery execution period to obtain the predicted total recoverable amount :

[0106]

[0107] where S(t) is the surplus amount prediction curve;

[0108] According to the difference between the target pressure and the current pressure of the oxygen storage device, and according to the pre-set pressure-volume conversion coefficient C, the absorbable amount of the oxygen storage device is calculated : ;

[0109] Taking the predicted total recoverable amount and the absorbable amount of the oxygen storage device the smaller value of the predicted total recoverable amount and the oxygen storage device's current absorption capacity as the total recovery amount .

[0110] the total amount of oxygen that is theoretically recoverable within the determined recovery execution period. This is achieved by integrating the surplus amount prediction curve S(t) over this time period. For discrete data points this is simply a summation. For example, if the recovery period is 8 minutes and the surplus amount per minute is 50, 65, 70, 80, 55, 40, 20, 10, then the predicted total recoverable amount is the sum of these values, i.e. 390 m3. the current oxygen storage device's current absorption capacity. This depends on the current tank pressure and the set target pressure . Assuming the current pressure is 7 bar and the target pressure is 8 bar, and the pre-set pressure-volume conversion coefficient C is 200 cubic meters per bar, indicating that 200 m3of oxygen needs to be charged for every 1 bar increase in pressure, then the oxygen storage device's current absorption capacity is 200 m3. The predicted total recoverable amount is compared with the oxygen storage device's current absorption capacity, and the smaller value of the two is determined as the total recovery amount . In the above example, is 390 m3, is 200 m3, and thus the total recovery amount is 200 m3. This value ensures that it will not attempt to recover a non-existent surplus, and that it will not over-charge the tank, ensuring the reality and safety of the operation, as . Figure 6 In a second embodiment, the present application also provides a vented oxygen recovery system based on a prediction model, comprising the following modules:

[0111] a first calculation module for decomposing historical oxygen demand data into a low-frequency trend component and a high-frequency disturbance component using wavelet packet transform, and calculating the energy value of the high-frequency component;

[0112] a second calculation module for constructing a B-spline function model for the low-frequency and high-frequency components respectively, and adjusting the weight coefficient of the model regularization term according to the energy value; superimposing the prediction results of each model to obtain a future oxygen demand prediction curve; and calculating a future vented oxygen surplus amount prediction curve according to the oxygen demand prediction curve and a pre-set oxygen production plan;

[0113] a determination module for determining an integral judgment period and a surplus amount accumulation threshold value according to the energy value; and determining the current time as a recovery opportunity when the integral value of the surplus amount prediction curve within the integral judgment period exceeds the threshold value;

[0114] a first calculation module for decomposing historical oxygen demand data into a low-frequency trend component and a high-frequency disturbance component using wavelet packet transform, and calculating the energy value of the high-frequency component; ​​

[0115] determining module, configured to determine a recovery execution period based on the recovery occasion; calculate a kurtosis coefficient of the surplus amount prediction curve in the recovery execution period, and determine a target pressure of the oxygen storage device according to the kurtosis coefficient; calculate a total recovery amount according to an integral value of the surplus amount in the recovery execution period, and a difference between a current pressure of the oxygen storage device and the target pressure.

[0116] In this specification, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations exist in any actual relationship or order. Moreover, the terms "include", "contain" or any other variants mean to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or further includes elements inherent to such a process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article or device including the element. In this document, "a", "an", "said", "the" and "it" can also include plural forms, unless the context clearly indicates otherwise. A plurality means at least two, such as 2, 3, 5 or 8, etc. "And / or" includes any and all combinations of the related listed items.

[0117] The various embodiments in the specification are described in a progressive manner, each embodiment focusing on the differences from other embodiments, and the various embodiments can be combined as needed, and the same and similar parts refer to each other.

[0118] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for recovering vented oxygen based on a predictive model, characterized in that, Includes the following steps: The historical oxygen demand data is decomposed into low-frequency trend components and high-frequency disturbance components using wavelet packet transform, and the energy value of the high-frequency components is calculated. For the low-frequency and high-frequency components, B-spline function models are constructed respectively, and the weight coefficients of the model regularization term are adjusted according to the energy value; the prediction results of each model are superimposed to obtain the future oxygen demand prediction curve; based on the oxygen demand prediction curve and the preset oxygen production plan, the future oxygen surplus prediction curve is calculated. Based on the energy value, the integration judgment period and the surplus accumulation threshold are determined; when the integral value of the surplus prediction curve within the integration judgment period exceeds the threshold, the current moment is determined as the recovery opportunity. The recovery execution cycle is determined based on the recovery timing; the kurtosis coefficient of the surplus prediction curve within the recovery execution cycle is calculated, and the target pressure of the oxygen storage equipment is determined based on the kurtosis coefficient; the total recovery amount is calculated based on the surplus integral value within the recovery execution cycle and the difference between the current pressure of the oxygen storage equipment and the target pressure.

2. The method according to claim 1, characterized in that, The step of decomposing historical oxygen demand data into low-frequency trend components and high-frequency disturbance components using wavelet packet transform, and calculating the energy value of the high-frequency components, includes: A preset wavelet basis is used to perform wavelet packet decomposition of historical oxygen demand data at a preset number of levels; The pre-defined low-frequency node coefficients after decomposition are reconstructed into low-frequency trend components, and the remaining node coefficients are reconstructed and added together to obtain high-frequency disturbance components. The energy value is obtained by calculating the sum of squares of the values ​​at each sampling point of the high-frequency disturbance component. .

3. The method according to claim 1, characterized in that, The step of constructing B-spline function models for the low-frequency and high-frequency components respectively, and adjusting the weight coefficients of the model regularization term according to the energy value, includes: For the model constructed using low-frequency components, the weight coefficients of the model regularization term are set. It is a fixed preset value; For the model constructed from high-frequency components, the weight coefficients of the model regularization term are set. Specifically: ; in, This is a preset proportional coefficient. The energy value is mentioned above.

4. The method according to claim 1, characterized in that, The step of calculating the future oxygen surplus forecast curve based on the oxygen demand forecast curve and the preset oxygen production plan includes: The predicted value at each moment on the future oxygen demand forecast curve is subtracted from the preset oxygen production plan to obtain the predicted value of the future oxygen surplus.

5. The method according to claim 1, characterized in that, The step of determining the integration judgment period and the surplus accumulation threshold based on the energy value includes: According to the energy value Determine the integral judgment period Specifically: ; According to the energy value Determine the cumulative threshold for surplus Specifically: ; in, and This is a preset scaling factor.

6. The method according to claim 1, characterized in that, The determination of the recycling execution cycle based on the recycling timing includes: Starting from the currently determined recovery timing, the surplus prediction curve is searched backwards. The moment when the curve drops from a positive value to a zero or negative value is determined as the recovery end time. The recovery execution cycle is the time period from the recovery timing to the recovery end time.

7. The method according to claim 1, characterized in that, The calculation of the kurtosis coefficient of the surplus prediction curve during the recovery execution cycle, and the determination of the target pressure of the oxygen storage equipment based on the kurtosis coefficient, includes: At the minimum safe pressure allowed by the oxygen storage device With the highest safety pressure Between, set the target pressure This makes the target pressure The kurtosis coefficient of the surplus prediction curve during the recovery execution cycle The functional relationship is monotonically increasing.

8. The method according to claim 1 or 7, characterized in that, The total recovery amount is calculated based on the integral value of the surplus amount within the recovery execution cycle and the difference between the current pressure of the oxygen storage equipment and the target pressure, including: Calculate the surplus prediction curve during the recovery execution cycle. The integral value within the range yields the predicted total recyclable amount. ; According to the target pressure Current pressure of oxygen storage equipment The difference is used to calculate the oxygen storage device's absorbable capacity based on the preset pressure-volume conversion coefficient C. : ; Take the predicted total amount of recyclables With the oxygen storage device's absorbable capacity The minimum value in the range is taken as the total recovery amount. .

9. An vented oxygen recovery system based on a predictive model, characterized in that, Includes the following modules: The first calculation module is used to decompose historical oxygen demand data into low-frequency trend components and high-frequency disturbance components using wavelet packet transform, and to calculate the energy value of the high-frequency components. The second calculation module is used to construct B-spline function models for the low-frequency and high-frequency components respectively, and adjust the weight coefficients of the model regularization term according to the energy value; superimpose the prediction results of each model to obtain the future oxygen demand prediction curve; and calculate the future oxygen surplus prediction curve based on the oxygen demand prediction curve and the preset oxygen production plan. The determination module is used to determine the integration judgment period and the surplus accumulation threshold based on the energy value; when the integral value of the surplus prediction curve in the integration judgment period exceeds the threshold, the current moment is determined as the recovery opportunity; The determination module is used to determine the recovery execution cycle based on the recovery timing; calculate the kurtosis coefficient of the surplus prediction curve within the recovery execution cycle, and determine the target pressure of the oxygen storage device based on the kurtosis coefficient; and calculate the total recovery amount based on the surplus integral value within the recovery execution cycle and the difference between the current pressure of the oxygen storage device and the target pressure.

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

Images