A method for predicting energy consumption of hydraulic system of die forging equipment with self-learning function

Abstract

The disclosed embodiment provides a method for predicting energy consumption of a die forging equipment hydraulic system with a self-learning function, which belongs to the field of data processing technology. Specifically, the method includes: analyzing the energy consumption characteristics of the energy conversion unit of the die forging equipment hydraulic system to obtain the key energy consumption characteristics of the die forging equipment hydraulic system; extracting the energy consumption data set and preprocessing it, and extracting the energy consumption feature set corresponding to the data information in combination with the key energy consumption characteristics; based on the key energy consumption characteristics and the energy consumption feature set, using different energy consumption prediction fusion algorithms to construct energy consumption prediction sub-models for each energy conversion unit; optimizing each energy consumption prediction sub-model through incremental learning and transfer learning algorithms; evaluating each optimized energy consumption prediction sub-model through cross-validation and preset evaluation indicators; and using each optimized energy consumption prediction sub-model to predict energy consumption of the energy conversion unit of the die forging equipment hydraulic system. Through the disclosed solution, the prediction efficiency, accuracy and adaptability are improved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of data processing technology, and in particular to a method for predicting energy consumption of a hydraulic system of die forging equipment with a self-learning function. Background Art

[0002] Currently, die forging equipment is essential core equipment for forming important components in fields such as aerospace, shipbuilding, and transportation. However, its operation process involves enormous energy losses. With the advancement of digital technology and industrial supply levels, and driven by scientific and technological innovation, the manufacturing industry is gradually developing towards high-end, intelligent, and green development. Current die forging equipment design guidelines are mainly focused on ensuring the reliability and serviceability of the forming process, and do not take into account the energy dissipation during equipment operation. This has also led to the low energy utilization rate of current die forging equipment. Die forging equipment systems are composed of many components and the energy conversion process is complex. In order to effectively manage and control its energy consumption, the first and most urgent problem to be solved is to accurately predict the energy losses of each core energy-consuming unit of the die forging equipment. Through energy consumption prediction models, the factors that cause high energy consumption during different working stages of its service life can be revealed, and the energy consumption status of the die forging equipment during service can be monitored in real time, thereby avoiding the occurrence of high-energy consumption operating conditions and achieving high-efficiency operation of the equipment.

[0003] Existing energy consumption prediction methods focus on predicting the total energy consumption of equipment, without taking into account the different energy consumption mechanisms and data distribution characteristics of different components, which makes the prediction results often lack physical interpretability. In addition, the prediction model does not take into account the problem of model prediction performance degradation caused by future changes in equipment operating status and working environment, resulting in the prediction accuracy being unable to meet the prediction requirements of the entire service life of the die forging equipment.

[0004] It can be seen that there is an urgent need for a method for predicting the energy consumption of the hydraulic system of die forging equipment with high prediction efficiency, accuracy and adaptability and self-learning function. Summary of the Invention

[0005] In view of this, an embodiment of the present disclosure provides a method for predicting energy consumption of a hydraulic system of die forging equipment with a self-learning function, which at least partially solves the problems of poor prediction efficiency, accuracy and adaptability in the prior art.

[0006] The present disclosure provides a method for predicting energy consumption of a hydraulic system of die forging equipment with a self-learning function, comprising:

[0007] Step 1: Analyze the energy consumption characteristics of the energy conversion unit of the die forging equipment hydraulic system, construct an energy consumption mechanism model, analyze the flow process of the energy die forging equipment hydraulic system and extract nonlinear complex links based on this model, and obtain the key energy consumption characteristics of the die forging equipment hydraulic system;

[0008] Step 2: Extract the energy consumption data set of each energy conversion unit based on the historical operation data of the hydraulic system of the die forging equipment and preprocess it to obtain a data information set, and extract the energy consumption feature set corresponding to the data information in combination with the key energy consumption characteristics;

[0009] Step 3: Based on the key energy consumption characteristics and energy consumption feature set, different energy consumption prediction fusion algorithms are used to construct energy consumption prediction sub-models for each energy conversion unit according to the different energy consumption characteristics of different energy consumption units;

[0010] Step 4: Collect real-time operating data of the die forging equipment hydraulic system, use a sliding window to store the data for a recent period of time, and optimize each energy consumption prediction sub-model based on this data through incremental learning and transfer learning algorithms;

[0011] Step 5: Evaluate each optimized energy consumption prediction sub-model through cross-validation and preset evaluation indicators. If the evaluation result does not meet the preset requirements, return to step 4 until the evaluation result meets the preset requirements;

[0012] Step 6: Use the optimized energy consumption prediction sub-models to predict the energy consumption of the energy conversion unit of the hydraulic system of the die forging equipment.

[0013] According to a specific implementation of the embodiment of the present disclosure, the energy conversion unit includes a servo motor, an electronically controlled plunger pump, and a control valve group;

[0014] The key energy consumption characteristics include the energy consumption characteristics of the servo motor, the energy consumption characteristics of the electronically controlled plunger pump and the energy consumption characteristics of the control valve group.

[0015] According to a specific implementation of the embodiment of the present disclosure, the steps of analyzing the flow process of the hydraulic system of the energy die forging equipment and extracting nonlinear complex links to obtain the key energy consumption characteristics of the hydraulic system of the die forging equipment specifically include:

[0016] Step 1.1: Calculate the energy consumption characteristics of the servo motor based on the servo motor's drive power loss, motor iron loss, and copper loss.

[0017] P Mloss =P SD +P Fe +P Cu

[0018]

[0019] E Mloss =∫P Min (1-η M )dt

[0020] Among them, P Mloss is the power loss of the motor, η M is the efficiency of the motor, EMloss is the energy consumption of the motor, P SD is the motor’s driving power loss, P Fe and P Cu They are the iron loss and copper loss of the motor, P Min is the power input to the motor, t is the time;

[0021] Step 1.2: Calculate the energy consumption characteristics of the electric piston pump based on mechanical efficiency and volumetric efficiency

[0022]

[0023] E Ploss =∫P Pin (1-η pv ·η pm )dt

[0024] Among them, η pm is the mechanical efficiency of the electronically controlled plunger pump, η pv is the volumetric efficiency of the electronically controlled plunger pump, E Ploss is the energy consumption of the electric piston pump, μ is the fluid dynamic viscosity, C v is the laminar drag coefficient, C f is the mechanical resistance coefficient, T s is the torque loss, Δp p is the pressure difference between the inlet and outlet of the electric control plunger pump, V max is the maximum displacement of the electric piston pump, C s is the laminar leakage coefficient, β is the ratio of the current working displacement of the electric control plunger pump to the maximum displacement, P Pin is the power input to the electronically controlled plunger pump;

[0025] Step 1.3, calculate the energy consumption characteristics of the control valve group based on the pressure loss of each valve in the control valve group

[0026]

[0027] Where Δp v is the pressure difference between the inlet and outlet of the valve, E Vloss is the total energy consumption of the control valve group, C q is the valve port flow coefficient, ρ is the density of the fluid, q v Flow rate of valve port, A v is the flow area of ​​the valve port, j is the valve number running during the forging process, and z is the number of valves used in the control valve group.

[0028] According to a specific implementation of the embodiment of the present disclosure, the step of extracting energy consumption data of each energy conversion unit based on historical operation data of the hydraulic system of the die forging equipment and preprocessing the data information includes:

[0029] Step 2.1: Perform stationarity analysis on the energy consumption dataset using the autocorrelation function:

[0030]

[0031] Where k is the time lag delay order, ACF(k) represents the value of ACF after k time intervals lag, yt is the energy consumption data value at time t, y y+k is the energy consumption data value of the lagged k time intervals, Cov(y t ,y y+k ) is the covariance between the two, σ z (t) is the standard deviation of energy consumption data at time t, σ z (t+k) is the standard deviation of energy consumption data lagged by k time intervals, μ is the overall mean, and E represents the expectation;

[0032] Step 2.2: Use the Z-score outlier detection method to calculate the Z-score value of each energy consumption data in the energy consumption data set, and remove the energy consumption data with a Z-score value greater than the threshold as an outlier. The expression of the Z-score value is:

[0033]

[0034] Where X is the value of the energy consumption dataset, μ is the mean of the dataset, and σ is the standard deviation of the dataset;

[0035] In step 2.3, Sigmoid normalization is used to normalize the energy consumption dataset after removing outliers to eliminate the dimension effect:

[0036]

[0037] Among them, a and b are constant parameters.

[0038] According to a specific implementation of the embodiment of the present disclosure, step 3 specifically includes:

[0039] In step 3.1, an autoencoder is used to extract features from the data of the servo motor and the electronically controlled plunger pump. The energy consumption feature set is mapped to a low-dimensional hidden representation through the encoder, and then the hidden representation is reconstructed into the energy consumption feature set through the decoder. The calculation relationship between the encoder and the decoder is as follows:

[0040] h=f(Wx+b)

[0041]

[0042] Among them, x is the energy consumption feature set, f and f' are nonlinear functions used for encoding and decoding, W and W' are weight matrices, b and b' are bias vectors, and the hidden layer neurons decode and output the reconstructed energy consumption feature set through the decoder. The energy consumption feature set obtained after feature extraction The feature sequence in is input to the informer algorithm as the input layer to obtain the energy consumption prediction sub-model of the servo motor and the electronically controlled plunger pump;

[0043] Step 3.2: Based on the key energy consumption characteristics and energy consumption feature set, the residual fusion mechanism data fusion modeling method is used to characterize the energy consumption generated by the control valve group, and the residual data based on the mechanism prediction and the actual energy consumption is obtained.

[0044] P′ res =f CGA (p v ,q v ,U v ,T o )

[0045] Among them, p v is the valve inlet pressure, U v is the electric control signal of the valve, T o is the oil temperature, f CGA It is a data-driven residual prediction model;

[0046] In step 3.3, the energy consumption mechanism model is used to output a preliminary energy consumption forecast value. After calculating the residual with the actual energy consumption value, the data-driven residual prediction model is used to predict the energy consumption residual value and compensate it into the energy consumption mechanism model to construct the energy consumption forecast sub-model corresponding to the control valve group:

[0047] ΔP′ pred =ΔP pred +P′ res

[0048] Among them, ΔP pred is the predicted value of the energy consumption mechanism model, ΔP′ pred It is the predicted value of the fusion model of mechanism and data.

[0049] According to a specific implementation of the embodiment of the present disclosure, step 4 specifically includes:

[0050] Step 4.1: Collect real-time operating data from the die forging equipment hydraulic system and use an offset detection algorithm to determine the data distribution changes of the real-time operating data. If the data distribution changes slightly and gradually, and the model error gradually increases, proceed to step 4.2. If the data distribution changes suddenly and the model performance drops sharply, proceed to step 4.3.

[0051] Step 4.2, define the loss function through the EWC incremental learning algorithm and optimize each energy consumption prediction sub-model accordingly. When the loss function is minimized, the optimized energy consumption prediction sub-model is obtained, where the expression of the loss function is:

[0052]

[0053] Among them, L B (θ) is the loss function under the new training data, λ is the regularization coefficient, F i is the Fisher information matrix, θ i are network parameters, is the optimal parameter obtained from the old data, and N is the number of samples;

[0054] Step 4.3: Use the transfer learning algorithm domain adversarial neural network to transfer each energy consumption prediction sub-model in the source domain to the target domain for energy consumption prediction. The source domain and target domain inputs are mapped to θ through the feature extraction layer. f Parameterize the new space, and the prediction layer maps this new space representation to a weighted θ y Parameterized output, then the prediction loss under the i-th task is converted into an optimization problem and each energy consumption prediction sub-model is updated accordingly, where the expression of the optimization problem is

[0055]

[0056] Among them, θ d is a parameterized domain classifier, is a logistic regressor that determines whether a given input comes from the source domain or the target domain.

[0057] According to a specific implementation of the embodiment of the present disclosure, step 6 specifically includes:

[0058] Deploy optimized energy consumption prediction sub-models on edge devices, build data storage and analysis modules on the cloud, and realize online update of energy consumption models through cloud-edge collaboration. The energy consumption prediction results of the energy conversion unit of the hydraulic system of die forging equipment will be output and displayed on a visual interface.

[0059] According to a specific implementation of the embodiment of the present disclosure, the preset evaluation indicators include root mean square error, mean absolute percentage error and R 2 coefficient;

[0060] The expression of the root mean square error is:

[0061]

[0062] Among them, y i is the actual energy consumption value of the i-th data point, is the predicted energy consumption value of the i-th data point, and n is the total amount of data;

[0063] The expression of the mean absolute percentage error is:

[0064]

[0065] The R 2 The expression of the coefficient is

[0066]

[0067] Here, SSR represents the regression sum of squares, and SST represents the total sum of squares.

[0068] The energy consumption prediction scheme of the die forging equipment hydraulic system with self-learning function in the embodiment of the present disclosure includes: step 1, analyzing the energy consumption characteristics of the energy conversion unit of the die forging equipment hydraulic system, constructing an energy consumption mechanism model, analyzing the flow process of the energy die forging equipment hydraulic system and extracting nonlinear complex links based on this, and obtaining the key energy consumption characteristics of the die forging equipment hydraulic system; step 2, extracting the energy consumption data set of each energy conversion unit according to the historical operation data of the die forging equipment hydraulic system and preprocessing it to obtain a data information set, and extracting the energy consumption feature set corresponding to the data information in combination with the key energy consumption characteristics; step 3, based on the key energy consumption characteristics and the energy consumption feature set, for different energy consumption units Different energy consumption characteristics are used, and different energy consumption prediction fusion algorithms are used to construct energy consumption prediction sub-models of various energy conversion units; Step 4, collect real-time operation data of the hydraulic system of the die forging equipment, use a sliding window to store the data in the most recent period, and optimize each energy consumption prediction sub-model based on this data through incremental learning and transfer learning algorithms; Step 5, evaluate each optimized energy consumption prediction sub-model through cross-validation and preset evaluation indicators, if the evaluation result does not meet the preset requirements, return to Step 4 until the evaluation result meets the preset requirements; Step 6, use each optimized energy consumption prediction sub-model to predict the energy consumption of the energy conversion unit of the hydraulic system of the die forging equipment.

[0069] The beneficial effects of the embodiments of the present disclosure are as follows: through the scheme of the present disclosure, the energy consumption mechanism of different energy consumption units of the hydraulic system of the die forging equipment is combined with the distribution characteristics of its historical operating data, and different mechanisms and data fusion modeling methods are adopted to ensure the model prediction accuracy while enhancing the physical interpretability of the model. At the same time, the introduction of incremental learning and transfer learning optimization algorithms enables the model to continuously realize self-optimization of the model according to the data flow of the equipment service cycle, eliminating the problem of prediction performance degradation caused by operating environment fluctuations and component performance attenuation, meeting the needs of real-time energy consumption prediction of the die forging equipment throughout its service cycle, and improving prediction efficiency, accuracy and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0071] Figure 1 A flow chart of a method for predicting energy consumption of a hydraulic system of die forging equipment with a self-learning function provided by an embodiment of the present disclosure;

[0072] Figure 2 A schematic diagram of a method for predicting energy consumption of a hydraulic system of die forging equipment with a self-learning function provided by an embodiment of the present disclosure;

[0073] Figure 3 A schematic diagram of the energy consumption mechanism analysis of a die forging equipment system provided in an embodiment of the present disclosure;

[0074] Figure 4 An energy consumption prediction algorithm framework based on dual-driven mechanism and data is provided in the embodiments of the present disclosure;

[0075] Figure 5 A flow chart of a model self-optimization algorithm based on incremental learning and transfer learning provided in an embodiment of the present disclosure;

[0076] Figure 6 A system architecture for real-time monitoring of energy consumption of die forging equipment is provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0077] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0078] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0079] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on this disclosure, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement an apparatus and / or practice a method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this apparatus and / or practice this method.

[0080] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The illustrations only show components related to the present disclosure and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0081] Additionally, in the following description, specific details are provided to provide a thorough understanding of the examples. However, one skilled in the art will appreciate that the aspects described can be practiced without these specific details.

[0082] An embodiment of the present disclosure provides a method for predicting energy consumption of a hydraulic system of die forging equipment with a self-learning function. The method can be applied to the energy consumption prediction process of die forging equipment in industrial scenarios.

[0083] See also Figure 1 , is a flow chart of a method for predicting energy consumption of a hydraulic system of a die forging equipment with a self-learning function provided by an embodiment of the present disclosure. Figure 1 and Figure 2 As shown, the method mainly includes the following steps:

[0084] Step 1: Analyze the energy consumption characteristics of the energy conversion unit of the die forging equipment hydraulic system, construct an energy consumption mechanism model, analyze the flow process of the energy die forging equipment hydraulic system and extract nonlinear complex links based on this model, and obtain the key energy consumption characteristics of the die forging equipment hydraulic system;

[0085] Optionally, the energy conversion unit includes a servo motor, an electronically controlled plunger pump, and a control valve group;

[0086] The key energy consumption characteristics include the energy consumption characteristics of the servo motor, the energy consumption characteristics of the electronically controlled plunger pump and the energy consumption characteristics of the control valve group.

[0087] Furthermore, the steps of analyzing the flow process of the hydraulic system of the energy die forging equipment and extracting nonlinear complex links to obtain the key energy consumption characteristics of the hydraulic system of the die forging equipment specifically include:

[0088] Step 1.1: Calculate the energy consumption characteristics of the servo motor based on the servo motor's drive power loss, motor iron loss, and copper loss.

[0089] P Mloss =P SD +P Fe +P Cu

[0090]

[0091] E Mloss =∫P Min (1-η M )dt

[0092] Among them, P Mloss is the power loss of the motor, η M is the efficiency of the motor, E Mloss is the energy consumption of the motor, P SD is the motor’s driving power loss, P Fe and P Cu They are the iron loss and copper loss of the motor, P Min is the power input to the motor, t is the time;

[0093] Step 1.2: Calculate the energy consumption characteristics of the electric piston pump based on mechanical efficiency and volumetric efficiency

[0094]

[0095]

[0096] E Ploss =∫P Pin (1-η pv ·η pm )dt

[0097] Among them, η pm is the mechanical efficiency of the electric piston pump, η pv is the volumetric efficiency of the electronically controlled plunger pump, E Ploss is the energy consumption of the electric piston pump, μ is the fluid dynamic viscosity, C v is the laminar drag coefficient, C f is the mechanical resistance coefficient, T s is the torque loss, Δp p is the pressure difference between the inlet and outlet of the electric control plunger pump, V max is the maximum displacement of the electric piston pump, C sis the laminar leakage coefficient, β is the ratio of the current working displacement of the electric control plunger pump to the maximum displacement, P Pin is the power input to the electronically controlled plunger pump;

[0098] Step 1.3, calculate the energy consumption characteristics of the control valve group based on the pressure loss of each valve in the control valve group

[0099]

[0100] Where Δp v is the pressure difference between the inlet and outlet of the valve, E Vloss is the total energy consumption of the control valve group, C q is the valve port flow coefficient, ρ is the density of the fluid, q v Flow rate of valve port, A v is the flow area of ​​the valve port, j is the valve number running during the forging process, and z is the number of valves used in the control valve group.

[0101] In the specific implementation, based on the energy dissipation principle of each core energy conversion unit of the hydraulic system of the die forging equipment, the system energy consumption mechanism model is constructed, and its energy flow process is analyzed according to the mechanism model. Figure 3 For example, the energy dissipation principles of the mechanical, electronic control, and hydraulic systems of die forging equipment were analyzed. The components that generate energy loss during the system energy flow include the motor, hydraulic pump, control valve group, hydraulic cylinder, and the loss caused by mechanical structure friction. The friction loss between the hydraulic cylinder and the structure is very small and can often be ignored. Therefore, the energy consumption unit is mainly composed of the motor, hydraulic pump, and control valve group. The energy consumption of the motor mainly includes drive power loss, motor iron loss, and copper loss. The energy consumption can be calculated from these losses to obtain the working efficiency of the motor. The energy consumption expression is:

[0102] P Mloss =P SD +P Fe +P Cu

[0103]

[0104] E Mloss =∫P Min (1-η M )dt

[0105] Among them, P Mloss is the power loss of the motor, η M is the efficiency of the motor, E Mloss is the energy consumption of the motor, P SD is the motor’s driving power loss, P Fe and P Cu They are the iron loss and copper loss of the motor, P Minis the power input to the motor, and t is the time.

[0106] The energy consumption of a plunger pump comes from mechanical losses caused by friction due to the movement of the pump's internal structural components. It also comes from volumetric losses caused by leakage from the pump's high-pressure chamber, the compressibility of the hydraulic oil, and pumping resistance. The above energy consumption can generally be expressed using mechanical efficiency and volumetric efficiency. The energy consumption characteristics of a pump can be expressed as:

[0107]

[0108] E Ploss =∫P Pin (1-η pv ·η pm )dt

[0109] Among them, η pm is the mechanical efficiency of the pump, η pv is the volumetric efficiency of the pump, E Ploss is the energy consumption of the pump, μ is the dynamic viscosity of the fluid, C v is the laminar drag coefficient, C f is the mechanical resistance coefficient, T s is the torque loss, Δp p is the pressure difference between the pump inlet and outlet, V max is the maximum displacement of the pump, C s is the laminar leakage coefficient, β is the ratio of the pump's current working displacement to its maximum displacement, P Pin is the power input to the pump.

[0110] There are many types and quantities of control valve groups in die forging equipment systems. For control valves such as cartridge valves, check valves, reversing valves, servo valves, and solenoid switching valves, energy loss is mainly reflected in the pressure loss at both ends of the valve. Therefore, during the operation of the press hydraulic system, the energy consumption generated by the entire control valve group can be expressed as:

[0111]

[0112] Where Δp v is the pressure difference between the inlet and outlet of the valve, E Vloss is the total energy consumption of the control valve group, C q is the valve port flow coefficient, ρ is the density of the fluid, q v Flow rate of valve port, A v is the flow area of ​​the valve port, i is the valve number running during the forging process, and z is the number of valves used in the control valve group.

[0113] Step 2: Extract the energy consumption data set of each energy conversion unit based on the historical operation data of the hydraulic system of the die forging equipment and preprocess it to obtain a data information set, and extract the energy consumption feature set corresponding to the data information in combination with the key energy consumption characteristics;

[0114] Based on the above embodiment, the step of extracting energy consumption data of each energy conversion unit based on historical operation data of the hydraulic system of the die forging equipment and preprocessing the data information includes:

[0115] Step 2.1: Perform stationarity analysis on the energy consumption dataset using the autocorrelation function:

[0116]

[0117] Where k is the time lag delay order, ACF(k) represents the value of ACF after k time intervals lag, yt is the energy consumption data value at time t, y y+k is the energy consumption data value of the lagged k time intervals, Cov(y t ,y y+k ) is the covariance between the two, σ z (t) is the standard deviation of energy consumption data at time t, σ z (t+k) is the standard deviation of energy consumption data lagged by k time intervals, k is the time lag delay order, μ is the overall mean, and E represents the expectation;

[0118] Step 2.2: Use the Z-score outlier detection method to calculate the Z-score value of each energy consumption data in the energy consumption data set, and remove the energy consumption data with a Z-score value greater than the threshold as an outlier. The expression of the Z-score value is:

[0119]

[0120] Where X is the value of the energy consumption dataset, μ is the mean of the dataset, and σ is the standard deviation of the dataset;

[0121] In step 2.3, Sigmoid normalization is used to normalize the energy consumption dataset after removing outliers to eliminate the dimension effect:

[0122]

[0123] Among them, a and b are constant parameters.

[0124] In the specific implementation, data processing and feature extraction are performed on the operating data of the hydraulic system of the die forging equipment. After the operating data of the equipment is collected, the data often has problems such as missing values, outliers, and duplicate values. Therefore, it is necessary to preprocess the data. The data processing methods used include stationarity analysis, outlier processing, and data normalization. Stationarity analysis is used to test whether the data has stable distribution characteristics and avoid the deviation of prediction results caused by non-stationary data. The stationarity analysis of time series data is performed through the autocorrelation function. The expression is as follows:

[0125]

[0126] Where k is the lag order, μ is the population mean, and E is the expectation. The stationarity of a sequence is determined by observing how quickly the autocorrelation coefficient decays as k increases. The autocorrelation coefficient of a stationary sequence decays rapidly to near zero and fluctuates randomly around zero.

[0127] The environmental factors during the operation of die forging equipment are complex. Due to factors such as vibration and impact, the data collected by the sensor often contains a lot of noise and abnormal data. The Z-score outlier detection method is used for unified outlier detection. The calculation formula is as follows:

[0128]

[0129] Where Z is the Z-score, X is the value of the dataset, μ is the mean of the dataset, and σ is the standard deviation of the dataset. If the absolute value of the Z-score is greater than a certain threshold, the data point is usually considered an outlier. Generally, the threshold is 2 or 3. All outliers in the collected data are removed, and the data is completed using linear interpolation of neighboring data.

[0130] Due to the difference in dimensions in the data, there will be a large gap in the values ​​of different data. In order to avoid the characteristic scale difference caused by this phenomenon, Sigmoid normalization is used to normalize the data to eliminate the dimension effect. The calculation formula is:

[0131]

[0132] Among them, a and b are parameters that need to be adjusted according to the specific problem.

[0133] Step 3: Based on the key energy consumption characteristics and energy consumption feature set, different energy consumption prediction fusion algorithms are used to construct energy consumption prediction sub-models for each energy conversion unit according to the different energy consumption characteristics of different energy consumption units;

[0134] Based on the above embodiment, step 3 specifically includes:

[0135] In step 3.1, an autoencoder is used to extract features from the data of the servo motor and the electronically controlled plunger pump. The energy consumption feature set is mapped to a low-dimensional hidden representation through the encoder, and then the hidden representation is reconstructed into the energy consumption feature set through the decoder. The calculation relationship between the encoder and the decoder is as follows:

[0136] h=f(Wx+b)

[0137]

[0138] Among them, x is the energy consumption feature set, f and f' are nonlinear functions used for encoding and decoding, W and W' are weight matrices, b and b' are bias vectors, and the hidden layer neurons decode and output the reconstructed energy consumption feature set through the decoder. The energy consumption feature set obtained after feature extraction The feature sequence in is input to the informer algorithm as the input layer to obtain the energy consumption prediction sub-model of the servo motor and the electronically controlled plunger pump;

[0139] Step 3.2: Based on the key energy consumption characteristics and energy consumption feature set, the residual fusion mechanism data fusion modeling method is used to characterize the energy consumption generated by the control valve group, and the residual data based on the mechanism prediction and the actual energy consumption is obtained.

[0140] P′ res =f CGA (p v ,q v ,U v ,T o )

[0141] Among them, p v is the valve inlet pressure, U v is the electric control signal of the valve, T o is the oil temperature, f CGA It is a data-driven residual prediction model;

[0142] In step 3.3, the energy consumption mechanism model is used to output a preliminary energy consumption forecast value. After calculating the residual with the actual energy consumption value, the data-driven residual prediction model is used to predict the energy consumption residual value and compensate it into the energy consumption mechanism model to construct the energy consumption forecast sub-model corresponding to the control valve group:

[0143] ΔP′ pred =ΔP pred +P′ res

[0144] Among them, ΔP pred is the predicted value of the energy consumption mechanism model, ΔP′ pred It is the predicted value of the fusion model of mechanism and data.

[0145] In specific implementation, the energy consumption prediction sub-model is constructed by integrating the mechanism and data of each energy transfer unit. Figure 4 Taking the fusion model construction process in [1] as an example for analysis, the two energy units, motors and hydraulic pumps, have fewer input variables, complex internal energy dissipation mechanisms, and many nonlinear parameters in the energy consumption model. It is often impossible to predict energy consumption by accurately setting their mechanism parameters. The feature fusion mechanism data fusion modeling method is more suitable for energy consumption prediction of motors and hydraulic pumps. First, an autoencoder is used to extract features from the data of the motor and hydraulic pump. The input data is mapped to a low-dimensional hidden representation through the encoder, and then the hidden representation is reconstructed into the original data through the decoder. The hidden representation is considered to be a valuable feature of the data. The calculation relationship between the encoder and decoder is as follows:

[0146] h=f(Wx+b)

[0147]

[0148] Among them, x is the original input data, f, f' are nonlinear functions used for encoding and decoding, W, W' are weight matrices, b, b' are bias vectors, and the hidden layer neurons are reconstructed by the decoder output. The feature sequence obtained after feature extraction is input into the informer algorithm as the input layer to predict the energy consumption of the motor and hydraulic pump.

[0149] For the valve components of die forging equipment, the energy consumption mechanism model for each valve component is relatively simple. However, the system contains a large number of valve components, making it difficult to obtain energy consumption data for each valve component before and after the data collection process. Therefore, by integrating the valve components into a control valve group and using the residual fusion mechanism data fusion modeling method based on the energy consumption characteristics of the valve group before and after the valve group, the energy consumption generated by the control valve group can be more accurately characterized. Based on the data structure analysis of the control valve group, the input feature categories are relatively small. Therefore, a convolutional neural network (CNN) layer is introduced to increase the dimensionality of the input features, blending low-dimensional features into high-dimensional features, thereby enhancing the expressive power of the existing features. It also converts noise and irrelevant information in the training data into useful features. The resulting features are then input to the GRU module for training. The attention mechanism is introduced to automatically learn and calculate the contribution of the input data to the output data. That is, when performing energy consumption prediction tasks, the model focuses on the energy consumption data that is more critical to the prediction results among the large number of input features, reducing the emphasis on other non-critical features. The final output is the residual data based on the mechanism prediction and the actual energy consumption. The prediction result is expressed as:

[0150] P′ res =f CGA (pv ,q v ,U v ,T o )

[0151] Among them, p v is the valve inlet pressure, U v is the electric control signal of the valve, T o is the oil temperature, P r ' es is the prediction target, i.e. the residual data between the mechanism prediction result and the actual energy consumption, f CGA Data-driven residual prediction model. The mechanism model is used to output a preliminary energy consumption prediction value. After calculating the residual with the actual energy consumption value, the data-driven residual prediction model is used to predict the energy consumption residual value and compensate it into the mechanism model to construct a comprehensive energy consumption prediction model. The expression is as follows:

[0152] ΔP′ pred =ΔP pred +P′ res

[0153] Among them, ΔP pred is the predicted value of the energy consumption mechanism model, ΔP′ pred The predicted value of the fusion model of mechanism and data is used to correct the errors caused by nonlinearity and uncertainty factors in the control valve group, ensuring the accuracy of the energy consumption of the entire control valve group.

[0154] Step 4: Collect real-time operating data of the die forging equipment hydraulic system, use a sliding window to store the data for a recent period of time, and optimize each energy consumption prediction sub-model based on this data through incremental learning and transfer learning algorithms;

[0155] Based on the above embodiment, step 4 specifically includes:

[0156] Step 4.1: Collect real-time operating data from the die forging equipment hydraulic system and use an offset detection algorithm to determine the data distribution changes of the real-time operating data. If the data distribution changes slightly and gradually, and the model error gradually increases, proceed to step 4.2. If the data distribution changes suddenly and the model performance drops sharply, proceed to step 4.3.

[0157] Step 4.2, define the loss function through the EWC incremental learning algorithm and optimize each energy consumption prediction sub-model accordingly. When the loss function is minimized, the optimized energy consumption prediction sub-model is obtained, where the expression of the loss function is:

[0158]

[0159] Among them, L B(θ) is the loss function under the new training data, λ is the regularization coefficient, F i is the Fisher information matrix, θ i are network parameters, is the optimal parameter obtained from the old data, and N is the number of samples;

[0160] Step 4.3: Use the transfer learning algorithm domain adversarial neural network to transfer each energy consumption prediction sub-model in the source domain to the target domain for energy consumption prediction. The source domain and target domain inputs are mapped to θ through the feature extraction layer. f Parameterize the new space, and the prediction layer maps this new space representation to a weighted θ y Parameterized output, then the prediction loss under the i-th task is converted into an optimization problem and each energy consumption prediction sub-model is updated accordingly, where the expression of the optimization problem is

[0161]

[0162] Among them, θ d is a parameterized domain classifier, is a logistic regressor that determines whether a given input comes from the source domain or the target domain.

[0163] In specific implementation, the energy consumption prediction model is optimized through incremental learning and transfer learning algorithms. Figure 5 The algorithm flow in the model is used to self-optimize the model. First, the offset detection algorithm is used to determine the changes in the data distribution of the real-time running data stream of the device. If there is a small gradual change or a gradual increase in the model error, incremental learning is used to adjust and optimize the model parameters to adapt to the slowly changing operating environment and keep the model updated in real time. The model is optimized using the EWC incremental learning algorithm. A weight value is assigned to each parameter to indicate the importance of the parameter to the old task. For each task, the penalty intensity of each parameter depends on the importance of the parameter to the task. For more important parameters, the penalty is higher. During the training process of the new task, the impact on the old task is controlled by modifying the weight. The loss function can be expressed as:

[0164]

[0165] Among them, L B (θ) is the loss function under the new training data, λ is the regularization coefficient, F i is the Fisher information matrix, θ i are network parameters, The optimal parameters obtained for the old data.

[0166] The Fisher information matrix used to measure the importance of network parameters is expressed as follows:

[0167]

[0168] Here, N is the number of samples. A larger value indicates less uncertainty in the network parameters and a greater importance for those parameters. This means the network parameters for the new task need to be closer to the optimal parameters obtained from the old task, ultimately retaining the important parameters learned by the model in the old task. The EWC incremental algorithm can find locations where the model's loss function is very small for both new and old data, allowing the model to retain important model parameters from the old data after learning the new data, thereby ensuring the model's prediction accuracy for consecutive tasks.

[0169] If the offset detection algorithm results show a sudden change in data distribution and a sharp drop in model performance, consider using a transfer learning algorithm to migrate the source domain model to the target domain for energy consumption prediction to meet the prediction needs in the new scenario. The transfer learning algorithm Domain Adversarial Neural Network (DANN) can combine domain adaptation and feature learning in a single training process. After training, the different domain features obtained are both discriminative and domain-invariant, allowing the learned domain-invariant features to be directly applied to the target domain. DANN primarily measures the distance between two domain distributions by calculating the H-divergence, which is calculated as follows:

[0170]

[0171] Where S and T are two different domain distributions, and I represents the discriminant function. The principle of DANN is to control the H-divergence by finding indistinguishable representations between the source domain and the target domain. Its loss function is also designed as a regularization factor that approximates the H-divergence. The source domain and target domain inputs are mapped to θ through the feature extraction layer. f Parameterize the new space, and the prediction layer maps this new space representation to a weighted θ y The parameterized output can therefore be transformed into the following optimization problem for the prediction loss under the i-th task:

[0172]

[0173]

[0174] Among them, θ d is a parameterized domain classifier, The logistic regressor is used to determine whether a given input comes from the source domain or the target domain. Domain-generative adversarial learning is used to obtain domain-invariant features across different datasets, thereby ensuring the prediction model's effectiveness in new scenarios.

[0175] Step 5: Evaluate each optimized energy consumption prediction sub-model through cross-validation and preset evaluation indicators. If the evaluation result does not meet the preset requirements, return to step 4 until the evaluation result meets the preset requirements;

[0176] Furthermore, the preset evaluation indicators include root mean square error, mean absolute percentage error and R 2 coefficient;

[0177] The expression of the root mean square error is:

[0178]

[0179] Among them, y i is the actual energy consumption value of the i-th data point, is the predicted energy consumption value of the i-th data point, and n is the total amount of data;

[0180] The expression of the mean absolute percentage error is:

[0181]

[0182] The R 2 The expression of the coefficient is

[0183]

[0184] Here, SSR represents the regression sum of squares, and SST represents the total sum of squares.

[0185] In the specific implementation, the prediction performance of the model is evaluated, and the effectiveness of the model is evaluated from two aspects: the accuracy of the prediction results and the stability of the prediction model. The evaluation indicators include root mean square error (RMSE), mean absolute percentage error (MAPE), R 2 Coefficient. RMSE is a commonly used forecast error metric that measures the square root of the average of the squares of the differences between the predicted value and the actual value. The smaller the RMSE value, the better the forecast performance of the model. Its specific calculation formula is as follows:

[0186]

[0187] where y i is the actual energy consumption value of the i-th data point, is the predicted energy consumption value of the i-th data point, and n is the total amount of data.

[0188] MAPE is a commonly used forecast accuracy evaluation indicator. It measures the average percentage error between the predicted value and the actual value. It helps to understand the degree to which the predicted value deviates from the actual value. The smaller the MAPE value, the smaller the deviation of the model's predicted value. The specific calculation formula is as follows:

[0189]

[0190] R 2 The coefficient, also known as the coefficient of determination or the square of the correlation coefficient, is an indicator used to evaluate the goodness of fit of a regression model and indicates the proportion of data variance that the model can explain. 2 The value of the coefficient is between 0 and 1. The closer it is to 1, the better the regression model fits the data. Conversely, the worse the model's ability to explain the data is. The specific calculation is mainly to calculate R by calculating the regression square (SSR) and the total sum of squares (SST). 2 coefficient:

[0191]

[0192] Step 6: Use the optimized energy consumption prediction sub-models to predict the energy consumption of the energy conversion unit of the hydraulic system of the die forging equipment.

[0193] Based on the above embodiment, step 6 specifically includes:

[0194] Deploy optimized energy consumption prediction sub-models on edge devices, build data storage and analysis modules on the cloud, and realize online update of energy consumption models through cloud-edge collaboration. The energy consumption prediction results of the energy conversion unit of the hydraulic system of die forging equipment will be output and displayed on a visual interface.

[0195] During the specific implementation, the development of the real-time energy consumption prediction system for die forging equipment was completed. Figure 6The system development architecture shown in the figure is used to develop a real-time energy consumption prediction system. Communication is established between the various sensors at the die forging equipment layer and the data acquisition equipment at the data layer. Data is collected to determine the operating speed and crossbar displacement of the die forging equipment's mechanical structure; the input current, input voltage, output torque, and output speed of the servo motor; the input torque, input speed, output flow, output pressure, and oil temperature of the hydraulic pump; and the upstream pressure, upstream flow, downstream pressure, and downstream flow of the control valve group. A cloud data storage platform is constructed to upload this data stored in the data acquisition equipment. Data storage and processing are implemented on the cloud data management platform. The established energy consumption prediction model is deployed at the model layer, using real-time data collected from the data platform to predict the equipment's energy consumption. The predicted energy consumption values ​​are then uploaded to the cloud platform for data storage. Simultaneously, the energy consumption results during equipment operation are displayed on a computer using a visual interface, facilitating real-time monitoring of equipment energy consumption and the formulation of system operation and maintenance strategies by decision-makers. Finally, the decision results are applied to the equipment layer to effectively manage equipment energy consumption.

[0196] The energy consumption prediction method for the hydraulic system of die forging equipment with self-learning function provided in this embodiment combines the energy consumption mechanism of different energy consumption units of the hydraulic system of die forging equipment with the distribution characteristics of its historical operation data, adopts different mechanism and data fusion modeling methods, ensures the model prediction accuracy while enhancing the physical interpretability of the model. At the same time, the introduction of incremental learning and transfer learning optimization algorithms enables the model to continuously realize self-optimization of the model according to the data stream of the equipment service cycle, eliminates the problem of prediction performance degradation caused by operating environment fluctuations and component performance degradation, meets the demand for real-time energy consumption prediction of die forging equipment throughout its service cycle, and improves prediction efficiency, accuracy and adaptability. Real-time prediction of the energy consumption of die forging equipment can effectively improve the energy management efficiency of the equipment, achieve energy saving and consumption reduction and intelligent production. Through real-time monitoring and data analysis, energy consumption trends can be predicted, which helps to detect anomalies in advance, optimize equipment operating parameters, avoid energy waste, not only reduce production costs, but also improve the stability and service life of equipment operation, and promote the transformation of the die forging manufacturing field to green manufacturing and intelligent manufacturing.

[0197] It should be understood that various parts of the present disclosure can be implemented in hardware, software, firmware, or a combination thereof.

[0198] The above description is merely a specific embodiment of the present disclosure, but the scope of protection of the present disclosure is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this disclosure should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure should be based on the scope of protection of the claims.

Claims

1. A method for predicting energy consumption of a hydraulic system of die forging equipment with a self-learning function, characterized in that: include: Step 1: Analyze the energy consumption characteristics of the energy conversion unit of the die forging equipment hydraulic system, construct an energy consumption mechanism model, analyze the flow process of the energy die forging equipment hydraulic system and extract nonlinear complex links based on this model, and obtain the key energy consumption characteristics of the die forging equipment hydraulic system; Step 2: Extract the energy consumption data set of each energy conversion unit based on the historical operation data of the hydraulic system of the die forging equipment and preprocess it to obtain a data information set, and extract the energy consumption feature set corresponding to the data information in combination with the key energy consumption characteristics; Step 3: Based on the key energy consumption characteristics and energy consumption feature set, different energy consumption prediction fusion algorithms are used to construct energy consumption prediction sub-models for each energy conversion unit according to the different energy consumption characteristics of different energy consumption units; Step 4: Collect real-time operating data of the die forging equipment hydraulic system, use a sliding window to store the real-time operating data for the most recent period, and determine the changes in its data distribution. If the data distribution changes slightly and gradually, and the model error gradually increases, then optimize each energy consumption prediction sub-model using an incremental learning algorithm. If the data distribution changes suddenly and the model performance drops sharply, then optimize each energy consumption prediction sub-model using a transfer learning algorithm. Step 5: Evaluate each optimized energy consumption prediction sub-model through cross-validation and preset evaluation indicators. If the evaluation result does not meet the preset requirements, return to step 4 until the evaluation result meets the preset requirements; Step 6: Use the optimized energy consumption prediction sub-models to predict the energy consumption of the energy conversion unit of the hydraulic system of the die forging equipment.

2. The method according to claim 1, characterized in that The energy conversion unit includes a servo motor, an electronically controlled plunger pump and a control valve group; The key energy consumption characteristics include the energy consumption characteristics of the servo motor, the energy consumption characteristics of the electronically controlled plunger pump and the energy consumption characteristics of the control valve group.

3. The method according to claim 2, characterized in that The steps of analyzing the flow process of the hydraulic system of the energy die forging equipment and extracting nonlinear complex links to obtain the key energy consumption characteristics of the hydraulic system of the die forging equipment specifically include: Step 1.1: Calculate the energy consumption characteristics of the servo motor based on the servo motor's drive power loss, motor iron loss, and copper loss. P Mloss =P SD +P Fe +P Cu E Mloss =∫P Min (1-η M )dt Among them, P Mloss is the power loss of the motor, η M is the efficiency of the motor, E Mloss is the energy consumption of the motor, P SD is the motor’s driving power loss, P Fe and P Cu They are the iron loss and copper loss of the motor, P Min is the power input to the motor, t is the time; Step 1.2: Calculate the energy consumption characteristics of the electric piston pump based on mechanical efficiency and volumetric efficiency E Ploss =∫P Pin (1-η pv ·η pm )dt Among them, η pm is the mechanical efficiency of the electronically controlled plunger pump, η pv is the volumetric efficiency of the electronically controlled plunger pump, E Ploss is the energy consumption of the electric piston pump, μ is the fluid dynamic viscosity, C v is the laminar drag coefficient, C f is the mechanical resistance coefficient, T s is the torque loss, Δp p is the pressure difference between the inlet and outlet of the electric control plunger pump, V max is the maximum displacement of the electric piston pump, C s is the laminar leakage coefficient, β is the ratio of the current working displacement of the electric control plunger pump to the maximum displacement, P Pin is the power input to the electronically controlled plunger pump; Step 1.3, calculate the energy consumption characteristics of the control valve group based on the pressure loss of each valve in the control valve group Where Δp v is the pressure difference between the inlet and outlet of the valve, E Vloss is the total energy consumption of the control valve group, C q is the valve port flow coefficient, ρ is the density of the fluid, q v Flow rate of valve port, A v is the flow area of ​​the valve port, j is the valve number running during the forging process, and z is the number of valves used in the control valve group.

4. The method according to claim 3, characterized in that The step of extracting energy consumption data of each energy conversion unit based on historical operation data of the hydraulic system of the die forging equipment and preprocessing the data information includes: Step 2.1: Perform stationarity analysis on the energy consumption dataset using the autocorrelation function: Where k is the time lag order, ACF(k) represents the value of ACF after k time intervals lag, and y t is the energy consumption data value at time t, y y+k is the energy consumption data value of the lagged k time intervals, Cov(y t ,y y+k ) is the covariance between the two, σ z (t) is the standard deviation of energy consumption data at time t, σ z (t+k) is the standard deviation of energy consumption data lagged by k time intervals, μ is the overall mean, and E represents the expectation; Step 2.2: Use the Z-score outlier detection method to calculate the Z-score value of each energy consumption data in the energy consumption data set, and remove the energy consumption data with a Z-score value greater than the threshold as an outlier. The expression of the Z-score value is: Where X is the value of the energy consumption dataset, μ is the mean of the dataset, and σ is the standard deviation of the dataset; In step 2.3, Sigmoid normalization is used to normalize the energy consumption dataset after removing outliers to eliminate the dimension effect: Among them, a and b are constant parameters.

5. The method according to claim 4, characterized in that The step 3 specifically includes: In step 3.1, an autoencoder is used to extract features from the data of the servo motor and the electronically controlled plunger pump. The energy consumption feature set is mapped to a low-dimensional hidden representation through the encoder, and then the hidden representation is reconstructed into the energy consumption feature set through the decoder. The calculation relationship between the encoder and the decoder is as follows: h=f(Wx+b) Among them, x is the energy consumption feature set, f and f' are nonlinear functions used for encoding and decoding, W and W' are weight matrices, b and b' are bias vectors, and the hidden layer neurons decode and output the reconstructed energy consumption feature set through the decoder. The energy consumption feature set obtained after feature extraction The feature sequence in is fed into the informer algorithm as the input layer to obtain the energy consumption prediction sub-model of the servo motor and the electronically controlled plunger pump; Step 3.2: Based on the key energy consumption characteristics and energy consumption feature set, the residual fusion mechanism data fusion modeling method is used to characterize the energy consumption generated by the control valve group, and the residual data based on the mechanism prediction and the actual energy consumption is obtained. P’ res =f CGA (p v ,q v ,U v ,T o ) Among them, p v is the valve inlet pressure, U v is the electric control signal of the valve, T o is the oil temperature, f CGA It is a data-driven residual prediction model; In step 3.3, the energy consumption mechanism model is used to output a preliminary energy consumption forecast value. After calculating the residual with the actual energy consumption value, the data-driven residual prediction model is used to predict the energy consumption residual value and compensate it into the energy consumption mechanism model to construct the energy consumption forecast sub-model corresponding to the control valve group: ΔP’ pred =ΔP pred +P res Among them, ΔP pred is the predicted value of the energy consumption mechanism model, ΔP' pred It is the predicted value of the fusion model of mechanism and data.

6. The method according to claim 5, characterized in that The step 4 specifically includes: Step 4.1: Collect real-time operating data from the die forging equipment hydraulic system and use an offset detection algorithm to determine the data distribution changes of the real-time operating data. If the data distribution changes slightly and gradually, and the model error gradually increases, proceed to step 4.

2. If the data distribution changes suddenly and the model performance drops sharply, proceed to step 4.

3. Step 4.2, define the loss function through the EWC incremental learning algorithm and optimize each energy consumption prediction sub-model accordingly. When the loss function is minimized, the optimized energy consumption prediction sub-model is obtained, where the expression of the loss function is: Among them, L B (θ) is the loss function under the new training data, λ is the regularization coefficient, F i is the Fisher information matrix, θ i are network parameters, is the optimal parameter obtained from the old data, and N is the number of samples; Step 4.3: Use the transfer learning algorithm domain adversarial neural network to transfer each energy consumption prediction sub-model in the source domain to the target domain for energy consumption prediction. The source domain and target domain inputs are mapped to θ through the feature extraction layer. f Parameterize the new space, and the prediction layer maps this new space representation to a weighted θ y Parameterized output, then the prediction loss under the i-th task is converted into an optimization problem and each energy consumption prediction sub-model is updated accordingly, where the expression of the optimization problem is Among them, θ d is a parameterized domain classifier, is a logistic regressor that determines whether a given input comes from the source domain or the target domain.

7. The method according to claim 6, characterized in that The step 6 specifically includes: Deploy optimized energy consumption prediction sub-models on edge devices, build data storage and analysis modules on the cloud, and realize online update of energy consumption models through cloud-edge collaboration. The energy consumption prediction results of the energy conversion unit of the hydraulic system of die forging equipment will be output and displayed on a visual interface.

8. The method according to claim 7, characterized in that The preset evaluation indicators include root mean square error, mean absolute percentage error and R 2 coefficient; The expression of the root mean square error is: Among them, y i is the actual energy consumption value of the i-th data point, is the predicted energy consumption value of the i-th data point, and n is the total amount of data; The expression of the mean absolute percentage error is: The R 2 The expression of the coefficient is Here, SSR represents the regression sum of squares, and SST represents the total sum of squares.

Images