Soft-sensing method for hydraulic pump output flow considering service performance degradation

Abstract

The present invention provides a soft-sensing method for hydraulic pump output flow that takes into account service performance degradation. The method comprises: obtaining the hydraulic pump's speed, inlet / outlet pressure differential, oil temperature, and actual flow rate under different operating conditions; constructing a hydraulic pump flow prediction model, the hydraulic pump flow prediction model comprising a neural network and a parameter correction module. The input parameters corresponding to the input layer of the neural network are the obtained hydraulic pump speed, inlet / outlet pressure differential, and oil temperature, and the output layer is the predicted flow rate. The parameter correction module corrects the optimal weight matrix and optimal threshold matrix of the neural network based on the hydraulic pump speed, inlet / outlet pressure differential, oil temperature, and actual flow rate obtained when the hydraulic pump experiences service performance degradation. Based on the corrected hydraulic pump flow prediction model, the flow rate output by the hydraulic pump after service performance degradation is predicted. The present invention can accurately predict the hydraulic pump output flow rate even after the hydraulic pump experiences service performance degradation.

Description

Technical Field

[0001] The invention belongs to the field of hydraulic pump flow measurement, and in particular relates to a hydraulic pump output flow soft measurement method taking into account service performance degradation. Background Art

[0002] As a core component of a hydraulic system, precise flow control of the hydraulic pump is crucial to the overall system's performance. In traditional hydraulic systems, pump flow is primarily measured using additional flow meters and sensors. However, these additions not only increase system purchase and maintenance costs but also result in additional energy losses. Furthermore, these flow meters and sensors are not suitable for applications where space is limited.

[0003] Compared to direct flow measurement methods, indirect flow measurement methods such as flow maps and polynomial fitting can effectively overcome these issues. However, due to external load variations, uncertainties in hydraulic oil parameters, and internal pump leakage, the actual output flow of a hydraulic pump deviates significantly from the expected output flow, and the actual output flow is highly nonlinear. Consequently, the interaction of these factors makes it difficult to establish an accurate mathematical model of the hydraulic pump. Furthermore, many internal pump parameters are difficult to measure directly and require estimation using complex identification algorithms, further complicating model development.

[0004] In addition, after long-term operation, the internal components of the hydraulic pump will wear and degrade, which will cause the flow characteristics to change. At this time, the indirect flow measurement model based on the flow map and polynomial fitting needs to be re-established. Summary of the Invention

[0005] The present invention provides a soft measurement method for hydraulic pump output flow considering service performance degradation, so as to solve the problem that when the service performance of the hydraulic pump degrades, it is difficult to accurately predict the output flow of the hydraulic pump using the existing hydraulic pump flow model.

[0006] According to a first aspect of an embodiment of the present invention, a soft measurement method for a hydraulic pump output flow rate considering service performance degradation is provided, comprising: step S100, obtaining a speed, an inlet and outlet pressure difference, an oil temperature, and an actual flow rate of the hydraulic pump under different working conditions;

[0007] Step S200: constructing a hydraulic pump flow prediction model, wherein the hydraulic pump flow prediction model includes a neural network and a parameter correction module. The input parameters of the neural network corresponding to the input layer are the obtained hydraulic pump speed, inlet and outlet pressure difference, and oil temperature, and the output layer is the predicted flow;

[0008] Step S300: The parameter correction module corrects the optimal weight matrix and the optimal threshold matrix of the neural network based on the speed, inlet and outlet pressure difference, oil temperature, and actual flow rate of the hydraulic pump obtained when the hydraulic pump has service performance degradation;

[0009] Step S400: Based on the revised hydraulic pump flow prediction model, predict the flow output by the hydraulic pump after service performance degradation occurs.

[0010] In an optional implementation, the nonlinear mapping from the input layer to the first hidden layer of the neural network in step S200 is:

[0011]

[0012] where u j is the jth neuron in the first hidden layer, j is an integer greater than 0 and less than or equal to the number of neurons in each hidden layer m, ω is the speed of the hydraulic pump, Δp is the inlet and outlet pressure difference, T is the oil temperature, w 1j is the weight of the hydraulic pump speed to the jth neuron in the first hidden layer, w 2j is the weight of the inlet and outlet pressure difference to the jth neuron in the first hidden layer, w 3j is the weight from oil temperature to the jth neuron in the first hidden layer, is the threshold of the jth neuron from the input layer to the first hidden layer, and f(x) is the activation function corresponding to the input layer to the first hidden layer;

[0013] The nonlinear mapping between two adjacent hidden layers is:

[0014]

[0015] where u p is the pth neuron in the hidden layer of this layer, p is an integer, which is greater than 0 and less than or equal to the number of neurons m in each hidden layer, w 1p is the weight from the first neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u 1p is the first neuron in the previous adjacent hidden layer, w 2p is the weight from the second neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u 2p is the second neuron in the previous adjacent hidden layer, w mp is the weight from the mth neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u mp is the mth neuron in the previous adjacent hidden layer, is the threshold of the pth neuron in the previous adjacent hidden layer to the current hidden layer, and f(x) is the activation function corresponding to the previous adjacent hidden layer to the current hidden layer;

[0016] The nonlinear mapping from the last hidden layer to the output layer is:

[0017]

[0018] Where Q′ is the output layer predicted flow, w k is the weight of the kth neuron in the last hidden layer to predict the flow to the output layer, u k is the kth neuron in the last hidden layer, k is an integer greater than 0 and less than or equal to the number of neurons in each hidden layer m, b is the threshold of the predicted flow from the last hidden layer to the output layer, and g(x) is the activation function corresponding to the last hidden layer to the output layer.

[0019] In another optional implementation, step S200 specifically includes:

[0020] Step S201, designing the number of hidden layers and the number of neurons in each hidden layer in the neural network: the number of hidden layers is set within a set range, and the number of neurons is set within a set range, wherein each number of layers within the set range is combined with each number of neurons within the set range;

[0021] Step S202: When the optimal weight matrix and the optimal threshold matrix in the neural network are constant, for each combination, the hydraulic pump speed, the inlet and outlet pressure difference, and the oil temperature under each operating condition obtained when the hydraulic pump has not experienced service performance degradation are used as a group of samples. Each group of samples is input into the neural network corresponding to the combination to obtain a predicted flow rate for each group of samples. Based on the actual flow rate and the predicted flow rate of each group of samples, the root mean square error of the flow rate corresponding to the combination is calculated;

[0022] Step S203: Determine the combination with the smallest flow root mean square error among all combinations, and use the number of layers and the number of neurons in the combination as the number of hidden layers and the number of neurons in each hidden layer in the neural network.

[0023] In another optional implementation, step S200 further specifically includes:

[0024] Step S204: When the number of hidden layers and the number of neurons in each hidden layer of the neural network are constant, each set of samples obtained when the hydraulic pump has not experienced service performance degradation is input into the neural network for training to obtain a predicted flow rate for each set of samples. The neural network determines the network performance of the neural network based on the actual flow rate and the predicted flow rate of each set of samples.

[0025] Step S205: Determine whether the network performance of the neural network meets the standard. If so, obtain the optimal weight matrix and optimal threshold matrix of the neural network. Otherwise, adjust the weight matrix and threshold matrix in the neural network and return to step S204.

[0026] The weight matrix includes the weight of each input parameter in the input layer to each neuron in the first hidden layer, the weight of each neuron in the hidden layer to each neuron in the next hidden layer, and the weight of each neuron in the last hidden layer to the predicted flow in the output layer; the threshold matrix includes the threshold of each neuron in the first hidden layer from the input layer, the threshold of each neuron in the next hidden layer from the hidden layer, and the threshold of the predicted flow from the last hidden layer to the output layer.

[0027] In another optional implementation, the method further includes: before constructing the neural network, normalizing the obtained rotational speed, inlet and outlet pressure difference and oil temperature of the hydraulic pump, and the output predicted flow rate.

[0028] In another optional implementation, step S300 specifically includes:

[0029] Step S310: Assume that the correction coefficients of the optimal weight matrix and the optimal threshold matrix are α and β respectively, and use the hydraulic pump speed, inlet and outlet pressure difference, and oil temperature under each working condition obtained when the hydraulic pump has service performance degradation as a group of samples. Input each group of samples into the neural network to obtain the predicted flow rate of each group of samples;

[0030] Step S320: Determine a loss function based on the actual flow and predicted flow of each group of samples;

[0031] Step S330: Based on the gradient descent method, the correction coefficients α and β are iteratively updated according to the loss function and the set descent rate;

[0032] Step S340: Determine whether the number of iterations is greater than the set number or whether the root mean square error is less than the set value. If so, the update of the correction coefficients α and β is completed; otherwise, return to step S310.

[0033] In another optional implementation, before step S320, the method further includes: obtaining a mathematical model related to the hydraulic pump speed, the inlet and outlet pressure difference, and the oil temperature based on a mathematical model of the hydraulic pump output flow rate and a relationship between oil viscosity, oil temperature, and the inlet and outlet pressure difference;

[0034] Adding samples different from the groups of samples obtained when the hydraulic pump experiences service performance degradation, inputting each group of added samples into the mathematical model to obtain a theoretical calculated flow rate for each group of added samples; inputting each group of added samples into the neural network to obtain a predicted flow rate for each group of added samples;

[0035] The step S320 includes determining a loss function based on the actual flow rate and predicted flow rate of each group of samples obtained when the service performance of the hydraulic pump degrades, and the theoretically calculated flow rate and predicted flow rate of each group of additional samples.

[0036] In another optional implementation, step S330 includes: iteratively optimizing the descent rate using an adaptive learning rate optimization algorithm; and iteratively updating the correction coefficients α and β according to the loss function and the optimized descent rate.

[0037] In another optional implementation, in step S320, the loss function J is determined according to the following formula based on the actual flow and predicted flow of each group of samples: t :

[0038]

[0039] where n t is the number of samples obtained when the hydraulic pump shows service performance degradation, These are samples obtained when the hydraulic pump's service performance degrades. is the actual flow rate obtained when the hydraulic pump has service performance degradation, f t () is a neural network;

[0040] In step S330, the correction coefficients α and β are iteratively updated according to the loss function and the set descent speed according to the following formula:

[0041]

[0042] Where v is the gradient descent speed, e is the number of iterations, is the partial differential;

[0043] The corrected weight matrix W t and threshold matrix B t They are:

[0044] W t =α * W s

[0045] B t =β * B s

[0046] where α *and β * are the correction coefficients of the optimal weight matrix and the optimal threshold matrix when the number of iterations is greater than the set number or the root mean square error is less than the set value, W s and B s are the optimal weight matrix and optimal threshold matrix of the neural network when the hydraulic pump has no service performance degradation.

[0047] In another optional implementation, the mathematical model Q related to the hydraulic pump speed, inlet and outlet pressure difference and oil temperature is pump for:

[0048]

[0049] Where D is the displacement of the hydraulic pump, ω is the speed of the hydraulic pump, μ is the oil viscosity, ρ is the oil density, Δp is the pressure difference between the inlet and outlet of the hydraulic pump, C a 、C b 、C c 、C d These are some coefficients related to the internal structure of the hydraulic pump;

[0050] a, b, and c are coefficients related to the hydraulic oil, and T is the oil temperature;

[0051] In step S320, the loss function J is determined according to the following formula based on the actual flow rate and predicted flow rate of each group of samples obtained when the service performance of the hydraulic pump degrades and the theoretical calculated flow rate and predicted flow rate of each group of additional samples:

[0052]

[0053] where n t is the number of samples obtained when the hydraulic pump shows service performance degradation, and These are samples obtained when the hydraulic pump's service performance deteriorated and additional samples. is the actual flow rate obtained when the hydraulic pump has service performance degradation, f t () is a neural network; After inputting the mathematical model into the corresponding group of additional samples, the theoretical calculated flow rate of the group of additional samples is obtained, where m is the number of additional samples;

[0054] In step S330, the correction coefficients α and β are iteratively updated according to the loss function and the optimized descent rate according to the following formula:

[0055]

[0056] Where υ′ is the gradient descent speed after this iterative optimization, e is the number of iterations, is the partial differential;

[0057] The corrected weight matrix W t and threshold matrix B t They are:

[0058] W t =α * W s

[0059] B t =β * B s

[0060] where α * and β * are the correction coefficients of the optimal weight matrix and the optimal threshold matrix when the number of iterations is greater than the set number or the root mean square error is less than the set value, W s and B s are the optimal weight matrix and optimal threshold matrix of the neural network when the hydraulic pump has no service performance degradation.

[0061] The beneficial effects of the present invention are:

[0062] 1. The present invention provides a parameter correction module in the hydraulic pump flow prediction model. Based on the hydraulic pump speed, inlet and outlet pressure difference, oil temperature, and actual flow rate obtained when the hydraulic pump has experienced service performance degradation, the optimal weight matrix and optimal threshold matrix of the neural network are corrected. In this way, the hydraulic pump output flow rate can be accurately predicted even when the hydraulic pump has experienced service performance degradation.

[0063] 2. The present invention designs the number of hidden layers and neurons in the neural network to better learn the features corresponding to different tasks. The design of the neural network structure based on actual flow calibration data can make the resulting neural network model more realistically reflect the actual flow characteristics of the hydraulic pump, thereby improving the flow prediction accuracy and generalization ability of the model.

[0064] 3. The present invention trains the optimal weight matrix and threshold matrix of the neural network based on the groups of samples obtained when the hydraulic pump has not experienced service performance degradation. The groups of samples obtained when the hydraulic pump has experienced service performance degradation are only used to modify the optimal weight matrix and threshold matrix, without involving the establishment of the optimal weight matrix and threshold matrix. In this way, the neural network can more realistically reflect the actual flow characteristics of the hydraulic pump.

[0065] 4. The present invention normalizes the obtained hydraulic pump speed, inlet and outlet pressure difference, oil temperature, and output predicted flow rate to prevent different dimensions and data scales from affecting model performance;

[0066] 5. The present invention sets correction coefficients α and β corresponding to the optimal weight matrix and optimal threshold matrix of the neural network, inputs each group of samples obtained when the service performance of the hydraulic pump degrades into the neural network, obtains the predicted flow rate of each group of samples, determines a loss function based on the actual flow rate and the predicted flow rate of each group of samples, and iteratively updates the correction coefficients α and β based on the loss function and a set descent rate. In this way, the optimal weight matrix and the optimal threshold matrix are corrected, so that the hydraulic pump flow prediction model adapts to changes in flow characteristics caused by wear of internal components after long-term operation of the hydraulic pump, and the predicted flow rate takes into account the service performance degradation of the hydraulic pump. When the service performance of the hydraulic pump degrades, the flow prediction accuracy is high;

[0067] 6. The present invention introduces a mathematical model of the hydraulic pump output flow rate related to the hydraulic pump speed, inlet and outlet pressure difference and oil temperature. When the number of samples obtained when the service performance of the hydraulic pump is degraded is small, additional samples different from the obtained samples are added, and each group of additional samples is input into the mathematical model respectively to obtain the theoretical calculated flow rate of each group of samples. Each group of additional samples is input into the neural network to obtain the predicted flow rate of each group of additional samples. According to the actual flow rate and predicted flow rate of each group of samples obtained when the service performance of the hydraulic pump is degraded, as well as the theoretical calculated flow rate and predicted flow rate of each group of additional samples, a loss function is determined. According to the loss function and the optimized descent rate, the correction coefficients α and β of the optimal weight matrix and the optimal threshold matrix are iteratively updated. In this way, with a small number of samples and a limited number of working conditions, the hydraulic pump output flow rate can also be accurately predicted after the service performance of the hydraulic pump is degraded. In addition, the present invention also adopts an adaptive learning rate optimization algorithm to iteratively optimize the descent rate, thereby further improving the flow prediction accuracy of the hydraulic pump after the service performance is degraded. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a flow chart of an embodiment of a soft measurement method for hydraulic pump output flow considering service performance degradation according to the present invention;

[0069] Figure 2 It is a structural diagram of the existing pump-controlled hydraulic system test bench;

[0070] Figure 3 Schematic diagram of the neural network structure of the present invention;

[0071] Figure 4 This is a design table of the number of hidden layers and the number of neurons in each hidden layer in the neural network of the present invention;

[0072] Figure 5 This is a principle diagram of an embodiment of the present invention for correcting the optimal weight matrix and the optimal threshold matrix of a neural network;

[0073] Figure 6 This is a schematic diagram of another embodiment of the present invention for correcting the optimal weight matrix and the optimal threshold matrix of a neural network;

[0074] Figure 7 This is a comparison chart of the average flow prediction errors of the improved transfer neural network model and the initial neural network model corresponding to the method of the present invention on all data sets after service performance degradation. DETAILED DESCRIPTION

[0075] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention and to make the above-mentioned purposes, features and advantages of the embodiments of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention are further described in detail below with reference to the accompanying drawings.

[0076] In the description of the present invention, unless otherwise specified and limited, it should be noted that the term "connection" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection between two elements. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meaning of the above terms can be understood according to the specific circumstances.

[0077] See also Figure 1 , is a flow chart of an embodiment of a hydraulic pump output flow soft measurement method considering service performance degradation according to the present invention. The hydraulic pump output flow soft measurement method considering service performance degradation may include the following steps:

[0078] Step S100, obtaining the speed, inlet and outlet pressure difference, oil temperature and actual flow rate of the hydraulic pump under different working conditions;

[0079] Step S200: constructing a hydraulic pump flow prediction model, wherein the hydraulic pump flow prediction model includes a neural network and a parameter correction module. The input parameters of the neural network corresponding to the input layer are the obtained hydraulic pump speed, inlet and outlet pressure difference, and oil temperature, and the output layer is the predicted flow;

[0080] Step S300: The parameter correction module corrects the optimal weight matrix and the optimal threshold matrix of the neural network based on the speed, inlet and outlet pressure difference, oil temperature, and actual flow rate of the hydraulic pump obtained when the hydraulic pump has service performance degradation;

[0081] Step S400: Based on the revised hydraulic pump flow prediction model, predict the flow output by the hydraulic pump after service performance degradation occurs.

[0082] It can be seen from the above embodiments that the present invention sets a parameter correction module in the hydraulic pump flow prediction model, and corrects the optimal weight matrix and optimal threshold matrix of the neural network based on the hydraulic pump speed, inlet and outlet pressure difference, oil temperature and actual flow obtained when the hydraulic pump has service performance degradation. In this way, even when the hydraulic pump has service performance degradation, the hydraulic pump output flow can be accurately predicted.

[0083] In the above step S100, combined with Figure 2 As shown, the oil tank outlet is connected to the oil return port of the oil tank via a hydraulic pump 1, a check valve 3, a flowmeter 8, a relief valve 5, and a return oil filter 6. The connection point between the check valve 3 and the flowmeter 8 is connected to the connection point between the relief valve 5 and the return oil filter 6 via a throttle valve 4. An air cooler 7 is located adjacent to the return oil filter 6. A servo motor 2 is connected to the hydraulic pump 1. The speed of the hydraulic pump 1 can be obtained by detecting the speed of the servo motor 2. The outlet of the hydraulic pump 1 is connected to a pressure detection device for detecting the outlet pressure of the hydraulic pump. Since the inlet of the hydraulic pump is at atmospheric pressure, the pressure difference between the outlet pressure and atmospheric pressure can be obtained by detecting the outlet pressure of the hydraulic pump. A temperature detector is installed in the oil tank to detect the oil temperature. The flowmeter 8 is used to detect the actual flow rate output by the hydraulic pump. The speed of the hydraulic pump is controlled by the host computer, and the outlet pressure is controlled by adjusting the relief pressure of the relief valve. The oil temperature is controlled by heating the relief valve to increase the temperature and cooling the temperature by the air cooler.

[0084] In the above step S200, the neural network may include an input layer, multiple hidden layers and an output layer. Figure 3 As shown, the nonlinear mapping from the input layer to the first hidden layer of the neural network can be expressed as:

[0085]

[0086] where u j is the jth neuron in the first hidden layer, j is an integer greater than 0 and less than or equal to the number of neurons in each hidden layer m, ω is the speed of the hydraulic pump, Δp is the inlet and outlet pressure difference, T is the oil temperature, w 1j is the weight of the hydraulic pump speed to the jth neuron in the first hidden layer, w 2j is the weight of the inlet and outlet pressure difference to the jth neuron in the first hidden layer, w 3j is the weight from oil temperature to the jth neuron in the first hidden layer, is the threshold of the jth neuron from the input layer to the first hidden layer, and f(x) is the activation function corresponding to the input layer to the first hidden layer;

[0087] The nonlinear mapping between two adjacent hidden layers can be expressed as:

[0088]

[0089] where u p is the pth neuron in the hidden layer of this layer, p is an integer, which is greater than 0 and less than or equal to the number of neurons m in each hidden layer, w 1p is the weight from the first neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u 1p is the first neuron in the previous adjacent hidden layer, w 2p is the weight from the second neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u 2p is the second neuron in the previous adjacent hidden layer, w mp is the weight from the mth neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u mp is the mth neuron in the previous adjacent hidden layer, is the threshold of the pth neuron in the previous adjacent hidden layer to the current hidden layer, and f(x) is the activation function corresponding to the previous adjacent hidden layer to the current hidden layer;

[0090] The nonlinear mapping from the last hidden layer to the output layer can be expressed as:

[0091]

[0092] Where Q′ is the output layer predicted flow, w k is the weight of the kth neuron in the last hidden layer to predict the flow to the output layer, u k is the kth neuron in the last hidden layer, k is an integer greater than 0 and less than or equal to the number of neurons in each hidden layer m, b is the threshold of the predicted flow from the last hidden layer to the output layer, and g(x) is the activation function corresponding to the last hidden layer to the output layer.

[0093] In the above step S200, the establishment of the neural network can be divided into designing the number of hidden layers and the number of neurons in each hidden layer in the neural network and determining the optimal weight matrix and the optimal threshold matrix of the neural network when the hydraulic pump does not experience service performance degradation. The step S200 can specifically include:

[0094] Step S201, design the number of hidden layers and the number of neurons in each hidden layer in the neural network: the number of hidden layers is set within a set range, and the number of neurons is set within a set range, wherein each number of layers within the set range is combined with each number of neurons within the set range. Figure 4As shown, the number of hidden layers can be in the range of 1 to 5, and the number of neurons in each hidden layer can be in the range of (2, 4, 6, 8, 10). Each number of layers in 1 to 5 is combined with the number of neurons respectively.

[0095] Step S202: When the optimal weight matrix and the optimal threshold matrix in the neural network are constant, for each combination, the hydraulic pump speed, inlet and outlet pressure difference, and oil temperature under each operating condition obtained when the hydraulic pump has not experienced service performance degradation are used as a set of samples. Each set of samples is input into the neural network corresponding to the combination to obtain the predicted flow rate of each set of samples. Based on the actual flow rate of each set of samples and the predicted flow rate, the root mean square error of the flow rate corresponding to the combination is calculated. The root mean square error RMSE can be expressed as:

[0096]

[0097] Where n is the number of samples, Q i is the actual flow value of the i-th group of samples, Q i ′ is the predicted flow value of the i-th group of samples.

[0098] Step S203: determine the combination with the smallest flow root mean square error among all combinations, and use the number of layers and the number of neurons in the combination as the number of hidden layers and the number of neurons in each hidden layer in the neural network. Figure 4 As shown in the figure, it can be seen that when the number of hidden layers is 3 and the number of neurons is 8, the RMSE of the network is the smallest, and the structural performance of the corresponding neural network is the best.

[0099] The present invention designs the number of hidden layers and the number of neurons in the hidden layer of the neural network, which can enable it to better learn the features corresponding to different tasks; the design of the neural network structure based on actual flow calibration data can make the obtained neural network model more realistically reflect the actual flow characteristics of the hydraulic pump, thereby improving the flow prediction accuracy and generalization ability of the model.

[0100] In addition, the step S200 may further specifically include:

[0101] Step S204: When the number of hidden layers of the neural network and the number of neurons in each hidden layer are constant, each group of samples obtained when the hydraulic pump has not experienced service performance degradation is input into the neural network for training to obtain a predicted flow rate for each group of samples. The neural network determines the network performance of the neural network based on the actual flow rate and the predicted flow rate of each group of samples.

[0102] Step S205: Determine whether the network performance of the neural network meets the standard. If so, obtain the optimal weight matrix and optimal threshold matrix of the neural network. Otherwise, adjust the optimal weight matrix and optimal threshold matrix in the neural network and return to step S204. -6 ), it can be determined that the network performance of the neural network meets the standard. The weight matrix can include the weight of each input parameter in the input layer to each neuron in the first hidden layer, the weight of each neuron in the hidden layer to each neuron in the next hidden layer, and the weight of each neuron in the last hidden layer to the predicted flow in the output layer; the threshold matrix can include the threshold from the input layer to each neuron in the first hidden layer, the threshold from the hidden layer to each neuron in the next hidden layer, and the threshold from the last hidden layer to the predicted flow in the output layer. The groups of samples in step S200 can be samples obtained when the hydraulic pump has not experienced service performance degradation.

[0103] In this embodiment, the data obtained when the hydraulic pump has no service performance degradation is used as the source domain data, and the data obtained when the hydraulic pump has service performance degradation is used as the target domain data; the source domain data D s and target domain data D t They can be expressed as:

[0104]

[0105] n s is the number of samples in the source domain, n t is the number of samples in the target domain, and n t <<n s , / is the separator.

[0106] The initial weight matrix W of the neural network s and the initial threshold matrix B s It can be expressed as:

[0107]

[0108] in and are the weight matrix and threshold matrix from the input layer to the first hidden layer, and They are the weight matrix and threshold matrix corresponding to the last hidden layer to the output layer respectively.

[0109] When training the optimal weight matrix and the optimal threshold matrix of the neural network, the present invention is based on the groups of samples obtained when the hydraulic pump has not experienced service performance degradation. The groups of samples obtained when the hydraulic pump has experienced service performance degradation are only used to correct the optimal weight matrix and the optimal threshold matrix, and do not involve the establishment of the optimal weight matrix and the optimal threshold matrix. In this way, the neural network can more realistically reflect the actual flow characteristics of the hydraulic pump.

[0110] In order to prevent different dimensions and data scales from affecting model performance, the method may further include: before constructing the neural network, normalizing the obtained speed, inlet and outlet pressure difference, oil temperature, and output predicted flow of the hydraulic pump, and the expression thereof is:

[0111]

[0112] is the expression for normalizing each input parameter separately; An expression for normalizing the output layer predicted flow.

[0113] Since the wear of the internal components of the hydraulic pump after long-term operation may cause its flow characteristics to change, the present invention designs a parameter correction module to correct the optimal weight matrix and optimal threshold matrix of the neural network based on the speed, inlet and outlet pressure difference, oil temperature and actual flow rate of the hydraulic pump obtained when the service performance of the hydraulic pump degrades. In one embodiment of step S300, combined with Figure 5 As shown, the step S300 may specifically include:

[0114] Step S310: Assume that the correction coefficients of the optimal weight matrix and the optimal threshold matrix are α and β, respectively. The hydraulic pump speed, inlet and outlet pressure difference, and oil temperature under each operating condition obtained when the hydraulic pump experiences service performance degradation are used as a set of samples. Each set of samples is input into the neural network to obtain a predicted flow rate for each set of samples. In this embodiment, the initial values ​​of the correction coefficients α and β can be arbitrary. Since the input parameters and the output layer predicted flow rate are normalized in advance, the initial values ​​of α and β are also limited to between (0, 1].

[0115] Step S320: Determine the loss function based on the actual flow and predicted flow of each group of samples. In this step, the loss function J can be determined according to the following formula: t :

[0116]

[0117] where n t is the number of samples obtained when the hydraulic pump shows service performance degradation, These are samples obtained when the hydraulic pump's service performance degrades. is the actual flow rate obtained when the hydraulic pump has service performance degradation, f t () is a neural network.

[0118] Step S330: Based on the gradient descent method, the correction coefficients α and β are iteratively updated according to the loss function and the set descent rate. In this step, the correction coefficients α and β are iteratively updated according to the following formula:

[0119]

[0120] Where υ is the gradient descent rate (which can be the same as the iterative descent rate during neural network training, set to 0.01), e is the number of iterations, is the partial differential symbol.

[0121] Step S340: Determine whether the number of iterations is greater than the set number or whether the root mean square error is less than the set value. If so, the update of the correction coefficients α and β is completed; otherwise, return to step S310.

[0122] The corrected weight matrix W t and threshold matrix B t They are:

[0123] W t =α * W s

[0124] B t =β * B s

[0125] where α * and β * are the correction coefficients of the optimal weight matrix and the optimal threshold matrix when the number of iterations is greater than the set number or the root mean square error is less than the set value, W s and B s are the optimal weight matrix and optimal threshold matrix of the neural network when the hydraulic pump has no service performance degradation.

[0126] The present invention sets correction coefficients α and β corresponding to the optimal weight matrix and the optimal threshold matrix of the neural network, inputs each group of samples obtained when the service performance of the hydraulic pump is degraded into the neural network, obtains the predicted flow of each group of samples, determines the loss function according to the actual flow and the predicted flow of each group of samples, and iteratively updates the correction coefficients α and β according to the loss function and the set descent rate. In this way, the correction of the optimal weight matrix and the optimal threshold matrix is ​​achieved, so that the hydraulic pump flow prediction model adapts to the changes in flow characteristics caused by wear of internal components after long-term operation of the hydraulic pump, and the predicted flow takes into account the service performance degradation of the hydraulic pump. When the service performance of the hydraulic pump is degraded, the flow prediction accuracy is high.

[0127] Although the embodiment of the above step S300 can also accurately predict the flow rate when the service performance of the hydraulic pump is degraded, the accuracy of the flow rate prediction needs to be further improved due to the small number of samples obtained when the service performance of the hydraulic pump is degraded and the limited coverage of working conditions. To this end, the present invention provides another embodiment of step S300, combined with Figure 6 As shown, step S300 in this embodiment may specifically include:

[0128] Step S310, assuming that the correction coefficients of the optimal weight matrix and the optimal threshold matrix are α and β respectively, the hydraulic pump speed, inlet and outlet pressure difference and oil temperature under each working condition obtained when the service performance of the hydraulic pump is degraded are taken as a group of samples, and each group of samples is input into the neural network respectively to obtain the predicted flow of each group of samples; according to the mathematical model of the hydraulic pump output flow and the relationship between the oil viscosity, the oil temperature and the inlet and outlet pressure difference, a mathematical model related to the hydraulic pump speed, inlet and outlet pressure difference and the oil temperature is obtained; additional samples different from the groups of samples obtained when the service performance of the hydraulic pump is degraded are added, and each group of additional samples is input into the mathematical model to obtain the theoretical calculated flow of each group of additional samples; each group of additional samples is input into the neural network to obtain the predicted flow of each group of additional samples.

[0129] The mathematical model Q related to the hydraulic pump speed, inlet and outlet pressure difference and oil temperature pump for:

[0130]

[0131] Where D is the displacement of the hydraulic pump, ω is the speed of the hydraulic pump, μ is the oil viscosity, ρ is the oil density, Δp is the pressure difference between the inlet and outlet of the hydraulic pump, C a 、C b 、C c 、C d These are some coefficients related to the internal structure of the hydraulic pump;

[0132] Since the neural network inputs in the hydraulic pump flow prediction model include three variables: ω, Δp, and T, and the mathematical model does not have the independent variable T, it is necessary to combine it with the loss function of the jointly modified neural network model and introduce the independent variable of temperature. Temperature and pressure differential variables mainly affect the viscosity of the oil, causing changes in the leakage of the hydraulic pump, and thus affecting its output flow. Therefore, the oil viscosity μ in the mathematical model is further improved based on the viscosity-temperature equation and the viscosity-pressure equation.

[0133] Since the viscosity-temperature equation is:

[0134]

[0135] Where: μ0 is the dynamic viscosity of the oil at a given temperature under one atmosphere of pressure, a, b, c are coefficients related to the hydraulic oil, and t is the oil temperature.

[0136] (2) Viscosity-pressure equation:

[0137] μ1=μ0e λΔp

[0138] Where: μ1 is the dynamic viscosity of the oil, λ is the viscosity-pressure coefficient, and Δp is the pressure difference between the inlet and outlet of the hydraulic pump.

[0139] Therefore, the oil viscosity μ in the mathematical model of the hydraulic pump output flow can be:

[0140] a, b, and c are coefficients related to the hydraulic oil, and T is the oil temperature.

[0141] Step S320: Determine a loss function based on the actual flow rate and predicted flow rate of each group of samples obtained when the hydraulic pump has service performance degradation, as well as the theoretically calculated flow rate and predicted flow rate of each group of additional samples. This step can determine the loss function J according to the following formula:

[0142]

[0143] where n t is the number of samples obtained when the hydraulic pump shows service performance degradation, and These are samples obtained when the hydraulic pump's service performance deteriorated and additional samples. is the actual flow rate obtained when the hydraulic pump has service performance degradation, f t () is a neural network; After inputting the mathematical model into the corresponding group of additional samples, the theoretical calculated flow rate of the group of additional samples is obtained, where m is the number of additional samples.

[0144] Step S330: Based on the gradient descent method, the correction coefficients α and β are iteratively updated according to the loss function and the set descent rate. In this step, the correction coefficients α and β can be iteratively updated according to the following formula:

[0145]

[0146] Where υ′ is the gradient descent speed after this iterative optimization, e is the number of iterations, is the partial differential.

[0147] Step S340: Determine whether the number of iterations is greater than the set number or whether the root mean square error is less than the set value. If so, the update of the correction coefficients α and β is completed; otherwise, return to step S310.

[0148] The corrected weight matrix W t and threshold matrix B t They can be:

[0149] W t =α * W S

[0150] B t =β * B s

[0151] where α * and β * are the correction coefficients of the optimal weight matrix and the optimal threshold matrix when the number of iterations is greater than the set number or the root mean square error is less than the set value, W s and B s are the optimal weight matrix and optimal threshold matrix of the neural network when the hydraulic pump has no service performance degradation.

[0152] The present invention introduces a mathematical model of the hydraulic pump output flow rate related to the hydraulic pump speed, inlet and outlet pressure difference and oil temperature. When the number of samples obtained when the service performance of the hydraulic pump is degraded is small, samples different from the obtained samples are added, and each group of added samples is input into the mathematical model respectively to obtain the theoretical calculated flow rate of each group of samples. Each group of added samples is input into the neural network to obtain the predicted flow rate of each group of added samples. According to the actual flow rate and predicted flow rate of each group of samples obtained when the service performance of the hydraulic pump is degraded and the theoretical calculated flow rate and predicted flow rate of each group of added samples, a loss function is determined. According to the loss function and the optimized descent rate, the correction coefficients α and β of the optimal weight matrix and the optimal threshold matrix are iteratively updated. In this way, with a small number of samples and a limited number of working conditions, the hydraulic pump output flow rate can be accurately predicted after the service performance of the hydraulic pump is degraded. In addition, the present invention also adopts an adaptive learning rate optimization algorithm to iteratively optimize the descent rate, thereby further improving the flow prediction accuracy of the hydraulic pump after the service performance is degraded.

[0153] Figure 7 This is a comparison chart of the average traffic prediction errors of the improved transfer neural network model and the initial neural network model corresponding to the method of the present invention on all target domain data sets. It can be seen from the figure that the prediction error of the improved neural network has been greatly reduced.

[0154] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the following claims.

[0155] It will be appreciated that the present invention is not limited to the precise construction that has been described above and shown in the accompanying drawings, and that various modifications and variations can be made without departing from its scope, which is governed solely by the appended claims.

Claims

1. A soft measurement method for hydraulic pump output flow considering service performance degradation, characterized in that: include: Step S100, obtaining the speed, inlet and outlet pressure difference, oil temperature and actual flow rate of the hydraulic pump under different working conditions; Step S200: constructing a hydraulic pump flow prediction model and a parameter correction module. The hydraulic pump flow prediction model includes a neural network. The input parameters of the neural network corresponding to the input layer are the obtained hydraulic pump speed, inlet and outlet pressure difference, and oil temperature. The output layer is the predicted flow. Step S300: The parameter correction module corrects the optimal weight matrix and the optimal threshold matrix of the neural network based on the speed, inlet and outlet pressure difference, oil temperature, and actual flow rate of the hydraulic pump obtained when the hydraulic pump has service performance degradation; Step S400: predicting the output flow of the hydraulic pump after service performance degradation based on the revised hydraulic pump flow prediction model; The step S300 specifically includes: Step S310: Assume that the correction coefficients of the optimal weight matrix and the optimal threshold matrix are α and β respectively, and use the hydraulic pump speed, inlet and outlet pressure difference, and oil temperature under each working condition obtained when the hydraulic pump has service performance degradation as a group of samples. Input each group of samples into the neural network to obtain the predicted flow rate of each group of samples; Step S320: Determine a loss function based on the actual flow and predicted flow of each group of samples; Step S330: Based on the gradient descent method, the correction coefficients α and β are iteratively updated according to the loss function and the set descent rate; Step S340: Determine whether the number of iterations is greater than the set number or whether the root mean square error is less than the set value. If so, the update of the correction coefficients α and β is completed; otherwise, return to step S310; In step S320, the loss function J is determined according to the following formula based on the actual flow and predicted flow of each group of samples: t : where n t is the number of samples obtained when the hydraulic pump shows service performance degradation, Y is the sample obtained when the hydraulic pump has service performance degradation. t i is the actual flow rate obtained when the hydraulic pump has service performance degradation, f t () is a neural network; In step S330, the correction coefficients α and β are iteratively updated according to the loss function and the set descent speed according to the following formula: Where υ is the gradient descent rate, e is the number of iterations, is the partial differential; The corrected weight matrix W t and threshold matrix b t They are: w t =a * W s b t =b * B s where α * and β * are the correction coefficients of the optimal weight matrix and the optimal threshold matrix when the number of iterations is greater than the set number or the root mean square error is less than the set value, W s and B s are the optimal weight matrix and optimal threshold matrix of the neural network when the hydraulic pump has no service performance degradation.

2. The hydraulic pump output flow soft measurement method considering service performance degradation according to claim 1 is characterized in that: The nonlinear mapping from the input layer to the first hidden layer of the neural network in step S200 is: where u j is the jth neuron in the first hidden layer, j is an integer greater than 0 and less than or equal to the number of neurons in each hidden layer m, ω is the speed of the hydraulic pump, Δp is the inlet and outlet pressure difference, T is the oil temperature, w 1j is the weight of the hydraulic pump speed to the jth neuron in the first hidden layer, w 2j is the weight of the inlet and outlet pressure difference to the jth neuron in the first hidden layer, w 3j is the weight from oil temperature to the jth neuron in the first hidden layer, is the threshold of the jth neuron from the input layer to the first hidden layer, and f(x) is the activation function corresponding to the input layer to the first hidden layer; The nonlinear mapping between two adjacent hidden layers is: where u p is the pth neuron in the hidden layer of this layer, p is an integer, which is greater than 0 and less than or equal to the number of neurons m in each hidden layer, w 1p is the weight from the first neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u 1p is the first neuron in the previous adjacent hidden layer, w 2p is the weight from the second neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u 2p is the second neuron in the previous adjacent hidden layer, w mp is the weight from the mth neuron in the previous adjacent hidden layer to the pth neuron in the current hidden layer, u mp is the mth neuron in the previous adjacent hidden layer, is the threshold of the pth neuron in the previous adjacent hidden layer to the current hidden layer, and f(x) is the activation function corresponding to the previous adjacent hidden layer to the current hidden layer; The nonlinear mapping from the last hidden layer to the output layer is: Where Q′ is the output layer predicted flow, w k is the weight of the kth neuron in the last hidden layer to predict the flow to the output layer, u k is the kth neuron in the last hidden layer, k is an integer greater than 0 and less than or equal to the number of neurons in each hidden layer m, b is the threshold of the predicted flow from the last hidden layer to the output layer, and g(x) is the activation function corresponding to the last hidden layer to the output layer.

3. The hydraulic pump output flow soft measurement method considering service performance degradation according to claim 1 or 2, characterized in that: The step S200 specifically includes: Step S201, designing the number of hidden layers and the number of neurons in each hidden layer in the neural network: the number of hidden layers is set within a set range, and the number of neurons is set within a set range, wherein each number of layers within the set range is combined with each number of neurons within the set range; Step S202: When the optimal weight matrix and the optimal threshold matrix in the neural network are constant, for each combination, the hydraulic pump speed, the inlet and outlet pressure difference, and the oil temperature under each operating condition obtained when the hydraulic pump has not experienced service performance degradation are used as a group of samples. Each group of samples is input into the neural network corresponding to the combination to obtain a predicted flow rate for each group of samples. Based on the actual flow rate and the predicted flow rate of each group of samples, the root mean square error of the flow rate corresponding to the combination is calculated; Step S203: Determine the combination with the smallest flow root mean square error among all combinations, and use the number of layers and the number of neurons in the combination as the number of hidden layers and the number of neurons in each hidden layer in the neural network.

4. The soft measurement method for hydraulic pump output flow considering service performance degradation according to claim 3 is characterized in that: The step S200 further specifically includes: Step S204: When the number of hidden layers and the number of neurons in each hidden layer of the neural network are constant, each set of samples obtained when the hydraulic pump has not experienced service performance degradation is input into the neural network for training to obtain a predicted flow rate for each set of samples. The neural network determines the network performance of the neural network based on the actual flow rate and the predicted flow rate of each set of samples. Step S205: Determine whether the network performance of the neural network meets the standard. If so, obtain the optimal weight matrix and optimal threshold matrix of the neural network. Otherwise, adjust the weight matrix and threshold matrix in the neural network and return to step S204. The weight matrix includes the weight of each input parameter in the input layer to each neuron in the first hidden layer, the weight of each neuron in the hidden layer to each neuron in the next hidden layer, and the weight of each neuron in the last hidden layer to the predicted flow in the output layer; the threshold matrix includes the threshold of each neuron in the first hidden layer from the input layer, the threshold of each neuron in the next hidden layer from the hidden layer, and the threshold of the predicted flow from the last hidden layer to the output layer.

5. The soft measurement method for hydraulic pump output flow considering service performance degradation according to claim 1, characterized in that: The method further includes: before constructing the neural network, normalizing the obtained rotation speed, inlet and outlet pressure difference and oil temperature of the hydraulic pump, and the output predicted flow.

6. The hydraulic pump output flow soft-sensing method considering service performance degradation according to claim 1, characterized in that: Before step S320, the method further includes: obtaining a mathematical model related to the hydraulic pump speed, the inlet and outlet pressure difference, and the oil temperature according to a mathematical model of the hydraulic pump output flow rate and the relationship between the oil viscosity, the oil temperature, and the inlet and outlet pressure difference; Adding samples different from the groups of samples obtained when the hydraulic pump experiences service performance degradation, inputting each group of added samples into the mathematical model to obtain a theoretical calculated flow rate for each group of added samples; inputting each group of added samples into the neural network to obtain a predicted flow rate for each group of added samples; The step S320 includes determining a loss function based on the actual flow rate and predicted flow rate of each group of samples obtained when the service performance of the hydraulic pump degrades, and the theoretically calculated flow rate and predicted flow rate of each group of additional samples.

7. The hydraulic pump output flow soft-sensing method considering service performance degradation according to claim 6, characterized in that: The step S330 includes: iteratively optimizing the descent speed using an adaptive learning rate optimization algorithm; and iteratively updating the correction coefficients α and β according to the loss function and the optimized descent speed.

8. The soft measurement method for hydraulic pump output flow considering service performance degradation according to claim 7, characterized in that: The mathematical model Q related to the hydraulic pump speed, inlet and outlet pressure difference and oil temperature pump for: Where D is the displacement of the hydraulic pump, ω is the speed of the hydraulic pump, μ is the oil viscosity, ρ is the oil density, Δp is the pressure difference between the inlet and outlet of the hydraulic pump, C a 、C b 、C c 、C d These are some coefficients related to the internal structure of the hydraulic pump; a, b, and c are coefficients related to the hydraulic oil, and T is the oil temperature; In step S320, the loss function J is determined according to the following formula based on the actual flow rate and predicted flow rate of each group of samples obtained when the service performance of the hydraulic pump degrades and the theoretical calculated flow rate and predicted flow rate of each group of additional samples: where n t is the number of samples obtained when the hydraulic pump shows service performance degradation, and are the samples obtained when the hydraulic pump has service performance degradation and the additional samples, Y t i is the actual flow rate obtained when the hydraulic pump has service performance degradation, f t () is a neural network; After inputting the mathematical model into the corresponding group of additional samples, the theoretical calculated flow rate of the group of additional samples is obtained, where m is the number of additional samples; In step S330, the correction coefficients α and β are iteratively updated according to the loss function and the optimized descent rate according to the following formula: Where v′ is the gradient descent speed after this iterative optimization, e is the number of iterations, is the partial differential; The corrected weight matrix W t and threshold matrix B t They are: W t =a * W s B t =b * B s where α * and β * are the correction coefficients of the optimal weight matrix and the optimal threshold matrix when the number of iterations is greater than the set number or the root mean square error is less than the set value, W s and B s are the optimal weight matrix and optimal threshold matrix of the neural network when the hydraulic pump has no service performance degradation.

Images